Original Research

Cancer patients experience depression at rates > 5 times that of the general population.^1-11 Despite an increase in palliative care use, depression rates continued to rise.^2-4 Between 5% to 16% of outpatients, 4% to 14% of inpatients, and up to 49% of patients receiving palliative care experience depression.⁵ This issue also impacts families and caregivers.¹ A 2021 meta-analysis found that 23% of active military personnel and 20% of veterans experience depression.¹¹

Antidepressants approved by the US Food and Drug Administration (FDA) target the serotonin, norepinephrine, or dopamine systems and include boxed warnings about an increased risk of suicidal thoughts in adults aged 18 to 24 years.^12,13 These medications are categorized into several classes: monoamine oxidase inhibitors (MAOIs), tricyclic antidepressants (TCAs), tetracyclic antidepressants (TeCAs), norepinephrine-dopamine reuptake inhibitors (NDRIs), selective serotonin reuptake inhibitors (SSRIs), serotonin-norepinephrine reuptake inhibitors (SNRIs), serotonin receptor modulators (SRMs), serotonin-melatonin receptor antagonists (SMRAs), and N—methyl-D-aspartate receptor antagonists (NMDARAs).^14,15 The first FDA-approved antidepressants, iproniazid (an MAOI) and imipramine (a TCA) laid the foundation for the development of newer classes like SSRIs and SNRIs.^15-17

Older antidepressants such as MAOIs and TCAs are used less due to their adverse effects (AEs) and drug interactions. MAOIs, such as iproniazid, selegiline, moclobemide, tranylcypromine, isocarboxazid, and phenelzine, have numerous AEs and drug interactions, making them unsuitable for first- or second-line treatment of depression.^14,18-21 TCAs such as doxepin, amitriptyline, nortriptyline, imipramine, desipramine, clomipramine, trimipramine, protriptyline, maprotiline, and amoxapine have a narrow therapeutic index requiring careful monitoring for signs of toxicity such as QRS widening, tremors, or confusion. Despite the issues, TCAs are generally classified as second-line agents for major depressive disorder (MDD). TCAs have off-label uses for migraine prophylaxis, treatment of obsessive-compulsive disorder (OCD), insomnia, and chronic pain management first-line.^14,22-29

Newer antidepressants, including TeCAs and NDRIs, are typically more effective, but also come with safety concerns. TeCAs like mirtazapine interact with several medications, including MAOIs, serotonin-increasing drugs, alcohol, cannabidiol, and marijuana. Mirtazapine is FDA-approved for the treatment of moderate to severe depression in adults. It is also used off-label to treat insomnia, panic disorder, posttraumatic stress disorder (PTSD), generalized anxiety disorder (GAD), social anxiety disorder (SAD), headaches, and migraines. Compared to other antidepressants, mirtazapine is effective for all stages of depression and addresses a broad range of related symptoms.^14,30-34 NDRIs, such as bupropion, also interact with various medications, including MAOIs, other antidepressants, stimulants, and alcohol. Bupropion is FDA-approved for smoking cessation and to treat depression and SAD. It is also used off-label for depression- related bipolar disorder or sexual dysfunction, attention-deficit/hyperactivity disorder (ADHD), and obesity.^14,35-42

SSRIs, SNRIs, and SRMs should be used with caution. SSRIs such as sertraline, citalopram, escitalopram, fluoxetine, paroxetine, and fluvoxamine are first-line treatments for depression and various psychiatric disorders due to their safety and efficacy. Common AEs of SSRIs include sexual dysfunction, sleep disturbances, weight changes, and gastrointestinal (GI) issues. SSRIs can prolong the QT interval, posing a risk of life-threatening arrhythmia, and may interact with other medications, necessitating treatment adjustments. The FDA approved SSRIs for MDD, GAD, bulimia nervosa, bipolar depression, OCD, panic disorder, premenstrual dysphoric disorder, treatment-resistant depression, PTSD, and SAD. Off-label uses include binge eating disorder, body dysmorphic disorder, fibromyalgia, premature ejaculation, paraphilias, autism, Raynaud phenomenon, and vasomotor symptoms associated with menopause. Among SSRIs, sertraline and escitalopram are noted for their effectiveness and tolerability.^14,43-53

SNRIs, including duloxetine, venlafaxine, desvenlafaxine, milnacipran, and levomilnacipran, may increase bleeding risk, especially when taken with blood thinners. They can also elevate blood pressure, which may worsen if combined with stimulants. SNRIs may interact with other medications that affect serotonin levels, increasing the risk of serotonin syndrome when taken with triptans, pain medications, or other antidepressants.¹⁴ Desvenlafaxine has been approved by the FDA (but not by the European Medicines Agency).^54-56 Duloxetine is FDA-approved for the treatment of depression, neuropathic pain, anxiety disorders, fibromyalgia, and musculoskeletal disorders. It is used off-label to treat chemotherapy-induced peripheral neuropathy and stress urinary incontinence.^57-61 Venlafaxine is FDA-approved for depression, SAD, and panic disorder, and is prescribed off-label to treat ADHD, neuropathy, fibromyalgia, cataplexy, and PTSD, either alone or in combination with other medications.^62,63 Milnacipran is not approved for MDD; levomilnacipran received approval in 2013.⁶⁴

SRMs such as trazodone, nefazodone, vilazodone, and vortioxetine also function as serotonin reuptake inhibitors.^14,15 Trazodone is FDA-approved for MDD. It has been used off-label to treat anxiety, Alzheimer disease, substance misuse, bulimia nervosa, insomnia, fibromyalgia, and PTSD when first-line SSRIs are ineffective. A notable AE of trazodone is orthostatic hypotension, which can lead to dizziness and increase the risk of falls, especially in geriatric patients.^65-70 Nefazodone was discontinued in Europe in 2003 due to rare cases of liver toxicity but remains available in the US.^71-74 Vilazodone and vortioxetine are FDA-approved.

The latest classes of antidepressants include SMRAs and NMDARAs.¹⁴ Agomelatine, an SMRA, was approved in Europe in 2009 but rejected by the FDA in 2011 due to liver toxicity.⁷⁵ NMDARAs like esketamine and a combination of dextromethorphan and bupropion received FDA approval in 2019 and 2022, respectively.^76,77

This retrospective study analyzes noncancer drugs used during systemic chemotherapy based on a dataset of 14 antineoplastic agents. It sought to identify the most dispensed noncancer drug groups, discuss findings, compare patients with and without antidepressant prescriptions, and examine trends in antidepressant use from 2002 to 2023. This analysis expands on prior research.^78-81

Methods

The Walter Reed National Military Medical Center Institutional Review Board approved the study protocol and ensured compliance with the Health Insurance Portability and Accountability Act as an exempt protocol. The Joint Pathology Center (JPC) of the US Department of Defense (DoD) Cancer Registry Program and Military Health System (MHS) data experts from the Comprehensive Ambulatory/Professional Encounter Record (CAPER) and Pharmacy Data Transaction Service (PDTS) provided data for the analysis.

Data Sources

The JPC DoD Cancer Registry Program contains data from 1998 to 2024. CAPER and PDTS are part of the MHS Data Repository/Management Analysis and Reporting Tool database. Each observation in CAPER represents an ambulatory encounter at a military treatment facility (MTF). CAPER records are available from 2003 to 2024. PDTS records are available from 2002 to 2004. Each observation in PDTS represents a prescription filled for an MHS beneficiary, excluding those filled at international civilian pharmacies and inpatient pharmacy prescriptions.

This cross-sectional analysis requested data extraction for specific cancer drugs from the DoD Cancer Registry, focusing on treatment details, diagnosis dates, patient demographics, and physicians’ comments on AEs. After identifying patients, CAPER was used to identify additional health conditions. PDTS was used to compile a list of prescription medications filled during systemic cancer treatment or < 2 years postdiagnosis.

The 2016 Surveillance, Epidemiology, and End Results Program Coding and Staging Manual and International Classification of Diseases for Oncology, 3rd edition, 1st revision, were used to decode disease and cancer types.^82,83 Data sorting and analysis were performed using Microsoft Excel. The percentage for the total was calculated by using the number of patients or data available within the subgroup divided by the total number of patients or data variables. To compare the mean number of dispensed antidepressants to those without antidepressants, a 2-tailed, 2-sample z test was used to calculate the P value and determine statistical significance (P < .05) using socscistatistics.com.

Data were extracted 3 times between 2021 and 2023. The initial 2021 protocol focused on erlotinib and gefitinib. A modified protocol in 2022 added paclitaxel, cisplatin, docetaxel, pemetrexed, and crizotinib; further modification in 2023 included 8 new antineoplastic agents and 2 anticoagulants. Sotorasib has not been prescribed in the MHS, and JPC lacks records for noncancer drugs. The 2023 dataset comprised 2210 patients with cancer treated with 14 antineoplastic agents; 2104 had documented diagnoses and 2113 had recorded prescriptions. Data for erlotinib, gefitinib, and paclitaxel have been published previously.^78,79

Results

Of 2113 patients with recorded prescriptions, 1297 patients (61.4%) received 109 cancer drugs, including 96 antineoplastics, 7 disease-modifying antirheumatic agents, 4 biologic response modifiers, and 2 calcitonin gene-related peptides. Fourteen antineoplastic agents had complete data from JPC, while others were noted for combination therapies or treatment switches from the PDTS (Table 1). Seventy-six cancer drugs were prescribed with antidepressants in 489 patients (eAppendix).

The JPC provided 2242 entries for 2210 patients, ranging in age from 2 months to 88 years (mean, 56 years), documenting treatment from September 1988 to January 2023. Thirty-two patients had duplicate entries due to multiple cancer locations or occurrences. Of the 2242 patients, 1541 (68.7%) were aged > 50 years, 975 patients (43.5%) had cancers that were stage III or IV, and 1267 (56.5%) had cancers that were stage 0, I, II, or not applicable/unknown. There were 51 different types of cancer: breast, lung, testicular, endometrial, and ovarian were most common (n ≥ 100 patients). Forty-two cancer types were documented among 750 patients prescribed antidepressants (Table 2).

The CAPER database recorded 8882 unique diagnoses for 2104 patients, while PDTS noted 1089 unique prescriptions within 273 therapeutic codes for 2113 patients. Nine therapeutic codes (opiate agonists, adrenals, cathartics-laxatives, nonsteroidal anti-inflammatory agents, antihistamines for GI conditions, 5-HT3 receptor antagonists, analgesics and antipyretic miscellanea, antineoplastic agents, and proton-pump inhibitors) and 8 drugs (dexamethasone, prochlorperazine, ondansetron, docusate, acetaminophen, ibuprofen, oxycodone, and polyethylene glycol 3350) were associated with > 1000 patients (≥ 50%). Patients had between 1 and 275 unique health conditions and filled 1 to 108 prescriptions. The mean (SD) number of diagnoses and prescriptions was 50 (28) and 29 (12), respectively. Of the 273 therapeutic codes, 30 groups were analyzed, with others categorized into miscellaneous groups such as lotions, vaccines, and devices. Significant differences in mean number of prescriptions were found for patients taking antidepressants compared to those not (P < .05), except for anticonvulsants and antipsychotics (P = .12 and .09, respectively) (Table 3).

Antidepressants

Of the 2113 patients with recorded prescriptions, 750 (35.5%) were dispensed 17 different antidepressants. Among these 17 antidepressants, 183 (8.7%) patients received duloxetine, 158 (7.5%) received venlafaxine, 118 (5.6%) received trazodone, and 107 (5.1%) received sertraline (Figure 1, Table 4). Of the 750 patients, 509 (67.9%) received 1 antidepressant, 168 (22.4%) received 2, 60 (8.0%) received 3, and 13 (1.7%) received > 3. Combinations varied, but only duloxetine and trazodone were prescribed to > 10 patients.

Antidepressants were prescribed annually at an overall mean (SD) rate of 23% (5%) from 2003 to 2022 (Figure 2). Patients on antidepressants during systemic therapy had a greater number of diagnosed medical conditions and received more prescription medications compared to those not taking antidepressants (P < .001) (Figure 3). The 745 patients taking antidepressants in CAPER data had between 1 and 275 diagnosed medical issues, with a mean (SD) of 55 (31) vs a range of 1 to 209 and a mean (SD) of 46 (26) for the 1359 patients not taking antidepressants. The 750 patients on antidepressants in PDTS data had between 8 and 108 prescriptions dispensed, with a mean (SD) of 32 (12), vs a range of 1 to 65 prescriptions and a mean (SD) of 29 (12) for 1363 patients not taking antidepressants.

Discussion

The JPC DoD Cancer Registry includes information on cancer types, stages, treatment regimens, and physicians’ notes, while noncancer drugs are sourced from the PDTS database. The pharmacy uses a different documentation system, leading to varied classifications.

Database reliance has its drawbacks. For example, megestrol is coded as a cancer drug, although it’s primarily used for endometrial or gynecologic cancers. Many drugs have multiple therapeutic codes assigned to them, including 10 antineoplastic agents: diclofenac, Bacillus Calmette-Guérin (BCG), megestrol acetate, tamoxifen, anastrozole, letrozole, leuprolide, goserelin, degarelix, and fluorouracil. Diclofenac, BCG, and mitomycin have been repurposed for cancer treatment.^84-87 From 2003 to 2023, diclofenac was prescribed to 350 patients for mild-to-moderate pain, with only 2 patients receiving it for cancer in 2018. FDA-approved for bladder cancer in 1990, BCG was prescribed for cancer treatment for 1 patient in 2021 after being used for vaccines between 2003 and 2018. Tamoxifen, used for hormone receptor-positive breast cancer from 2004 to 2017 with 53 patients, switched to estrogen agonist-antagonists from 2017 to 2023 with 123 patients. Only a few of the 168 patients were prescribed tamoxifen using both codes.^88-91 Anastrozole and letrozole were coded as antiestrogens for 7 and 18 patients, respectively, while leuprolide and goserelin were coded as gonadotropins for 59 and 18 patients. Degarelix was coded as antigonadotropins, fluorouracil as skin and mucous membrane agents miscellaneous, and megestrol acetate as progestins for 7, 6, and 3 patients, respectively. Duloxetine was given to 186 patients, primarily for depression from 2005 to 2023, with 7 patients treated for fibromyalgia from 2022 to 2023.

Antidepressants Observed

Tables 1 and 5 provide insight into the FDA approval of 14 antineoplastics and antidepressants and their CYP metabolic pathways.^92-122 In Table 4, the most prescribed antidepressant classes are SNRIs, SRMs, SSRIs, TeCAs, NDRIs, and TCAs. This trend highlights a preference for newer medications with weak CYP inhibition. A total of 349 patients were prescribed SSRIs, 343 SNRIs, 119 SRMs, 109 TCAs, 83 TeCAs, and 79 NDRIs. MAOIs, SMRAs, and NMDARAs were not observed in this dataset. While there are instances of dextromethorphan-bupropion and sertraline-escitalopram being dispensed together, it remains unclear whether these were NMDARA combinations.

Among the 14 specific antineoplastic agents, 10 are metabolized by CYP isoenzymes, primarily CYP3A4. Duloxetine neither inhibits nor is metabolized by CYP3A4, a reason it is often recommended, following venlafaxine.

Both duloxetine and venlafaxine are used off-label for chemotherapy-induced peripheral neuropathy related to paclitaxel and docetaxel. According to the CYP metabolized pathway, duloxetine tends to have more favorable DDIs than venlafaxine. In PDTS data, 371 patients were treated with paclitaxel and 180 with docetaxel, with respective antidepressant prescriptions of 156 and 70. Of the 156 patients dispensed paclitaxel, 62 (40%) were dispensed with duloxetine compared to 43 (28%) with venlafaxine. Of the 70 patients dispensed docetaxel, 23 (33%) received duloxetine vs 24 (34%) with venlafaxine.

Of 85 patients prescribed duloxetine, 75 received it with either paclitaxel or docetaxel (5 received both). Five patients had documented AEs (1 neuropathy related). Of 67 patients prescribed venlafaxine, 66 received it with either paclitaxel or docetaxel. Two patients had documented AEs (1 was neuropathy related, the same patient who received duloxetine). Of the 687 patients treated with paclitaxel and 337 with docetaxel in all databases, 4 experienced neuropathic AEs from both medications.⁷⁹

Antidepressants can increase the risk of bleeding, especially when combined with blood thinners, and may elevate blood pressure, particularly alongside stimulants. Of the 554 patients prescribed 9 different anticoagulants, enoxaparin, apixaban, and rivaroxaban were the most common (each > 100 patients). Among these, 201 patients (36%) received both anticoagulants and antidepressants: duloxetine for 64 patients, venlafaxine for 30, trazodone for 35, and sertraline for 26. There were no data available to assess bleeding rates related to the evaluation of DDIs between these medication classes.

Antidepressants can be prescribed for erectile dysfunction. Of the 148 patients prescribed an antidepressant for erectile dysfunction, duloxetine, trazodone, and mirtazapine were the most common. Antidepressant preferences varied by cancer type. Duloxetine was the only antidepressant used for all types of cancer. Venlafaxine, duloxetine, trazodone, sertraline, and escitalopram were the most prescribed antidepressants for breast cancer, while duloxetine, mirtazapine, citalopram, sertraline, and trazodone were the most prescribed for lung cancer. Sertraline, duloxetine, trazodone, amitriptyline, and escitalopram were most common for testicular cancer. Duloxetine, venlafaxine, trazodone, amitriptyline, and sertraline were the most prescribed for endometrial cancer, while duloxetine, venlafaxine, amitriptyline, citalopram, and sertraline were most prescribed for ovarian cancer.

The broadness of International Statistical Classification of Diseases, Tenth Revision codes made it challenging to identify nondepression diagnoses in the analyzed population. However, if all antidepressants were prescribed to treat depression, service members with cancer exhibited a higher depression rate (35%) than the general population (25%). Of 2104 patients, 191 (9.1%) had mood disorders, and 706 (33.6%) had mental disorders: 346 (49.0%) had 1 diagnosis, and 360 (51.0%) had multiple diagnoses. The percentage of diagnoses varied yearly, with notable drops in 2003, 2007, 2011, 2014, and 2018, and peaks in 2006, 2008, 2013, 2017, and 2022. This fluctuation was influenced by events like the establishment of PDTS in 2002, the 2008 economic recession, a hospital relocation in 2011, the 2014 Ebola outbreak, and the COVID-19 pandemic. Although the number of patients receiving antidepressants increased from 2019 to 2022, the overall percentage of patients receiving them did not significantly change from 2003 to 2022, aligning with previous research.^5,125

Many medications have potential uses beyond what is detailed in the prescribing information. Antidepressants can relieve pain, while pain medications may help with depression. Opioids were once thought to effectively treat depression, but this perspective has changed with a greater understanding of their risks, including misuse.^126-131 Pain is a severe and often unbearable AE of cancer. Of 2113 patients, 92% received opioids; 34% received both opioids and antidepressants; 2% received only antidepressants; and 7% received neither. This study didn’t clarify whether those on opioids alone recognized their depression or if those on both were aware of their dependence. While SSRIs are generally not addictive, they can lead to physical dependence, and any medication can be abused if not managed properly.^132-134

Conclusions

This retrospective study analyzes data from antineoplastic agents used in systemic cancer treatment between 1988 and 2023, with a particular focus on the use of antidepressants. Data on antidepressant prescriptions are incomplete and specific to these agents, which means the findings cannot be generalized to all antidepressants. Hence, the results indicate that patients taking antidepressants had more diagnosed health issues and received more medications compared to patients who were not on these drugs.

This study underscores the need for further research into the effects of antidepressants on cancer treatment, utilizing all data from the DoD Cancer Registry. Future research should explore DDIs between antidepressants and other cancer and noncancer medications, as this study did not assess AE documentation, unlike in studies involving erlotinib, gefitinib, and paclitaxel.^78,79 Further investigation is needed to evaluate the impact of discontinuing antidepressant use during cancer treatment. This comprehensive overview provides insights for clinicians to help them make informed decisions regarding the prescription of antidepressants in the context of cancer treatment.

Cancer patients experience depression at rates > 5 times that of the general population.^1-11 Despite an increase in palliative care use, depression rates continued to rise.^2-4 Between 5% to 16% of outpatients, 4% to 14% of inpatients, and up to 49% of patients receiving palliative care experience depression.⁵ This issue also impacts families and caregivers.¹ A 2021 meta-analysis found that 23% of active military personnel and 20% of veterans experience depression.¹¹

Antidepressants approved by the US Food and Drug Administration (FDA) target the serotonin, norepinephrine, or dopamine systems and include boxed warnings about an increased risk of suicidal thoughts in adults aged 18 to 24 years.^12,13 These medications are categorized into several classes: monoamine oxidase inhibitors (MAOIs), tricyclic antidepressants (TCAs), tetracyclic antidepressants (TeCAs), norepinephrine-dopamine reuptake inhibitors (NDRIs), selective serotonin reuptake inhibitors (SSRIs), serotonin-norepinephrine reuptake inhibitors (SNRIs), serotonin receptor modulators (SRMs), serotonin-melatonin receptor antagonists (SMRAs), and N—methyl-D-aspartate receptor antagonists (NMDARAs).^14,15 The first FDA-approved antidepressants, iproniazid (an MAOI) and imipramine (a TCA) laid the foundation for the development of newer classes like SSRIs and SNRIs.^15-17

Older antidepressants such as MAOIs and TCAs are used less due to their adverse effects (AEs) and drug interactions. MAOIs, such as iproniazid, selegiline, moclobemide, tranylcypromine, isocarboxazid, and phenelzine, have numerous AEs and drug interactions, making them unsuitable for first- or second-line treatment of depression.^14,18-21 TCAs such as doxepin, amitriptyline, nortriptyline, imipramine, desipramine, clomipramine, trimipramine, protriptyline, maprotiline, and amoxapine have a narrow therapeutic index requiring careful monitoring for signs of toxicity such as QRS widening, tremors, or confusion. Despite the issues, TCAs are generally classified as second-line agents for major depressive disorder (MDD). TCAs have off-label uses for migraine prophylaxis, treatment of obsessive-compulsive disorder (OCD), insomnia, and chronic pain management first-line.^14,22-29

Newer antidepressants, including TeCAs and NDRIs, are typically more effective, but also come with safety concerns. TeCAs like mirtazapine interact with several medications, including MAOIs, serotonin-increasing drugs, alcohol, cannabidiol, and marijuana. Mirtazapine is FDA-approved for the treatment of moderate to severe depression in adults. It is also used off-label to treat insomnia, panic disorder, posttraumatic stress disorder (PTSD), generalized anxiety disorder (GAD), social anxiety disorder (SAD), headaches, and migraines. Compared to other antidepressants, mirtazapine is effective for all stages of depression and addresses a broad range of related symptoms.^14,30-34 NDRIs, such as bupropion, also interact with various medications, including MAOIs, other antidepressants, stimulants, and alcohol. Bupropion is FDA-approved for smoking cessation and to treat depression and SAD. It is also used off-label for depression- related bipolar disorder or sexual dysfunction, attention-deficit/hyperactivity disorder (ADHD), and obesity.^14,35-42

SSRIs, SNRIs, and SRMs should be used with caution. SSRIs such as sertraline, citalopram, escitalopram, fluoxetine, paroxetine, and fluvoxamine are first-line treatments for depression and various psychiatric disorders due to their safety and efficacy. Common AEs of SSRIs include sexual dysfunction, sleep disturbances, weight changes, and gastrointestinal (GI) issues. SSRIs can prolong the QT interval, posing a risk of life-threatening arrhythmia, and may interact with other medications, necessitating treatment adjustments. The FDA approved SSRIs for MDD, GAD, bulimia nervosa, bipolar depression, OCD, panic disorder, premenstrual dysphoric disorder, treatment-resistant depression, PTSD, and SAD. Off-label uses include binge eating disorder, body dysmorphic disorder, fibromyalgia, premature ejaculation, paraphilias, autism, Raynaud phenomenon, and vasomotor symptoms associated with menopause. Among SSRIs, sertraline and escitalopram are noted for their effectiveness and tolerability.^14,43-53

SNRIs, including duloxetine, venlafaxine, desvenlafaxine, milnacipran, and levomilnacipran, may increase bleeding risk, especially when taken with blood thinners. They can also elevate blood pressure, which may worsen if combined with stimulants. SNRIs may interact with other medications that affect serotonin levels, increasing the risk of serotonin syndrome when taken with triptans, pain medications, or other antidepressants.¹⁴ Desvenlafaxine has been approved by the FDA (but not by the European Medicines Agency).^54-56 Duloxetine is FDA-approved for the treatment of depression, neuropathic pain, anxiety disorders, fibromyalgia, and musculoskeletal disorders. It is used off-label to treat chemotherapy-induced peripheral neuropathy and stress urinary incontinence.^57-61 Venlafaxine is FDA-approved for depression, SAD, and panic disorder, and is prescribed off-label to treat ADHD, neuropathy, fibromyalgia, cataplexy, and PTSD, either alone or in combination with other medications.^62,63 Milnacipran is not approved for MDD; levomilnacipran received approval in 2013.⁶⁴

SRMs such as trazodone, nefazodone, vilazodone, and vortioxetine also function as serotonin reuptake inhibitors.^14,15 Trazodone is FDA-approved for MDD. It has been used off-label to treat anxiety, Alzheimer disease, substance misuse, bulimia nervosa, insomnia, fibromyalgia, and PTSD when first-line SSRIs are ineffective. A notable AE of trazodone is orthostatic hypotension, which can lead to dizziness and increase the risk of falls, especially in geriatric patients.^65-70 Nefazodone was discontinued in Europe in 2003 due to rare cases of liver toxicity but remains available in the US.^71-74 Vilazodone and vortioxetine are FDA-approved.

The latest classes of antidepressants include SMRAs and NMDARAs.¹⁴ Agomelatine, an SMRA, was approved in Europe in 2009 but rejected by the FDA in 2011 due to liver toxicity.⁷⁵ NMDARAs like esketamine and a combination of dextromethorphan and bupropion received FDA approval in 2019 and 2022, respectively.^76,77

This retrospective study analyzes noncancer drugs used during systemic chemotherapy based on a dataset of 14 antineoplastic agents. It sought to identify the most dispensed noncancer drug groups, discuss findings, compare patients with and without antidepressant prescriptions, and examine trends in antidepressant use from 2002 to 2023. This analysis expands on prior research.^78-81

Methods

The Walter Reed National Military Medical Center Institutional Review Board approved the study protocol and ensured compliance with the Health Insurance Portability and Accountability Act as an exempt protocol. The Joint Pathology Center (JPC) of the US Department of Defense (DoD) Cancer Registry Program and Military Health System (MHS) data experts from the Comprehensive Ambulatory/Professional Encounter Record (CAPER) and Pharmacy Data Transaction Service (PDTS) provided data for the analysis.

Data Sources

The JPC DoD Cancer Registry Program contains data from 1998 to 2024. CAPER and PDTS are part of the MHS Data Repository/Management Analysis and Reporting Tool database. Each observation in CAPER represents an ambulatory encounter at a military treatment facility (MTF). CAPER records are available from 2003 to 2024. PDTS records are available from 2002 to 2004. Each observation in PDTS represents a prescription filled for an MHS beneficiary, excluding those filled at international civilian pharmacies and inpatient pharmacy prescriptions.

This cross-sectional analysis requested data extraction for specific cancer drugs from the DoD Cancer Registry, focusing on treatment details, diagnosis dates, patient demographics, and physicians’ comments on AEs. After identifying patients, CAPER was used to identify additional health conditions. PDTS was used to compile a list of prescription medications filled during systemic cancer treatment or < 2 years postdiagnosis.

The 2016 Surveillance, Epidemiology, and End Results Program Coding and Staging Manual and International Classification of Diseases for Oncology, 3rd edition, 1st revision, were used to decode disease and cancer types.^82,83 Data sorting and analysis were performed using Microsoft Excel. The percentage for the total was calculated by using the number of patients or data available within the subgroup divided by the total number of patients or data variables. To compare the mean number of dispensed antidepressants to those without antidepressants, a 2-tailed, 2-sample z test was used to calculate the P value and determine statistical significance (P < .05) using socscistatistics.com.

Data were extracted 3 times between 2021 and 2023. The initial 2021 protocol focused on erlotinib and gefitinib. A modified protocol in 2022 added paclitaxel, cisplatin, docetaxel, pemetrexed, and crizotinib; further modification in 2023 included 8 new antineoplastic agents and 2 anticoagulants. Sotorasib has not been prescribed in the MHS, and JPC lacks records for noncancer drugs. The 2023 dataset comprised 2210 patients with cancer treated with 14 antineoplastic agents; 2104 had documented diagnoses and 2113 had recorded prescriptions. Data for erlotinib, gefitinib, and paclitaxel have been published previously.^78,79

Results

Of 2113 patients with recorded prescriptions, 1297 patients (61.4%) received 109 cancer drugs, including 96 antineoplastics, 7 disease-modifying antirheumatic agents, 4 biologic response modifiers, and 2 calcitonin gene-related peptides. Fourteen antineoplastic agents had complete data from JPC, while others were noted for combination therapies or treatment switches from the PDTS (Table 1). Seventy-six cancer drugs were prescribed with antidepressants in 489 patients (eAppendix).

The JPC provided 2242 entries for 2210 patients, ranging in age from 2 months to 88 years (mean, 56 years), documenting treatment from September 1988 to January 2023. Thirty-two patients had duplicate entries due to multiple cancer locations or occurrences. Of the 2242 patients, 1541 (68.7%) were aged > 50 years, 975 patients (43.5%) had cancers that were stage III or IV, and 1267 (56.5%) had cancers that were stage 0, I, II, or not applicable/unknown. There were 51 different types of cancer: breast, lung, testicular, endometrial, and ovarian were most common (n ≥ 100 patients). Forty-two cancer types were documented among 750 patients prescribed antidepressants (Table 2).

The CAPER database recorded 8882 unique diagnoses for 2104 patients, while PDTS noted 1089 unique prescriptions within 273 therapeutic codes for 2113 patients. Nine therapeutic codes (opiate agonists, adrenals, cathartics-laxatives, nonsteroidal anti-inflammatory agents, antihistamines for GI conditions, 5-HT3 receptor antagonists, analgesics and antipyretic miscellanea, antineoplastic agents, and proton-pump inhibitors) and 8 drugs (dexamethasone, prochlorperazine, ondansetron, docusate, acetaminophen, ibuprofen, oxycodone, and polyethylene glycol 3350) were associated with > 1000 patients (≥ 50%). Patients had between 1 and 275 unique health conditions and filled 1 to 108 prescriptions. The mean (SD) number of diagnoses and prescriptions was 50 (28) and 29 (12), respectively. Of the 273 therapeutic codes, 30 groups were analyzed, with others categorized into miscellaneous groups such as lotions, vaccines, and devices. Significant differences in mean number of prescriptions were found for patients taking antidepressants compared to those not (P < .05), except for anticonvulsants and antipsychotics (P = .12 and .09, respectively) (Table 3).

Antidepressants

Of the 2113 patients with recorded prescriptions, 750 (35.5%) were dispensed 17 different antidepressants. Among these 17 antidepressants, 183 (8.7%) patients received duloxetine, 158 (7.5%) received venlafaxine, 118 (5.6%) received trazodone, and 107 (5.1%) received sertraline (Figure 1, Table 4). Of the 750 patients, 509 (67.9%) received 1 antidepressant, 168 (22.4%) received 2, 60 (8.0%) received 3, and 13 (1.7%) received > 3. Combinations varied, but only duloxetine and trazodone were prescribed to > 10 patients.

Antidepressants were prescribed annually at an overall mean (SD) rate of 23% (5%) from 2003 to 2022 (Figure 2). Patients on antidepressants during systemic therapy had a greater number of diagnosed medical conditions and received more prescription medications compared to those not taking antidepressants (P < .001) (Figure 3). The 745 patients taking antidepressants in CAPER data had between 1 and 275 diagnosed medical issues, with a mean (SD) of 55 (31) vs a range of 1 to 209 and a mean (SD) of 46 (26) for the 1359 patients not taking antidepressants. The 750 patients on antidepressants in PDTS data had between 8 and 108 prescriptions dispensed, with a mean (SD) of 32 (12), vs a range of 1 to 65 prescriptions and a mean (SD) of 29 (12) for 1363 patients not taking antidepressants.

Discussion

The JPC DoD Cancer Registry includes information on cancer types, stages, treatment regimens, and physicians’ notes, while noncancer drugs are sourced from the PDTS database. The pharmacy uses a different documentation system, leading to varied classifications.

Database reliance has its drawbacks. For example, megestrol is coded as a cancer drug, although it’s primarily used for endometrial or gynecologic cancers. Many drugs have multiple therapeutic codes assigned to them, including 10 antineoplastic agents: diclofenac, Bacillus Calmette-Guérin (BCG), megestrol acetate, tamoxifen, anastrozole, letrozole, leuprolide, goserelin, degarelix, and fluorouracil. Diclofenac, BCG, and mitomycin have been repurposed for cancer treatment.^84-87 From 2003 to 2023, diclofenac was prescribed to 350 patients for mild-to-moderate pain, with only 2 patients receiving it for cancer in 2018. FDA-approved for bladder cancer in 1990, BCG was prescribed for cancer treatment for 1 patient in 2021 after being used for vaccines between 2003 and 2018. Tamoxifen, used for hormone receptor-positive breast cancer from 2004 to 2017 with 53 patients, switched to estrogen agonist-antagonists from 2017 to 2023 with 123 patients. Only a few of the 168 patients were prescribed tamoxifen using both codes.^88-91 Anastrozole and letrozole were coded as antiestrogens for 7 and 18 patients, respectively, while leuprolide and goserelin were coded as gonadotropins for 59 and 18 patients. Degarelix was coded as antigonadotropins, fluorouracil as skin and mucous membrane agents miscellaneous, and megestrol acetate as progestins for 7, 6, and 3 patients, respectively. Duloxetine was given to 186 patients, primarily for depression from 2005 to 2023, with 7 patients treated for fibromyalgia from 2022 to 2023.

Antidepressants Observed

Tables 1 and 5 provide insight into the FDA approval of 14 antineoplastics and antidepressants and their CYP metabolic pathways.^92-122 In Table 4, the most prescribed antidepressant classes are SNRIs, SRMs, SSRIs, TeCAs, NDRIs, and TCAs. This trend highlights a preference for newer medications with weak CYP inhibition. A total of 349 patients were prescribed SSRIs, 343 SNRIs, 119 SRMs, 109 TCAs, 83 TeCAs, and 79 NDRIs. MAOIs, SMRAs, and NMDARAs were not observed in this dataset. While there are instances of dextromethorphan-bupropion and sertraline-escitalopram being dispensed together, it remains unclear whether these were NMDARA combinations.

Among the 14 specific antineoplastic agents, 10 are metabolized by CYP isoenzymes, primarily CYP3A4. Duloxetine neither inhibits nor is metabolized by CYP3A4, a reason it is often recommended, following venlafaxine.

Both duloxetine and venlafaxine are used off-label for chemotherapy-induced peripheral neuropathy related to paclitaxel and docetaxel. According to the CYP metabolized pathway, duloxetine tends to have more favorable DDIs than venlafaxine. In PDTS data, 371 patients were treated with paclitaxel and 180 with docetaxel, with respective antidepressant prescriptions of 156 and 70. Of the 156 patients dispensed paclitaxel, 62 (40%) were dispensed with duloxetine compared to 43 (28%) with venlafaxine. Of the 70 patients dispensed docetaxel, 23 (33%) received duloxetine vs 24 (34%) with venlafaxine.

Of 85 patients prescribed duloxetine, 75 received it with either paclitaxel or docetaxel (5 received both). Five patients had documented AEs (1 neuropathy related). Of 67 patients prescribed venlafaxine, 66 received it with either paclitaxel or docetaxel. Two patients had documented AEs (1 was neuropathy related, the same patient who received duloxetine). Of the 687 patients treated with paclitaxel and 337 with docetaxel in all databases, 4 experienced neuropathic AEs from both medications.⁷⁹

Antidepressants can increase the risk of bleeding, especially when combined with blood thinners, and may elevate blood pressure, particularly alongside stimulants. Of the 554 patients prescribed 9 different anticoagulants, enoxaparin, apixaban, and rivaroxaban were the most common (each > 100 patients). Among these, 201 patients (36%) received both anticoagulants and antidepressants: duloxetine for 64 patients, venlafaxine for 30, trazodone for 35, and sertraline for 26. There were no data available to assess bleeding rates related to the evaluation of DDIs between these medication classes.

Antidepressants can be prescribed for erectile dysfunction. Of the 148 patients prescribed an antidepressant for erectile dysfunction, duloxetine, trazodone, and mirtazapine were the most common. Antidepressant preferences varied by cancer type. Duloxetine was the only antidepressant used for all types of cancer. Venlafaxine, duloxetine, trazodone, sertraline, and escitalopram were the most prescribed antidepressants for breast cancer, while duloxetine, mirtazapine, citalopram, sertraline, and trazodone were the most prescribed for lung cancer. Sertraline, duloxetine, trazodone, amitriptyline, and escitalopram were most common for testicular cancer. Duloxetine, venlafaxine, trazodone, amitriptyline, and sertraline were the most prescribed for endometrial cancer, while duloxetine, venlafaxine, amitriptyline, citalopram, and sertraline were most prescribed for ovarian cancer.

The broadness of International Statistical Classification of Diseases, Tenth Revision codes made it challenging to identify nondepression diagnoses in the analyzed population. However, if all antidepressants were prescribed to treat depression, service members with cancer exhibited a higher depression rate (35%) than the general population (25%). Of 2104 patients, 191 (9.1%) had mood disorders, and 706 (33.6%) had mental disorders: 346 (49.0%) had 1 diagnosis, and 360 (51.0%) had multiple diagnoses. The percentage of diagnoses varied yearly, with notable drops in 2003, 2007, 2011, 2014, and 2018, and peaks in 2006, 2008, 2013, 2017, and 2022. This fluctuation was influenced by events like the establishment of PDTS in 2002, the 2008 economic recession, a hospital relocation in 2011, the 2014 Ebola outbreak, and the COVID-19 pandemic. Although the number of patients receiving antidepressants increased from 2019 to 2022, the overall percentage of patients receiving them did not significantly change from 2003 to 2022, aligning with previous research.^5,125

Many medications have potential uses beyond what is detailed in the prescribing information. Antidepressants can relieve pain, while pain medications may help with depression. Opioids were once thought to effectively treat depression, but this perspective has changed with a greater understanding of their risks, including misuse.^126-131 Pain is a severe and often unbearable AE of cancer. Of 2113 patients, 92% received opioids; 34% received both opioids and antidepressants; 2% received only antidepressants; and 7% received neither. This study didn’t clarify whether those on opioids alone recognized their depression or if those on both were aware of their dependence. While SSRIs are generally not addictive, they can lead to physical dependence, and any medication can be abused if not managed properly.^132-134

Conclusions

This retrospective study analyzes data from antineoplastic agents used in systemic cancer treatment between 1988 and 2023, with a particular focus on the use of antidepressants. Data on antidepressant prescriptions are incomplete and specific to these agents, which means the findings cannot be generalized to all antidepressants. Hence, the results indicate that patients taking antidepressants had more diagnosed health issues and received more medications compared to patients who were not on these drugs.

This study underscores the need for further research into the effects of antidepressants on cancer treatment, utilizing all data from the DoD Cancer Registry. Future research should explore DDIs between antidepressants and other cancer and noncancer medications, as this study did not assess AE documentation, unlike in studies involving erlotinib, gefitinib, and paclitaxel.^78,79 Further investigation is needed to evaluate the impact of discontinuing antidepressant use during cancer treatment. This comprehensive overview provides insights for clinicians to help them make informed decisions regarding the prescription of antidepressants in the context of cancer treatment.

Cancer patients experience depression at rates > 5 times that of the general population.^1-11 Despite an increase in palliative care use, depression rates continued to rise.^2-4 Between 5% to 16% of outpatients, 4% to 14% of inpatients, and up to 49% of patients receiving palliative care experience depression.⁵ This issue also impacts families and caregivers.¹ A 2021 meta-analysis found that 23% of active military personnel and 20% of veterans experience depression.¹¹

Antidepressants approved by the US Food and Drug Administration (FDA) target the serotonin, norepinephrine, or dopamine systems and include boxed warnings about an increased risk of suicidal thoughts in adults aged 18 to 24 years.^12,13 These medications are categorized into several classes: monoamine oxidase inhibitors (MAOIs), tricyclic antidepressants (TCAs), tetracyclic antidepressants (TeCAs), norepinephrine-dopamine reuptake inhibitors (NDRIs), selective serotonin reuptake inhibitors (SSRIs), serotonin-norepinephrine reuptake inhibitors (SNRIs), serotonin receptor modulators (SRMs), serotonin-melatonin receptor antagonists (SMRAs), and N—methyl-D-aspartate receptor antagonists (NMDARAs).^14,15 The first FDA-approved antidepressants, iproniazid (an MAOI) and imipramine (a TCA) laid the foundation for the development of newer classes like SSRIs and SNRIs.^15-17

Older antidepressants such as MAOIs and TCAs are used less due to their adverse effects (AEs) and drug interactions. MAOIs, such as iproniazid, selegiline, moclobemide, tranylcypromine, isocarboxazid, and phenelzine, have numerous AEs and drug interactions, making them unsuitable for first- or second-line treatment of depression.^14,18-21 TCAs such as doxepin, amitriptyline, nortriptyline, imipramine, desipramine, clomipramine, trimipramine, protriptyline, maprotiline, and amoxapine have a narrow therapeutic index requiring careful monitoring for signs of toxicity such as QRS widening, tremors, or confusion. Despite the issues, TCAs are generally classified as second-line agents for major depressive disorder (MDD). TCAs have off-label uses for migraine prophylaxis, treatment of obsessive-compulsive disorder (OCD), insomnia, and chronic pain management first-line.^14,22-29

Newer antidepressants, including TeCAs and NDRIs, are typically more effective, but also come with safety concerns. TeCAs like mirtazapine interact with several medications, including MAOIs, serotonin-increasing drugs, alcohol, cannabidiol, and marijuana. Mirtazapine is FDA-approved for the treatment of moderate to severe depression in adults. It is also used off-label to treat insomnia, panic disorder, posttraumatic stress disorder (PTSD), generalized anxiety disorder (GAD), social anxiety disorder (SAD), headaches, and migraines. Compared to other antidepressants, mirtazapine is effective for all stages of depression and addresses a broad range of related symptoms.^14,30-34 NDRIs, such as bupropion, also interact with various medications, including MAOIs, other antidepressants, stimulants, and alcohol. Bupropion is FDA-approved for smoking cessation and to treat depression and SAD. It is also used off-label for depression- related bipolar disorder or sexual dysfunction, attention-deficit/hyperactivity disorder (ADHD), and obesity.^14,35-42

SSRIs, SNRIs, and SRMs should be used with caution. SSRIs such as sertraline, citalopram, escitalopram, fluoxetine, paroxetine, and fluvoxamine are first-line treatments for depression and various psychiatric disorders due to their safety and efficacy. Common AEs of SSRIs include sexual dysfunction, sleep disturbances, weight changes, and gastrointestinal (GI) issues. SSRIs can prolong the QT interval, posing a risk of life-threatening arrhythmia, and may interact with other medications, necessitating treatment adjustments. The FDA approved SSRIs for MDD, GAD, bulimia nervosa, bipolar depression, OCD, panic disorder, premenstrual dysphoric disorder, treatment-resistant depression, PTSD, and SAD. Off-label uses include binge eating disorder, body dysmorphic disorder, fibromyalgia, premature ejaculation, paraphilias, autism, Raynaud phenomenon, and vasomotor symptoms associated with menopause. Among SSRIs, sertraline and escitalopram are noted for their effectiveness and tolerability.^14,43-53

SNRIs, including duloxetine, venlafaxine, desvenlafaxine, milnacipran, and levomilnacipran, may increase bleeding risk, especially when taken with blood thinners. They can also elevate blood pressure, which may worsen if combined with stimulants. SNRIs may interact with other medications that affect serotonin levels, increasing the risk of serotonin syndrome when taken with triptans, pain medications, or other antidepressants.¹⁴ Desvenlafaxine has been approved by the FDA (but not by the European Medicines Agency).^54-56 Duloxetine is FDA-approved for the treatment of depression, neuropathic pain, anxiety disorders, fibromyalgia, and musculoskeletal disorders. It is used off-label to treat chemotherapy-induced peripheral neuropathy and stress urinary incontinence.^57-61 Venlafaxine is FDA-approved for depression, SAD, and panic disorder, and is prescribed off-label to treat ADHD, neuropathy, fibromyalgia, cataplexy, and PTSD, either alone or in combination with other medications.^62,63 Milnacipran is not approved for MDD; levomilnacipran received approval in 2013.⁶⁴

SRMs such as trazodone, nefazodone, vilazodone, and vortioxetine also function as serotonin reuptake inhibitors.^14,15 Trazodone is FDA-approved for MDD. It has been used off-label to treat anxiety, Alzheimer disease, substance misuse, bulimia nervosa, insomnia, fibromyalgia, and PTSD when first-line SSRIs are ineffective. A notable AE of trazodone is orthostatic hypotension, which can lead to dizziness and increase the risk of falls, especially in geriatric patients.^65-70 Nefazodone was discontinued in Europe in 2003 due to rare cases of liver toxicity but remains available in the US.^71-74 Vilazodone and vortioxetine are FDA-approved.

The latest classes of antidepressants include SMRAs and NMDARAs.¹⁴ Agomelatine, an SMRA, was approved in Europe in 2009 but rejected by the FDA in 2011 due to liver toxicity.⁷⁵ NMDARAs like esketamine and a combination of dextromethorphan and bupropion received FDA approval in 2019 and 2022, respectively.^76,77

This retrospective study analyzes noncancer drugs used during systemic chemotherapy based on a dataset of 14 antineoplastic agents. It sought to identify the most dispensed noncancer drug groups, discuss findings, compare patients with and without antidepressant prescriptions, and examine trends in antidepressant use from 2002 to 2023. This analysis expands on prior research.^78-81

Methods

The Walter Reed National Military Medical Center Institutional Review Board approved the study protocol and ensured compliance with the Health Insurance Portability and Accountability Act as an exempt protocol. The Joint Pathology Center (JPC) of the US Department of Defense (DoD) Cancer Registry Program and Military Health System (MHS) data experts from the Comprehensive Ambulatory/Professional Encounter Record (CAPER) and Pharmacy Data Transaction Service (PDTS) provided data for the analysis.

Data Sources

The JPC DoD Cancer Registry Program contains data from 1998 to 2024. CAPER and PDTS are part of the MHS Data Repository/Management Analysis and Reporting Tool database. Each observation in CAPER represents an ambulatory encounter at a military treatment facility (MTF). CAPER records are available from 2003 to 2024. PDTS records are available from 2002 to 2004. Each observation in PDTS represents a prescription filled for an MHS beneficiary, excluding those filled at international civilian pharmacies and inpatient pharmacy prescriptions.

This cross-sectional analysis requested data extraction for specific cancer drugs from the DoD Cancer Registry, focusing on treatment details, diagnosis dates, patient demographics, and physicians’ comments on AEs. After identifying patients, CAPER was used to identify additional health conditions. PDTS was used to compile a list of prescription medications filled during systemic cancer treatment or < 2 years postdiagnosis.

The 2016 Surveillance, Epidemiology, and End Results Program Coding and Staging Manual and International Classification of Diseases for Oncology, 3rd edition, 1st revision, were used to decode disease and cancer types.^82,83 Data sorting and analysis were performed using Microsoft Excel. The percentage for the total was calculated by using the number of patients or data available within the subgroup divided by the total number of patients or data variables. To compare the mean number of dispensed antidepressants to those without antidepressants, a 2-tailed, 2-sample z test was used to calculate the P value and determine statistical significance (P < .05) using socscistatistics.com.

Data were extracted 3 times between 2021 and 2023. The initial 2021 protocol focused on erlotinib and gefitinib. A modified protocol in 2022 added paclitaxel, cisplatin, docetaxel, pemetrexed, and crizotinib; further modification in 2023 included 8 new antineoplastic agents and 2 anticoagulants. Sotorasib has not been prescribed in the MHS, and JPC lacks records for noncancer drugs. The 2023 dataset comprised 2210 patients with cancer treated with 14 antineoplastic agents; 2104 had documented diagnoses and 2113 had recorded prescriptions. Data for erlotinib, gefitinib, and paclitaxel have been published previously.^78,79

Results

Of 2113 patients with recorded prescriptions, 1297 patients (61.4%) received 109 cancer drugs, including 96 antineoplastics, 7 disease-modifying antirheumatic agents, 4 biologic response modifiers, and 2 calcitonin gene-related peptides. Fourteen antineoplastic agents had complete data from JPC, while others were noted for combination therapies or treatment switches from the PDTS (Table 1). Seventy-six cancer drugs were prescribed with antidepressants in 489 patients (eAppendix).

The JPC provided 2242 entries for 2210 patients, ranging in age from 2 months to 88 years (mean, 56 years), documenting treatment from September 1988 to January 2023. Thirty-two patients had duplicate entries due to multiple cancer locations or occurrences. Of the 2242 patients, 1541 (68.7%) were aged > 50 years, 975 patients (43.5%) had cancers that were stage III or IV, and 1267 (56.5%) had cancers that were stage 0, I, II, or not applicable/unknown. There were 51 different types of cancer: breast, lung, testicular, endometrial, and ovarian were most common (n ≥ 100 patients). Forty-two cancer types were documented among 750 patients prescribed antidepressants (Table 2).

The CAPER database recorded 8882 unique diagnoses for 2104 patients, while PDTS noted 1089 unique prescriptions within 273 therapeutic codes for 2113 patients. Nine therapeutic codes (opiate agonists, adrenals, cathartics-laxatives, nonsteroidal anti-inflammatory agents, antihistamines for GI conditions, 5-HT3 receptor antagonists, analgesics and antipyretic miscellanea, antineoplastic agents, and proton-pump inhibitors) and 8 drugs (dexamethasone, prochlorperazine, ondansetron, docusate, acetaminophen, ibuprofen, oxycodone, and polyethylene glycol 3350) were associated with > 1000 patients (≥ 50%). Patients had between 1 and 275 unique health conditions and filled 1 to 108 prescriptions. The mean (SD) number of diagnoses and prescriptions was 50 (28) and 29 (12), respectively. Of the 273 therapeutic codes, 30 groups were analyzed, with others categorized into miscellaneous groups such as lotions, vaccines, and devices. Significant differences in mean number of prescriptions were found for patients taking antidepressants compared to those not (P < .05), except for anticonvulsants and antipsychotics (P = .12 and .09, respectively) (Table 3).

Antidepressants

Of the 2113 patients with recorded prescriptions, 750 (35.5%) were dispensed 17 different antidepressants. Among these 17 antidepressants, 183 (8.7%) patients received duloxetine, 158 (7.5%) received venlafaxine, 118 (5.6%) received trazodone, and 107 (5.1%) received sertraline (Figure 1, Table 4). Of the 750 patients, 509 (67.9%) received 1 antidepressant, 168 (22.4%) received 2, 60 (8.0%) received 3, and 13 (1.7%) received > 3. Combinations varied, but only duloxetine and trazodone were prescribed to > 10 patients.

Antidepressants were prescribed annually at an overall mean (SD) rate of 23% (5%) from 2003 to 2022 (Figure 2). Patients on antidepressants during systemic therapy had a greater number of diagnosed medical conditions and received more prescription medications compared to those not taking antidepressants (P < .001) (Figure 3). The 745 patients taking antidepressants in CAPER data had between 1 and 275 diagnosed medical issues, with a mean (SD) of 55 (31) vs a range of 1 to 209 and a mean (SD) of 46 (26) for the 1359 patients not taking antidepressants. The 750 patients on antidepressants in PDTS data had between 8 and 108 prescriptions dispensed, with a mean (SD) of 32 (12), vs a range of 1 to 65 prescriptions and a mean (SD) of 29 (12) for 1363 patients not taking antidepressants.

Discussion

The JPC DoD Cancer Registry includes information on cancer types, stages, treatment regimens, and physicians’ notes, while noncancer drugs are sourced from the PDTS database. The pharmacy uses a different documentation system, leading to varied classifications.

Database reliance has its drawbacks. For example, megestrol is coded as a cancer drug, although it’s primarily used for endometrial or gynecologic cancers. Many drugs have multiple therapeutic codes assigned to them, including 10 antineoplastic agents: diclofenac, Bacillus Calmette-Guérin (BCG), megestrol acetate, tamoxifen, anastrozole, letrozole, leuprolide, goserelin, degarelix, and fluorouracil. Diclofenac, BCG, and mitomycin have been repurposed for cancer treatment.^84-87 From 2003 to 2023, diclofenac was prescribed to 350 patients for mild-to-moderate pain, with only 2 patients receiving it for cancer in 2018. FDA-approved for bladder cancer in 1990, BCG was prescribed for cancer treatment for 1 patient in 2021 after being used for vaccines between 2003 and 2018. Tamoxifen, used for hormone receptor-positive breast cancer from 2004 to 2017 with 53 patients, switched to estrogen agonist-antagonists from 2017 to 2023 with 123 patients. Only a few of the 168 patients were prescribed tamoxifen using both codes.^88-91 Anastrozole and letrozole were coded as antiestrogens for 7 and 18 patients, respectively, while leuprolide and goserelin were coded as gonadotropins for 59 and 18 patients. Degarelix was coded as antigonadotropins, fluorouracil as skin and mucous membrane agents miscellaneous, and megestrol acetate as progestins for 7, 6, and 3 patients, respectively. Duloxetine was given to 186 patients, primarily for depression from 2005 to 2023, with 7 patients treated for fibromyalgia from 2022 to 2023.

Antidepressants Observed

Tables 1 and 5 provide insight into the FDA approval of 14 antineoplastics and antidepressants and their CYP metabolic pathways.^92-122 In Table 4, the most prescribed antidepressant classes are SNRIs, SRMs, SSRIs, TeCAs, NDRIs, and TCAs. This trend highlights a preference for newer medications with weak CYP inhibition. A total of 349 patients were prescribed SSRIs, 343 SNRIs, 119 SRMs, 109 TCAs, 83 TeCAs, and 79 NDRIs. MAOIs, SMRAs, and NMDARAs were not observed in this dataset. While there are instances of dextromethorphan-bupropion and sertraline-escitalopram being dispensed together, it remains unclear whether these were NMDARA combinations.

Among the 14 specific antineoplastic agents, 10 are metabolized by CYP isoenzymes, primarily CYP3A4. Duloxetine neither inhibits nor is metabolized by CYP3A4, a reason it is often recommended, following venlafaxine.

Both duloxetine and venlafaxine are used off-label for chemotherapy-induced peripheral neuropathy related to paclitaxel and docetaxel. According to the CYP metabolized pathway, duloxetine tends to have more favorable DDIs than venlafaxine. In PDTS data, 371 patients were treated with paclitaxel and 180 with docetaxel, with respective antidepressant prescriptions of 156 and 70. Of the 156 patients dispensed paclitaxel, 62 (40%) were dispensed with duloxetine compared to 43 (28%) with venlafaxine. Of the 70 patients dispensed docetaxel, 23 (33%) received duloxetine vs 24 (34%) with venlafaxine.

Of 85 patients prescribed duloxetine, 75 received it with either paclitaxel or docetaxel (5 received both). Five patients had documented AEs (1 neuropathy related). Of 67 patients prescribed venlafaxine, 66 received it with either paclitaxel or docetaxel. Two patients had documented AEs (1 was neuropathy related, the same patient who received duloxetine). Of the 687 patients treated with paclitaxel and 337 with docetaxel in all databases, 4 experienced neuropathic AEs from both medications.⁷⁹

Antidepressants can increase the risk of bleeding, especially when combined with blood thinners, and may elevate blood pressure, particularly alongside stimulants. Of the 554 patients prescribed 9 different anticoagulants, enoxaparin, apixaban, and rivaroxaban were the most common (each > 100 patients). Among these, 201 patients (36%) received both anticoagulants and antidepressants: duloxetine for 64 patients, venlafaxine for 30, trazodone for 35, and sertraline for 26. There were no data available to assess bleeding rates related to the evaluation of DDIs between these medication classes.

Antidepressants can be prescribed for erectile dysfunction. Of the 148 patients prescribed an antidepressant for erectile dysfunction, duloxetine, trazodone, and mirtazapine were the most common. Antidepressant preferences varied by cancer type. Duloxetine was the only antidepressant used for all types of cancer. Venlafaxine, duloxetine, trazodone, sertraline, and escitalopram were the most prescribed antidepressants for breast cancer, while duloxetine, mirtazapine, citalopram, sertraline, and trazodone were the most prescribed for lung cancer. Sertraline, duloxetine, trazodone, amitriptyline, and escitalopram were most common for testicular cancer. Duloxetine, venlafaxine, trazodone, amitriptyline, and sertraline were the most prescribed for endometrial cancer, while duloxetine, venlafaxine, amitriptyline, citalopram, and sertraline were most prescribed for ovarian cancer.

The broadness of International Statistical Classification of Diseases, Tenth Revision codes made it challenging to identify nondepression diagnoses in the analyzed population. However, if all antidepressants were prescribed to treat depression, service members with cancer exhibited a higher depression rate (35%) than the general population (25%). Of 2104 patients, 191 (9.1%) had mood disorders, and 706 (33.6%) had mental disorders: 346 (49.0%) had 1 diagnosis, and 360 (51.0%) had multiple diagnoses. The percentage of diagnoses varied yearly, with notable drops in 2003, 2007, 2011, 2014, and 2018, and peaks in 2006, 2008, 2013, 2017, and 2022. This fluctuation was influenced by events like the establishment of PDTS in 2002, the 2008 economic recession, a hospital relocation in 2011, the 2014 Ebola outbreak, and the COVID-19 pandemic. Although the number of patients receiving antidepressants increased from 2019 to 2022, the overall percentage of patients receiving them did not significantly change from 2003 to 2022, aligning with previous research.^5,125

Many medications have potential uses beyond what is detailed in the prescribing information. Antidepressants can relieve pain, while pain medications may help with depression. Opioids were once thought to effectively treat depression, but this perspective has changed with a greater understanding of their risks, including misuse.^126-131 Pain is a severe and often unbearable AE of cancer. Of 2113 patients, 92% received opioids; 34% received both opioids and antidepressants; 2% received only antidepressants; and 7% received neither. This study didn’t clarify whether those on opioids alone recognized their depression or if those on both were aware of their dependence. While SSRIs are generally not addictive, they can lead to physical dependence, and any medication can be abused if not managed properly.^132-134

Conclusions

This retrospective study analyzes data from antineoplastic agents used in systemic cancer treatment between 1988 and 2023, with a particular focus on the use of antidepressants. Data on antidepressant prescriptions are incomplete and specific to these agents, which means the findings cannot be generalized to all antidepressants. Hence, the results indicate that patients taking antidepressants had more diagnosed health issues and received more medications compared to patients who were not on these drugs.

This study underscores the need for further research into the effects of antidepressants on cancer treatment, utilizing all data from the DoD Cancer Registry. Future research should explore DDIs between antidepressants and other cancer and noncancer medications, as this study did not assess AE documentation, unlike in studies involving erlotinib, gefitinib, and paclitaxel.^78,79 Further investigation is needed to evaluate the impact of discontinuing antidepressant use during cancer treatment. This comprehensive overview provides insights for clinicians to help them make informed decisions regarding the prescription of antidepressants in the context of cancer treatment.

Screening for colorectal cancer (CRC) with colonoscopy enables the identification and removal of CRC precursors (colonic adenomas) and has been associated with reduced risk of CRC incidence and mortality.^1-3 Furthermore, there is consensus that diet and lifestyle may be associated with forestalling CRC pathogenesis at the intermediate adenoma stages.^4-7 However, studies have shown that US veterans have poorer diet quality and a higher risk for neoplasia compared with nonveterans, reinforcing the need for tailored clinical approaches.^8,9 Combining screening with conversations about modifiable environmental and lifestyle risk factors, such as poor diet, is a highly relevant and possibly easily leveraged prevention for those at high risk. However, there is limited evidence for any particular dietary patterns or dietary features that are most important over time.⁷

Several dietary components have been shown to be associated with CRC risk,10 either as potentially chemopreventive (fiber, fruits and vegetables,¹¹ dairy,¹² supplemental vitamin D,¹³ calcium,¹⁴ and multivitamins¹⁵) or carcinogenic (red meat¹⁶ and alcohol¹⁷). Previous studies of veterans have similarly shown that higher intake of fiber and vitamin D reduced risk, and red meat is associated with an increased risk for finding CRC precursors during colonoscopy.¹⁸ However, these dietary categories are often analyzed in isolation. Studying healthy dietary patterns in aggregate may be more clinically relevant and easier to implement for prevention of CRC and its precursors.^19-21 Healthy dietary patterns, such as the US Dietary Guidelines for Americans represented by the Healthy Eating Index (HEI), the Mediterranean diet (MD), and the Dietary Approaches to Stop Hypertension (DASH) diet, have been associated with lower risk for chronic disease.^22-24 Despite the extant literature, no known studies have compared these dietary patterns for associations with risk of CRC precursor or CRC development among US veterans undergoing long-term screening and follow-up after a baseline colonoscopy.

The objective of this study was to test for associations between baseline scores of healthy dietary patterns and the most severe colonoscopy findings (MSCFs) over ≥ 10 years following a baseline screening colonoscopy in veterans.

Methods

Participants in the Cooperative Studies Program (CSP) #380 cohort study included 3121 asymptomatic veterans aged 50 to 75 years at baseline who had consented to initial screening colonoscopy between 1994 and 1997, with subsequent follow-up and surveillance.²⁵ Prior to their colonoscopy, all participants completed a baseline study survey that included questions about cancer risk factors including family history of CRC, diet, physical activity, and medication use.

Included in this cross-sectional analysis were data from a sample of veteran participants of the CSP #380 cohort with 1 baseline colonoscopy, follow-up surveillance through 2009, a cancer risk factor survey collected at baseline, and complete demographic and clinical indicator data. Excluded from the analysis were 67 participants with insufficient responses to the dietary food frequency questionnaire (FFQ) and 31 participants with missing body mass index (BMI), 3023 veterans.

Measures

MSCF. The outcome of interest in this study was the MSCF recorded across all participant colonoscopies during the study period. MSCF was categorized as either (1) no neoplasia; (2) < 2 nonadvanced adenomas, including small adenomas (diameter < 10 mm) with tubular histology; or (3) advanced neoplasia (AN), which is characterized by adenomas > 10 mm in diameter, with villous histology, with high-grade dysplasia, or CRC.

Dietary patterns. Dietary pattern scores representing dietary quality and calculated based on recommendations of the US Dietary Guidelines for Americans using the HEI, MD, and DASH diets were independent variables.^26-28 These 3 dietary patterns were chosen for their hypothesized relationship with CRC risk, but each weighs food categories differently (Appendix 1).^22-24,29 Dietary pattern scores were calculated using the CSP #380 self-reported responses to 129 baseline survey questions adapted from a well-established and previously validated semiquantitative FFQ.³⁰ The form was administered by mail twice to a sample of 127 participants at baseline and at 1 year. During this interval, men completed 1-week diet records twice, spaced about 6 months apart. Mean values for intake of most nutrients assessed by the 2 methods were similar. Intraclass correlation coefficients for the baseline and 1-year FFQ-assessed nutrient intakes that ranged from 0.47 for vitamin E (without supplements) to 0.80 for vitamin C (with supplements). Correlation coefficients between the energy-adjusted nutrient intakes were measured by diet records and the 1-year FFQ, which asked about diet during the year encompassing the diet records. Higher raw and percent scores indicated better alignment with recommendations from each respective dietary pattern. Percent scores were calculated as a standardizing method and used in analyses for ease of comparing the dietary patterns. Scoring can be found in Appendix 2.

Demographic characteristics and clinical indicators. Demographic characteristics included age categories, sex, and race/ethnicity. Clinical indicators included BMI, the number of comorbid conditions used to calculate the Charlson Comorbidity Index, family history of CRC in first-degree relatives, number of follow-up colonoscopies across the study period, and food-based vitamin D intake.³¹ These variables were chosen for their applicability found in previous CSP #380 cohort studies.^18,32,33 Self-reported race and ethnicity were collapsed due to small numbers in some groups. The authors acknowledge these are distinct concepts and the variable has limited utility other than for controlling for systemic racism in the model.

Statistical Analyses

Descriptive statistics were used to describe distributional assumptions for all variables, including demographics, clinical indicators, colonoscopy results, and dietary patterns. Pairwise correlations between the total dietary pattern scores and food category scores were calculated with Pearson correlation (r).

Multinomial logistic regression models were created using SAS procedure LOGISTIC with the outcome of the categorical MSCF (no neoplasia, nonadvanced adenoma, or AN).³⁴ A model was created for each independent predictor variable of interest (ie, the HEI, MD, or DASH percentage-standardized dietary pattern score and each food category comprising each dietary pattern score). All models were adjusted for age, sex, race/ethnicity, BMI, number of comorbidities, family history of CRC, number of follow-up colonoscopies, and estimated daily food-derived vitamin D intake. The demographic and clinical indicators were included in the models as they are known to be associated with CRC risk.¹⁸ The number of colonoscopies was included to control for surveillance intensity presuming risk for AN is reduced as polyps are removed. Because colonoscopy findings from an initial screening have unique clinical implications compared with follow- up and surveillance, MSCF was observed in 2 ways in sensitivity analyses: (1) baseline and (2) aggregate follow-up and surveillance only, excluding baseline findings.

Adjusted odds ratios (aORs) and 95% CIs for each of the MSCF outcomes with a reference finding of no neoplasia for the models are presented. We chose not to adjust for multiple comparisons across the different dietary patterns given the correlation between dietary pattern total and category scores but did adjust for multiple comparisons for dietary categories within each dietary pattern. Tests for statistical significance used α= .05 for the dietary pattern total scores and P values for the dietary category scores for each dietary pattern controlled for false discovery rate using the MULTTEST SAS procedure.35 All data manipulations and analyses were performed using SAS version 9.4.

Results

The study included 3023 patients. All were aged 50 to 75 years, 2923 (96.7%) were male and 2532 (83.8%) were non-Hispanic White (Table 1). Most participants were overweight or obese (n = 2535 [83.8%]), 2024 (67.0%) had ≤ 2 comorbidities, and 2602 (86.1%) had no family history of CRC. The MSCF for 1628 patients (53.9%) was no neoplasia, 966 patients (32.0%) was nonadvanced adenoma, and 429 participants (14.2%) had AN.

Mean percent scores were 58.5% for HEI, 38.2% for MD, and 63.1% for the DASH diet, with higher percentages indicating greater alignment with the recommendations for each diet (Table 2). All 3 dietary patterns scores standardized to percentages were strongly and significantly correlated in pairwise comparisons: HEI:MD, r = 0.62 (P < .001); HEI:DASH, r = 0.60 (P < .001); and MD:DASH, r = 0.72 (P < .001). Likewise, food category scores were significantly correlated across dietary patterns. For example, whole grain and fiber values from each dietary score were strongly correlated in pairwise comparisons: HEI Whole Grain:MD Grain, r = 0.64 (P < .001); HEI Whole Grain:DASH Fiber, r = 0.71 (P < .001); and MD Grain:DASH Fiber, r = 0.70 (P < .001).

Associations between individual participants' dietary pattern scores and the outcome of their pooled MSCF from baseline screening and ≥ 10 years of surveillance are presented in Table 3. For each single-point increases in dietary pattern scores (reflecting better dietary quality), aORs for nonadvanced adenoma vs no neoplasia were slightly lower but not statistically significantly: HEI, aOR, 1.00 (95% CI, 0.99-1.01); MD, aOR, 0.98 (95% CI, 0.94-1.02); and DASH, aOR, 0.99 (95% CI, 0.99-1.00). aORs for AN vs no neoplasia were slightly lower for each dietary pattern assessed, and only the MD and DASH scores were significantly different from 1.00: HEI, aOR, 1.00 (95% CI, 0.99-1.01); MD, aOR, 0.95 (95% CI, 0.90-1.00); and DASH, aOR, 0.99 (95% CI, 0.98-1.00).

We observed lower odds for nonadvanced adenoma and AN among all these dietary patterns when there was greater alignment with the recommended intake of whole grains and fiber. In separate models conducted using food categories comprising the dietary patterns as independent variables and after correcting for multiple tests, higher scores for the HEI Refined Grain category were associated with higher odds for nonadvanced adenoma (aOR, 1.03 [95% CI, 1.01-1.05]; P = .01) and AN (aOR, 1.05 [95% CI, 1.02-1.08]; P < .001). Higher scores for the HEI Whole Grain category were associated with lower odds for nonadvanced adenoma (aOR, 0.97 [95% CI, 0.95-0.99]; P = .01) and AN (aOR, 0.96 [95% CI, 0.93-0.99]; P = .01). Higher scores for the MD Grain category were significantly associated with lower odds for nonadvanced adenoma (aOR, 0.44 [95% CI, 0.26-0.75]; P = .002) and AN (aOR, 0.29 [95% CI, 0.14-0.62]; P = .001). The DASH Grains category also was significantly associated with lower odds for AN (aOR, 0.86 [95% CI, 0.78-0.95]; P = .002).

Discussion

In this study of 3023 veterans undergoing first-time screening colonoscopy and ≥ 10 years of surveillance, we found that healthy dietary patterns, as assessed by the MD and DASH diet, were significantly associated with lower risk of AN. Additionally, we identified lower odds for AN and nonadvanced adenoma compared with no neoplasia for higher grain scores for all the dietary patterns studied. Other food categories that comprise the dietary pattern scores had mixed associations with the MSCF outcomes. Several other studies have examined associations between dietary patterns and risk for CRC but to our knowledge, no studies have explored these associations among US veterans.

These results also indicate study participants had better than average (based on a 50% threshold) dietary quality according to the HEI and DASH diet scoring methods we used, but poor dietary quality according to the MD scoring method. The mean HEI scores for the present study were higher than a US Department of Agriculture study by Dong et al that compared dietary quality between veterans and nonveterans using the HEI, for which veterans’ expected HEI score was 45.6 of 100.⁸ This could be explained by the fact that the participants needed to be healthy to be eligible and those with healthier behaviors overall may have self-selected into the study due to motivation for screening during a time when screening was not yet commonplace. ³⁶ Similarly, participants of the present study had higher adherence to the DASH diet (63.1%) than adolescents with diabetes in a study by Günther et al. Conversely, firefighters who were coached to use a Mediterranean-style dietary pattern and dietary had higher adherence to MD than did participants in this study.²⁷

A closer examination of specific food category component scores that comprise the 3 distinct dietary patterns revealed mixed results from the multinomial modeling, which may have to do with the guideline thresholds used to calculate the dietary scores. When analyzed separately in the logistic regression models for their associations with nonadvanced adenomas and AN compared with no neoplasia, higher MD and DASH fruit scores (but not HEI fruit scores) were found to be significant. Other studies have had mixed findings when attempting to test for associations of fruit intake with adenoma recurrence.^10,37

This study had some unexpected findings. Vegetable intake was not associated with nonadvanced adenomas or AN risk. Studies of food categories have consistently found vegetable (specifically cruciferous ones) intake to be linked with lower odds for cancers.³⁸ Likewise, the red meat category, which was only a unique food category in the MD score, was not associated with nonadvanced adenomas or AN. Despite consistent literature suggesting higher intake of red meat and processed meats increases CRC risk, in 2019 the Nutritional Recommendations Consortium indicated that the evidence was weak.^39,40 This study showed higher DASH diet scores for low-fat dairy, which were maximized when participants reported at least 50% of their dairy servings per day as being low-fat, had lower odds for AN. Yet, the MD scores for low-fat dairy had no association with either outcome; their calculation was based on total number of servings per week. This difference in findings suggests the fat intake ratio may be more relevant to CRC risk than intake quantity.

The literature is mixed regarding fatty acid intake and CRC risk, which may be relevant to both dairy and meat intake. One systematic review and meta-analysis found dietary fat and types of fatty acid intake had no association with CRC risk.⁴¹ However, a more recent meta-analysis that assessed both dietary intake and plasma levels of fatty acids did find some statistically significant differences for various types of fatty acids and CRC risk.⁴²

The findings in the present study that grain intake is associated with lower odds for more severe colonoscopy findings among veterans are notable.⁴³ Lieberman et al, using the CSP #380 data, found that cereal fiber intake was associated with a lower odds for AN compared with hyperplastic polyps (OR, 0.98 [95% CI, 0.96- 1.00]).¹⁸ Similarly, Hullings et al determined that older adults in the highest quintile of cereal fiber intake had significantly lower odds of CRC than those in lower odds for CRC when compared with lowest quintile (OR, 0.89 [95% CI, 0.83- 0.96]; P < .001).⁴⁴ These findings support existing guidance that prioritizes whole grains as a key source of dietary fiber for CRC prevention.

A recent literature review on fiber, fat, and CRC risk suggested a consensus regarding one protective mechanism: dietary fiber from grains modulates the gut microbiota by promoting butyrate synthesis.⁴⁵ Butyrate is a short-chain fatty acid that supports energy production in colonocytes and has tumor-suppressing properties.⁴⁶ Our findings suggest there could be more to learn about the relationship between butyrate production and reduction of CRC risk through metabolomic studies that use measurements of plasma butyrate. These studies may examine associations between not just a singular food or food category, but rather food patterns that include fruits, vegetables, nuts and seeds, and whole grains known to promote butyrate production and plasma butyrate.⁴⁷

Improved understanding of mechanisms and risk-modifying lifestyle factors such as dietary patterns may enhance prevention strategies. Identifying the collective chemopreventive characteristics of a specific dietary pattern (eg, MD) will be helpful to clinicians and health care staff to promote healthy eating to reduce cancer risk. More studies are needed to understand whether such promotion is more clinically applicable and effective for patients, as compared with eating more or less of specific foods (eg, more whole grains, less red meat). Furthermore, considering important environmental factors collectively beyond dietary patterns may offer a way to better tailor screening and implement a variety of lifestyle interventions. In the literature, this is often referred to as a teachable moment when patients’ attentions are captured and may position them to be more receptive to guidance.⁴⁸

Limitations

This study has several important limitations and leaves opportunities for future studies that explore the role of dietary patterns and AN or CRC risk. First, the FFQ data used to calculate dietary pattern scores used in analysis were only captured at baseline, and there are nearly 3 decades across the study period. However, it is widely assumed that the diets of older adults, like those included in this study, remain stable over time which is appropriate given our sample population was aged 50 to 75 years when the baseline FFQ data were collected.^49-51 Additionally, while the HEI is a well-documented, standard scoring method for dietary quality, there are multitudes of dietary pattern scoring approaches for MD and DASH.^23,52,53 Finally, findings from this study using the sample of veterans may not be generalizable to a broader population. Future longitudinal studies that test for a clinically significant change threshold are warranted.

Conclusion

Results of this study suggest future research should further explore the effects of dietary patterns, particularly intake of specific food groups in combination, as opposed to individual nutrients or food items, on AN and CRC risk. Possible studies might explore these dietary patterns for their mechanistic role in altering the microbiome metabolism, which may influence CRC outcomes or include diet in a more comprehensive, holistic risk score that could be used to predict colonic neoplasia risk or in intervention studies that assess the effects of dietary changes on long-term CRC prevention. We suggest there are differences in people’s dietary intake patterns that might be important to consider when implementing tailored approaches to CRC risk mitigation.

Screening for colorectal cancer (CRC) with colonoscopy enables the identification and removal of CRC precursors (colonic adenomas) and has been associated with reduced risk of CRC incidence and mortality.^1-3 Furthermore, there is consensus that diet and lifestyle may be associated with forestalling CRC pathogenesis at the intermediate adenoma stages.^4-7 However, studies have shown that US veterans have poorer diet quality and a higher risk for neoplasia compared with nonveterans, reinforcing the need for tailored clinical approaches.^8,9 Combining screening with conversations about modifiable environmental and lifestyle risk factors, such as poor diet, is a highly relevant and possibly easily leveraged prevention for those at high risk. However, there is limited evidence for any particular dietary patterns or dietary features that are most important over time.⁷

Several dietary components have been shown to be associated with CRC risk,10 either as potentially chemopreventive (fiber, fruits and vegetables,¹¹ dairy,¹² supplemental vitamin D,¹³ calcium,¹⁴ and multivitamins¹⁵) or carcinogenic (red meat¹⁶ and alcohol¹⁷). Previous studies of veterans have similarly shown that higher intake of fiber and vitamin D reduced risk, and red meat is associated with an increased risk for finding CRC precursors during colonoscopy.¹⁸ However, these dietary categories are often analyzed in isolation. Studying healthy dietary patterns in aggregate may be more clinically relevant and easier to implement for prevention of CRC and its precursors.^19-21 Healthy dietary patterns, such as the US Dietary Guidelines for Americans represented by the Healthy Eating Index (HEI), the Mediterranean diet (MD), and the Dietary Approaches to Stop Hypertension (DASH) diet, have been associated with lower risk for chronic disease.^22-24 Despite the extant literature, no known studies have compared these dietary patterns for associations with risk of CRC precursor or CRC development among US veterans undergoing long-term screening and follow-up after a baseline colonoscopy.

The objective of this study was to test for associations between baseline scores of healthy dietary patterns and the most severe colonoscopy findings (MSCFs) over ≥ 10 years following a baseline screening colonoscopy in veterans.

Methods

Participants in the Cooperative Studies Program (CSP) #380 cohort study included 3121 asymptomatic veterans aged 50 to 75 years at baseline who had consented to initial screening colonoscopy between 1994 and 1997, with subsequent follow-up and surveillance.²⁵ Prior to their colonoscopy, all participants completed a baseline study survey that included questions about cancer risk factors including family history of CRC, diet, physical activity, and medication use.

Included in this cross-sectional analysis were data from a sample of veteran participants of the CSP #380 cohort with 1 baseline colonoscopy, follow-up surveillance through 2009, a cancer risk factor survey collected at baseline, and complete demographic and clinical indicator data. Excluded from the analysis were 67 participants with insufficient responses to the dietary food frequency questionnaire (FFQ) and 31 participants with missing body mass index (BMI), 3023 veterans.

Measures

MSCF. The outcome of interest in this study was the MSCF recorded across all participant colonoscopies during the study period. MSCF was categorized as either (1) no neoplasia; (2) < 2 nonadvanced adenomas, including small adenomas (diameter < 10 mm) with tubular histology; or (3) advanced neoplasia (AN), which is characterized by adenomas > 10 mm in diameter, with villous histology, with high-grade dysplasia, or CRC.

Dietary patterns. Dietary pattern scores representing dietary quality and calculated based on recommendations of the US Dietary Guidelines for Americans using the HEI, MD, and DASH diets were independent variables.^26-28 These 3 dietary patterns were chosen for their hypothesized relationship with CRC risk, but each weighs food categories differently (Appendix 1).^22-24,29 Dietary pattern scores were calculated using the CSP #380 self-reported responses to 129 baseline survey questions adapted from a well-established and previously validated semiquantitative FFQ.³⁰ The form was administered by mail twice to a sample of 127 participants at baseline and at 1 year. During this interval, men completed 1-week diet records twice, spaced about 6 months apart. Mean values for intake of most nutrients assessed by the 2 methods were similar. Intraclass correlation coefficients for the baseline and 1-year FFQ-assessed nutrient intakes that ranged from 0.47 for vitamin E (without supplements) to 0.80 for vitamin C (with supplements). Correlation coefficients between the energy-adjusted nutrient intakes were measured by diet records and the 1-year FFQ, which asked about diet during the year encompassing the diet records. Higher raw and percent scores indicated better alignment with recommendations from each respective dietary pattern. Percent scores were calculated as a standardizing method and used in analyses for ease of comparing the dietary patterns. Scoring can be found in Appendix 2.

Demographic characteristics and clinical indicators. Demographic characteristics included age categories, sex, and race/ethnicity. Clinical indicators included BMI, the number of comorbid conditions used to calculate the Charlson Comorbidity Index, family history of CRC in first-degree relatives, number of follow-up colonoscopies across the study period, and food-based vitamin D intake.³¹ These variables were chosen for their applicability found in previous CSP #380 cohort studies.^18,32,33 Self-reported race and ethnicity were collapsed due to small numbers in some groups. The authors acknowledge these are distinct concepts and the variable has limited utility other than for controlling for systemic racism in the model.

Statistical Analyses

Descriptive statistics were used to describe distributional assumptions for all variables, including demographics, clinical indicators, colonoscopy results, and dietary patterns. Pairwise correlations between the total dietary pattern scores and food category scores were calculated with Pearson correlation (r).

Multinomial logistic regression models were created using SAS procedure LOGISTIC with the outcome of the categorical MSCF (no neoplasia, nonadvanced adenoma, or AN).³⁴ A model was created for each independent predictor variable of interest (ie, the HEI, MD, or DASH percentage-standardized dietary pattern score and each food category comprising each dietary pattern score). All models were adjusted for age, sex, race/ethnicity, BMI, number of comorbidities, family history of CRC, number of follow-up colonoscopies, and estimated daily food-derived vitamin D intake. The demographic and clinical indicators were included in the models as they are known to be associated with CRC risk.¹⁸ The number of colonoscopies was included to control for surveillance intensity presuming risk for AN is reduced as polyps are removed. Because colonoscopy findings from an initial screening have unique clinical implications compared with follow- up and surveillance, MSCF was observed in 2 ways in sensitivity analyses: (1) baseline and (2) aggregate follow-up and surveillance only, excluding baseline findings.

Adjusted odds ratios (aORs) and 95% CIs for each of the MSCF outcomes with a reference finding of no neoplasia for the models are presented. We chose not to adjust for multiple comparisons across the different dietary patterns given the correlation between dietary pattern total and category scores but did adjust for multiple comparisons for dietary categories within each dietary pattern. Tests for statistical significance used α= .05 for the dietary pattern total scores and P values for the dietary category scores for each dietary pattern controlled for false discovery rate using the MULTTEST SAS procedure.35 All data manipulations and analyses were performed using SAS version 9.4.

Results

The study included 3023 patients. All were aged 50 to 75 years, 2923 (96.7%) were male and 2532 (83.8%) were non-Hispanic White (Table 1). Most participants were overweight or obese (n = 2535 [83.8%]), 2024 (67.0%) had ≤ 2 comorbidities, and 2602 (86.1%) had no family history of CRC. The MSCF for 1628 patients (53.9%) was no neoplasia, 966 patients (32.0%) was nonadvanced adenoma, and 429 participants (14.2%) had AN.

Mean percent scores were 58.5% for HEI, 38.2% for MD, and 63.1% for the DASH diet, with higher percentages indicating greater alignment with the recommendations for each diet (Table 2). All 3 dietary patterns scores standardized to percentages were strongly and significantly correlated in pairwise comparisons: HEI:MD, r = 0.62 (P < .001); HEI:DASH, r = 0.60 (P < .001); and MD:DASH, r = 0.72 (P < .001). Likewise, food category scores were significantly correlated across dietary patterns. For example, whole grain and fiber values from each dietary score were strongly correlated in pairwise comparisons: HEI Whole Grain:MD Grain, r = 0.64 (P < .001); HEI Whole Grain:DASH Fiber, r = 0.71 (P < .001); and MD Grain:DASH Fiber, r = 0.70 (P < .001).

Associations between individual participants' dietary pattern scores and the outcome of their pooled MSCF from baseline screening and ≥ 10 years of surveillance are presented in Table 3. For each single-point increases in dietary pattern scores (reflecting better dietary quality), aORs for nonadvanced adenoma vs no neoplasia were slightly lower but not statistically significantly: HEI, aOR, 1.00 (95% CI, 0.99-1.01); MD, aOR, 0.98 (95% CI, 0.94-1.02); and DASH, aOR, 0.99 (95% CI, 0.99-1.00). aORs for AN vs no neoplasia were slightly lower for each dietary pattern assessed, and only the MD and DASH scores were significantly different from 1.00: HEI, aOR, 1.00 (95% CI, 0.99-1.01); MD, aOR, 0.95 (95% CI, 0.90-1.00); and DASH, aOR, 0.99 (95% CI, 0.98-1.00).

We observed lower odds for nonadvanced adenoma and AN among all these dietary patterns when there was greater alignment with the recommended intake of whole grains and fiber. In separate models conducted using food categories comprising the dietary patterns as independent variables and after correcting for multiple tests, higher scores for the HEI Refined Grain category were associated with higher odds for nonadvanced adenoma (aOR, 1.03 [95% CI, 1.01-1.05]; P = .01) and AN (aOR, 1.05 [95% CI, 1.02-1.08]; P < .001). Higher scores for the HEI Whole Grain category were associated with lower odds for nonadvanced adenoma (aOR, 0.97 [95% CI, 0.95-0.99]; P = .01) and AN (aOR, 0.96 [95% CI, 0.93-0.99]; P = .01). Higher scores for the MD Grain category were significantly associated with lower odds for nonadvanced adenoma (aOR, 0.44 [95% CI, 0.26-0.75]; P = .002) and AN (aOR, 0.29 [95% CI, 0.14-0.62]; P = .001). The DASH Grains category also was significantly associated with lower odds for AN (aOR, 0.86 [95% CI, 0.78-0.95]; P = .002).

Discussion

In this study of 3023 veterans undergoing first-time screening colonoscopy and ≥ 10 years of surveillance, we found that healthy dietary patterns, as assessed by the MD and DASH diet, were significantly associated with lower risk of AN. Additionally, we identified lower odds for AN and nonadvanced adenoma compared with no neoplasia for higher grain scores for all the dietary patterns studied. Other food categories that comprise the dietary pattern scores had mixed associations with the MSCF outcomes. Several other studies have examined associations between dietary patterns and risk for CRC but to our knowledge, no studies have explored these associations among US veterans.

These results also indicate study participants had better than average (based on a 50% threshold) dietary quality according to the HEI and DASH diet scoring methods we used, but poor dietary quality according to the MD scoring method. The mean HEI scores for the present study were higher than a US Department of Agriculture study by Dong et al that compared dietary quality between veterans and nonveterans using the HEI, for which veterans’ expected HEI score was 45.6 of 100.⁸ This could be explained by the fact that the participants needed to be healthy to be eligible and those with healthier behaviors overall may have self-selected into the study due to motivation for screening during a time when screening was not yet commonplace. ³⁶ Similarly, participants of the present study had higher adherence to the DASH diet (63.1%) than adolescents with diabetes in a study by Günther et al. Conversely, firefighters who were coached to use a Mediterranean-style dietary pattern and dietary had higher adherence to MD than did participants in this study.²⁷

A closer examination of specific food category component scores that comprise the 3 distinct dietary patterns revealed mixed results from the multinomial modeling, which may have to do with the guideline thresholds used to calculate the dietary scores. When analyzed separately in the logistic regression models for their associations with nonadvanced adenomas and AN compared with no neoplasia, higher MD and DASH fruit scores (but not HEI fruit scores) were found to be significant. Other studies have had mixed findings when attempting to test for associations of fruit intake with adenoma recurrence.^10,37

This study had some unexpected findings. Vegetable intake was not associated with nonadvanced adenomas or AN risk. Studies of food categories have consistently found vegetable (specifically cruciferous ones) intake to be linked with lower odds for cancers.³⁸ Likewise, the red meat category, which was only a unique food category in the MD score, was not associated with nonadvanced adenomas or AN. Despite consistent literature suggesting higher intake of red meat and processed meats increases CRC risk, in 2019 the Nutritional Recommendations Consortium indicated that the evidence was weak.^39,40 This study showed higher DASH diet scores for low-fat dairy, which were maximized when participants reported at least 50% of their dairy servings per day as being low-fat, had lower odds for AN. Yet, the MD scores for low-fat dairy had no association with either outcome; their calculation was based on total number of servings per week. This difference in findings suggests the fat intake ratio may be more relevant to CRC risk than intake quantity.

The literature is mixed regarding fatty acid intake and CRC risk, which may be relevant to both dairy and meat intake. One systematic review and meta-analysis found dietary fat and types of fatty acid intake had no association with CRC risk.⁴¹ However, a more recent meta-analysis that assessed both dietary intake and plasma levels of fatty acids did find some statistically significant differences for various types of fatty acids and CRC risk.⁴²

The findings in the present study that grain intake is associated with lower odds for more severe colonoscopy findings among veterans are notable.⁴³ Lieberman et al, using the CSP #380 data, found that cereal fiber intake was associated with a lower odds for AN compared with hyperplastic polyps (OR, 0.98 [95% CI, 0.96- 1.00]).¹⁸ Similarly, Hullings et al determined that older adults in the highest quintile of cereal fiber intake had significantly lower odds of CRC than those in lower odds for CRC when compared with lowest quintile (OR, 0.89 [95% CI, 0.83- 0.96]; P < .001).⁴⁴ These findings support existing guidance that prioritizes whole grains as a key source of dietary fiber for CRC prevention.

A recent literature review on fiber, fat, and CRC risk suggested a consensus regarding one protective mechanism: dietary fiber from grains modulates the gut microbiota by promoting butyrate synthesis.⁴⁵ Butyrate is a short-chain fatty acid that supports energy production in colonocytes and has tumor-suppressing properties.⁴⁶ Our findings suggest there could be more to learn about the relationship between butyrate production and reduction of CRC risk through metabolomic studies that use measurements of plasma butyrate. These studies may examine associations between not just a singular food or food category, but rather food patterns that include fruits, vegetables, nuts and seeds, and whole grains known to promote butyrate production and plasma butyrate.⁴⁷

Improved understanding of mechanisms and risk-modifying lifestyle factors such as dietary patterns may enhance prevention strategies. Identifying the collective chemopreventive characteristics of a specific dietary pattern (eg, MD) will be helpful to clinicians and health care staff to promote healthy eating to reduce cancer risk. More studies are needed to understand whether such promotion is more clinically applicable and effective for patients, as compared with eating more or less of specific foods (eg, more whole grains, less red meat). Furthermore, considering important environmental factors collectively beyond dietary patterns may offer a way to better tailor screening and implement a variety of lifestyle interventions. In the literature, this is often referred to as a teachable moment when patients’ attentions are captured and may position them to be more receptive to guidance.⁴⁸

Limitations

This study has several important limitations and leaves opportunities for future studies that explore the role of dietary patterns and AN or CRC risk. First, the FFQ data used to calculate dietary pattern scores used in analysis were only captured at baseline, and there are nearly 3 decades across the study period. However, it is widely assumed that the diets of older adults, like those included in this study, remain stable over time which is appropriate given our sample population was aged 50 to 75 years when the baseline FFQ data were collected.^49-51 Additionally, while the HEI is a well-documented, standard scoring method for dietary quality, there are multitudes of dietary pattern scoring approaches for MD and DASH.^23,52,53 Finally, findings from this study using the sample of veterans may not be generalizable to a broader population. Future longitudinal studies that test for a clinically significant change threshold are warranted.

Conclusion

Results of this study suggest future research should further explore the effects of dietary patterns, particularly intake of specific food groups in combination, as opposed to individual nutrients or food items, on AN and CRC risk. Possible studies might explore these dietary patterns for their mechanistic role in altering the microbiome metabolism, which may influence CRC outcomes or include diet in a more comprehensive, holistic risk score that could be used to predict colonic neoplasia risk or in intervention studies that assess the effects of dietary changes on long-term CRC prevention. We suggest there are differences in people’s dietary intake patterns that might be important to consider when implementing tailored approaches to CRC risk mitigation.

Screening for colorectal cancer (CRC) with colonoscopy enables the identification and removal of CRC precursors (colonic adenomas) and has been associated with reduced risk of CRC incidence and mortality.^1-3 Furthermore, there is consensus that diet and lifestyle may be associated with forestalling CRC pathogenesis at the intermediate adenoma stages.^4-7 However, studies have shown that US veterans have poorer diet quality and a higher risk for neoplasia compared with nonveterans, reinforcing the need for tailored clinical approaches.^8,9 Combining screening with conversations about modifiable environmental and lifestyle risk factors, such as poor diet, is a highly relevant and possibly easily leveraged prevention for those at high risk. However, there is limited evidence for any particular dietary patterns or dietary features that are most important over time.⁷

Several dietary components have been shown to be associated with CRC risk,10 either as potentially chemopreventive (fiber, fruits and vegetables,¹¹ dairy,¹² supplemental vitamin D,¹³ calcium,¹⁴ and multivitamins¹⁵) or carcinogenic (red meat¹⁶ and alcohol¹⁷). Previous studies of veterans have similarly shown that higher intake of fiber and vitamin D reduced risk, and red meat is associated with an increased risk for finding CRC precursors during colonoscopy.¹⁸ However, these dietary categories are often analyzed in isolation. Studying healthy dietary patterns in aggregate may be more clinically relevant and easier to implement for prevention of CRC and its precursors.^19-21 Healthy dietary patterns, such as the US Dietary Guidelines for Americans represented by the Healthy Eating Index (HEI), the Mediterranean diet (MD), and the Dietary Approaches to Stop Hypertension (DASH) diet, have been associated with lower risk for chronic disease.^22-24 Despite the extant literature, no known studies have compared these dietary patterns for associations with risk of CRC precursor or CRC development among US veterans undergoing long-term screening and follow-up after a baseline colonoscopy.

The objective of this study was to test for associations between baseline scores of healthy dietary patterns and the most severe colonoscopy findings (MSCFs) over ≥ 10 years following a baseline screening colonoscopy in veterans.

Methods

Participants in the Cooperative Studies Program (CSP) #380 cohort study included 3121 asymptomatic veterans aged 50 to 75 years at baseline who had consented to initial screening colonoscopy between 1994 and 1997, with subsequent follow-up and surveillance.²⁵ Prior to their colonoscopy, all participants completed a baseline study survey that included questions about cancer risk factors including family history of CRC, diet, physical activity, and medication use.

Included in this cross-sectional analysis were data from a sample of veteran participants of the CSP #380 cohort with 1 baseline colonoscopy, follow-up surveillance through 2009, a cancer risk factor survey collected at baseline, and complete demographic and clinical indicator data. Excluded from the analysis were 67 participants with insufficient responses to the dietary food frequency questionnaire (FFQ) and 31 participants with missing body mass index (BMI), 3023 veterans.

Measures

MSCF. The outcome of interest in this study was the MSCF recorded across all participant colonoscopies during the study period. MSCF was categorized as either (1) no neoplasia; (2) < 2 nonadvanced adenomas, including small adenomas (diameter < 10 mm) with tubular histology; or (3) advanced neoplasia (AN), which is characterized by adenomas > 10 mm in diameter, with villous histology, with high-grade dysplasia, or CRC.

Dietary patterns. Dietary pattern scores representing dietary quality and calculated based on recommendations of the US Dietary Guidelines for Americans using the HEI, MD, and DASH diets were independent variables.^26-28 These 3 dietary patterns were chosen for their hypothesized relationship with CRC risk, but each weighs food categories differently (Appendix 1).^22-24,29 Dietary pattern scores were calculated using the CSP #380 self-reported responses to 129 baseline survey questions adapted from a well-established and previously validated semiquantitative FFQ.³⁰ The form was administered by mail twice to a sample of 127 participants at baseline and at 1 year. During this interval, men completed 1-week diet records twice, spaced about 6 months apart. Mean values for intake of most nutrients assessed by the 2 methods were similar. Intraclass correlation coefficients for the baseline and 1-year FFQ-assessed nutrient intakes that ranged from 0.47 for vitamin E (without supplements) to 0.80 for vitamin C (with supplements). Correlation coefficients between the energy-adjusted nutrient intakes were measured by diet records and the 1-year FFQ, which asked about diet during the year encompassing the diet records. Higher raw and percent scores indicated better alignment with recommendations from each respective dietary pattern. Percent scores were calculated as a standardizing method and used in analyses for ease of comparing the dietary patterns. Scoring can be found in Appendix 2.

Demographic characteristics and clinical indicators. Demographic characteristics included age categories, sex, and race/ethnicity. Clinical indicators included BMI, the number of comorbid conditions used to calculate the Charlson Comorbidity Index, family history of CRC in first-degree relatives, number of follow-up colonoscopies across the study period, and food-based vitamin D intake.³¹ These variables were chosen for their applicability found in previous CSP #380 cohort studies.^18,32,33 Self-reported race and ethnicity were collapsed due to small numbers in some groups. The authors acknowledge these are distinct concepts and the variable has limited utility other than for controlling for systemic racism in the model.

Statistical Analyses

Descriptive statistics were used to describe distributional assumptions for all variables, including demographics, clinical indicators, colonoscopy results, and dietary patterns. Pairwise correlations between the total dietary pattern scores and food category scores were calculated with Pearson correlation (r).

Multinomial logistic regression models were created using SAS procedure LOGISTIC with the outcome of the categorical MSCF (no neoplasia, nonadvanced adenoma, or AN).³⁴ A model was created for each independent predictor variable of interest (ie, the HEI, MD, or DASH percentage-standardized dietary pattern score and each food category comprising each dietary pattern score). All models were adjusted for age, sex, race/ethnicity, BMI, number of comorbidities, family history of CRC, number of follow-up colonoscopies, and estimated daily food-derived vitamin D intake. The demographic and clinical indicators were included in the models as they are known to be associated with CRC risk.¹⁸ The number of colonoscopies was included to control for surveillance intensity presuming risk for AN is reduced as polyps are removed. Because colonoscopy findings from an initial screening have unique clinical implications compared with follow- up and surveillance, MSCF was observed in 2 ways in sensitivity analyses: (1) baseline and (2) aggregate follow-up and surveillance only, excluding baseline findings.

Adjusted odds ratios (aORs) and 95% CIs for each of the MSCF outcomes with a reference finding of no neoplasia for the models are presented. We chose not to adjust for multiple comparisons across the different dietary patterns given the correlation between dietary pattern total and category scores but did adjust for multiple comparisons for dietary categories within each dietary pattern. Tests for statistical significance used α= .05 for the dietary pattern total scores and P values for the dietary category scores for each dietary pattern controlled for false discovery rate using the MULTTEST SAS procedure.35 All data manipulations and analyses were performed using SAS version 9.4.

Results

The study included 3023 patients. All were aged 50 to 75 years, 2923 (96.7%) were male and 2532 (83.8%) were non-Hispanic White (Table 1). Most participants were overweight or obese (n = 2535 [83.8%]), 2024 (67.0%) had ≤ 2 comorbidities, and 2602 (86.1%) had no family history of CRC. The MSCF for 1628 patients (53.9%) was no neoplasia, 966 patients (32.0%) was nonadvanced adenoma, and 429 participants (14.2%) had AN.

Mean percent scores were 58.5% for HEI, 38.2% for MD, and 63.1% for the DASH diet, with higher percentages indicating greater alignment with the recommendations for each diet (Table 2). All 3 dietary patterns scores standardized to percentages were strongly and significantly correlated in pairwise comparisons: HEI:MD, r = 0.62 (P < .001); HEI:DASH, r = 0.60 (P < .001); and MD:DASH, r = 0.72 (P < .001). Likewise, food category scores were significantly correlated across dietary patterns. For example, whole grain and fiber values from each dietary score were strongly correlated in pairwise comparisons: HEI Whole Grain:MD Grain, r = 0.64 (P < .001); HEI Whole Grain:DASH Fiber, r = 0.71 (P < .001); and MD Grain:DASH Fiber, r = 0.70 (P < .001).

Associations between individual participants' dietary pattern scores and the outcome of their pooled MSCF from baseline screening and ≥ 10 years of surveillance are presented in Table 3. For each single-point increases in dietary pattern scores (reflecting better dietary quality), aORs for nonadvanced adenoma vs no neoplasia were slightly lower but not statistically significantly: HEI, aOR, 1.00 (95% CI, 0.99-1.01); MD, aOR, 0.98 (95% CI, 0.94-1.02); and DASH, aOR, 0.99 (95% CI, 0.99-1.00). aORs for AN vs no neoplasia were slightly lower for each dietary pattern assessed, and only the MD and DASH scores were significantly different from 1.00: HEI, aOR, 1.00 (95% CI, 0.99-1.01); MD, aOR, 0.95 (95% CI, 0.90-1.00); and DASH, aOR, 0.99 (95% CI, 0.98-1.00).

We observed lower odds for nonadvanced adenoma and AN among all these dietary patterns when there was greater alignment with the recommended intake of whole grains and fiber. In separate models conducted using food categories comprising the dietary patterns as independent variables and after correcting for multiple tests, higher scores for the HEI Refined Grain category were associated with higher odds for nonadvanced adenoma (aOR, 1.03 [95% CI, 1.01-1.05]; P = .01) and AN (aOR, 1.05 [95% CI, 1.02-1.08]; P < .001). Higher scores for the HEI Whole Grain category were associated with lower odds for nonadvanced adenoma (aOR, 0.97 [95% CI, 0.95-0.99]; P = .01) and AN (aOR, 0.96 [95% CI, 0.93-0.99]; P = .01). Higher scores for the MD Grain category were significantly associated with lower odds for nonadvanced adenoma (aOR, 0.44 [95% CI, 0.26-0.75]; P = .002) and AN (aOR, 0.29 [95% CI, 0.14-0.62]; P = .001). The DASH Grains category also was significantly associated with lower odds for AN (aOR, 0.86 [95% CI, 0.78-0.95]; P = .002).

Discussion

In this study of 3023 veterans undergoing first-time screening colonoscopy and ≥ 10 years of surveillance, we found that healthy dietary patterns, as assessed by the MD and DASH diet, were significantly associated with lower risk of AN. Additionally, we identified lower odds for AN and nonadvanced adenoma compared with no neoplasia for higher grain scores for all the dietary patterns studied. Other food categories that comprise the dietary pattern scores had mixed associations with the MSCF outcomes. Several other studies have examined associations between dietary patterns and risk for CRC but to our knowledge, no studies have explored these associations among US veterans.

These results also indicate study participants had better than average (based on a 50% threshold) dietary quality according to the HEI and DASH diet scoring methods we used, but poor dietary quality according to the MD scoring method. The mean HEI scores for the present study were higher than a US Department of Agriculture study by Dong et al that compared dietary quality between veterans and nonveterans using the HEI, for which veterans’ expected HEI score was 45.6 of 100.⁸ This could be explained by the fact that the participants needed to be healthy to be eligible and those with healthier behaviors overall may have self-selected into the study due to motivation for screening during a time when screening was not yet commonplace. ³⁶ Similarly, participants of the present study had higher adherence to the DASH diet (63.1%) than adolescents with diabetes in a study by Günther et al. Conversely, firefighters who were coached to use a Mediterranean-style dietary pattern and dietary had higher adherence to MD than did participants in this study.²⁷

A closer examination of specific food category component scores that comprise the 3 distinct dietary patterns revealed mixed results from the multinomial modeling, which may have to do with the guideline thresholds used to calculate the dietary scores. When analyzed separately in the logistic regression models for their associations with nonadvanced adenomas and AN compared with no neoplasia, higher MD and DASH fruit scores (but not HEI fruit scores) were found to be significant. Other studies have had mixed findings when attempting to test for associations of fruit intake with adenoma recurrence.^10,37

This study had some unexpected findings. Vegetable intake was not associated with nonadvanced adenomas or AN risk. Studies of food categories have consistently found vegetable (specifically cruciferous ones) intake to be linked with lower odds for cancers.³⁸ Likewise, the red meat category, which was only a unique food category in the MD score, was not associated with nonadvanced adenomas or AN. Despite consistent literature suggesting higher intake of red meat and processed meats increases CRC risk, in 2019 the Nutritional Recommendations Consortium indicated that the evidence was weak.^39,40 This study showed higher DASH diet scores for low-fat dairy, which were maximized when participants reported at least 50% of their dairy servings per day as being low-fat, had lower odds for AN. Yet, the MD scores for low-fat dairy had no association with either outcome; their calculation was based on total number of servings per week. This difference in findings suggests the fat intake ratio may be more relevant to CRC risk than intake quantity.

The literature is mixed regarding fatty acid intake and CRC risk, which may be relevant to both dairy and meat intake. One systematic review and meta-analysis found dietary fat and types of fatty acid intake had no association with CRC risk.⁴¹ However, a more recent meta-analysis that assessed both dietary intake and plasma levels of fatty acids did find some statistically significant differences for various types of fatty acids and CRC risk.⁴²

The findings in the present study that grain intake is associated with lower odds for more severe colonoscopy findings among veterans are notable.⁴³ Lieberman et al, using the CSP #380 data, found that cereal fiber intake was associated with a lower odds for AN compared with hyperplastic polyps (OR, 0.98 [95% CI, 0.96- 1.00]).¹⁸ Similarly, Hullings et al determined that older adults in the highest quintile of cereal fiber intake had significantly lower odds of CRC than those in lower odds for CRC when compared with lowest quintile (OR, 0.89 [95% CI, 0.83- 0.96]; P < .001).⁴⁴ These findings support existing guidance that prioritizes whole grains as a key source of dietary fiber for CRC prevention.

A recent literature review on fiber, fat, and CRC risk suggested a consensus regarding one protective mechanism: dietary fiber from grains modulates the gut microbiota by promoting butyrate synthesis.⁴⁵ Butyrate is a short-chain fatty acid that supports energy production in colonocytes and has tumor-suppressing properties.⁴⁶ Our findings suggest there could be more to learn about the relationship between butyrate production and reduction of CRC risk through metabolomic studies that use measurements of plasma butyrate. These studies may examine associations between not just a singular food or food category, but rather food patterns that include fruits, vegetables, nuts and seeds, and whole grains known to promote butyrate production and plasma butyrate.⁴⁷

Improved understanding of mechanisms and risk-modifying lifestyle factors such as dietary patterns may enhance prevention strategies. Identifying the collective chemopreventive characteristics of a specific dietary pattern (eg, MD) will be helpful to clinicians and health care staff to promote healthy eating to reduce cancer risk. More studies are needed to understand whether such promotion is more clinically applicable and effective for patients, as compared with eating more or less of specific foods (eg, more whole grains, less red meat). Furthermore, considering important environmental factors collectively beyond dietary patterns may offer a way to better tailor screening and implement a variety of lifestyle interventions. In the literature, this is often referred to as a teachable moment when patients’ attentions are captured and may position them to be more receptive to guidance.⁴⁸

Limitations

This study has several important limitations and leaves opportunities for future studies that explore the role of dietary patterns and AN or CRC risk. First, the FFQ data used to calculate dietary pattern scores used in analysis were only captured at baseline, and there are nearly 3 decades across the study period. However, it is widely assumed that the diets of older adults, like those included in this study, remain stable over time which is appropriate given our sample population was aged 50 to 75 years when the baseline FFQ data were collected.^49-51 Additionally, while the HEI is a well-documented, standard scoring method for dietary quality, there are multitudes of dietary pattern scoring approaches for MD and DASH.^23,52,53 Finally, findings from this study using the sample of veterans may not be generalizable to a broader population. Future longitudinal studies that test for a clinically significant change threshold are warranted.

Conclusion

Results of this study suggest future research should further explore the effects of dietary patterns, particularly intake of specific food groups in combination, as opposed to individual nutrients or food items, on AN and CRC risk. Possible studies might explore these dietary patterns for their mechanistic role in altering the microbiome metabolism, which may influence CRC outcomes or include diet in a more comprehensive, holistic risk score that could be used to predict colonic neoplasia risk or in intervention studies that assess the effects of dietary changes on long-term CRC prevention. We suggest there are differences in people’s dietary intake patterns that might be important to consider when implementing tailored approaches to CRC risk mitigation.

About 530,000 to 628,000 episodes of bloodstream infections (BSI) occur annually in the US.¹ Early identification and treatment of bacteremia are essential to improve patient outcomes because it allows for more timely targeted antibiotic therapy.² Organism identification and susceptibility testing can take 2 to 5 days, prolonging the use of broad-spectrum empiric antibiotics and increasing the risk of adverse events.^3,4 The Infectious Disease Society of America recommends the use of rapid diagnostic testing and antimicrobial stewardship programs (ASPs) to improve rates of antibiotic susceptibilities to targeted antibiotics and optimize resource utilization.³ Rapid blood culture identification (BCID) technologies reduce the duration of empiric antibiotics in patients with contaminated blood cultures, resulting in shorter hospital stays and saving money per each patient tested.⁴

In March 2023, Veteran Health Indiana (VHI) implemented the BioFire FilmArray Blood Culture Identification (BCID2), a BSI panel test that identifies select gram-negative bacteria, gram-positive bacteria, yeast, and antimicrobial resistance genes with an aggregate sensitivity of 99% and a specificity of 99.8%. The BCID2 presents clinically relevant information faster than traditional culture methods, allowing clinicians to make more efficient and educated antibiotic regimen decisions than with previous methods.⁵

It takes 24 to 48 hours from blood collection for culture incubation, positivity, and gram staining to occur at VHI. If the gram stain is positive, the blood culture is placed on the BioFire BCID2 in addition to traditional culture medium. BioFire BCID2 results are ready in 45 to 60 minutes. Results are uploaded into the electronic health record (EHR) ≤ 2 hours after they are obtained and the primary team is notified if the test is positive for certain critical results. Susceptibility testing of an identified organism typically requires an additional 24 to 48 hours for finalization. VHI Infectious Disease created an evidence-based antibiotic recommendation chart for certain medication(s) and alternate therapies based on the reported organism and its interpreted presence of resistance markers (eg, ceftriaxone for Escherichia coli when extended-spectrum beta lactamases are not detected vs meropenem if extended-spectrum beta lactamases marker are present). These charts optimize the antibiotic regimen while awaiting susceptibility finalizations.

Two previous studies describe the impact of rapid diagnostic testing technology at US Department of Veterans Affairs (VA) medical centers.^6,7 In Texas, the ASP reviewed BCID panel results via clinical decision support software for about 1 hour per day.⁶ A Los Angeles study analyzed the impact of Biofire BCID with an interpretation guide centered on unnecessary vancomycin use and determined that shorter duration of the medication may have been the result of more frequent infectious disease consultation.⁷

This study assessed the time to optimal antibiotic de-escalation before and after the implementation of BioFire BCID2 with results reviewed by the ASP without active notification or assistance of any clinical decision support technology. The primary objective was to evaluate difference in time to optimal antibiotics from blood culture draw pre- vs postintervention. Secondary objectives included differences in time to organism identification, difference in time on broad-spectrum antibiotics, and difference in time to appropriate antibiotics.

Methods

This quasi-experimental retrospective chart review assessed the impact of BioFire BCID2 use on timely antibiotic de-escalation for patients who experienced a BSI at VHI between March 1, 2022, and October 1, 2023. Microbiology laboratory records identified eligible patients with positive blood cultures within the study time frame. Data were collected from the VHI EHR.

Patients were included if they had a positive bacterial blood culture and received ≥ 1 antibiotic indicated for bacteremia while receiving inpatient care. Patients were excluded if they died prior to blood culture results, transferred out of VHI, left against medical advice, or had untreated contaminants in blood culture results (ie, never received antibiotics aimed at the contaminated culture).

Patient lists were generated for before and after implementation of BioFire BCID2 (pre- and postintervention) using the VHI EHR and microbiology laboratory record system. The pre- and postinterventions groups were different sizes. As a result, a random sampling of the preintervention group was selected and included patients from March 1, 2022, through March 26, 2023. The postintervention group was smaller due to time constraints between initiation of BioFire BCID2 for data collection and included all patients from March 27, 2023, through October 1, 2023.

Optimal antibiotics were defined as escalation from inappropriate therapy to broader agent(s), de-escalation from broad-spectrum therapy to targeted agent(s), discontinuation of therapy due to an organism being identified as a contaminant, or optimization of a regimen to the preferred antimicrobial agent based on evidence-based consensus guidelines. Broad-spectrum antibiotics included: piperacillin/tazobactam, cefepime, ceftazidime, ceftazidime-avibactam, cefiderocol, carbapenems, fluroquinolones, vancomycin, daptomycin, ceftaroline, linezolid, or aztreonam. Appropriate antibiotics were defined as those with activity toward the final identified organism(s).

Deidentified participant data were entered into Microsoft Excel and kept on a secure VA server to complete statistical analyses. Parametric continuous data, such as age, were analyzed using the t-test, while nonparametric continuous data, such as time to optimal antibiotics, were analyzed using the Mann-Whitney U test. Categorical data, like sex and race, were analyzed using either Fisher exact test for small sample sizes or X² test for a larger sample size. Statistical significance levels was defined as P < .05.

Results

Using patient lists drawn from the EHR and the microbiology laboratory records, 110 electronic charts were randomly selected for review. Fifteen patients were excluded: 8 had untreated contaminants, 4 died, and 3 were transferred out of VHI. Of the 95 patients included, 48 were in the preintervention group and 47 were in the postintervention group (Figure 1).

Baseline characteristics were similar between the 2 groups (Table 1). Most patients were White males aged > 70 years in the EHR. The urinary tract was the most common source of infection, impacting 12 patients in each group (Figure 2). Escherichia coli, Klebsiella, Staphylococcus, and Streptococcus were the most common bloodstream isolates identified.

The median time to optimal antibiotics in the preintervention group was 58.5 hours vs 43.4 hours in the postintervention group (P = .11). The median time to organism identification was 37.8 hours in the preintervention group vs 16.9 hours in the postintervention group (P < .001). The median time on broad-spectrum antibiotics was 45.2 hours in the preintervention group vs 46.6 hours in the postintervention group (P = .99). The median time on appropriate antibiotics in the preintervention group was 2.3 hours vs 1.9 hours in the postintervention group (P = .79). Differences in other measured outcomes between the groups were not statistically significant (Table 2).

Although implementation of rapid diagnostic technology reduced the median time to optimal antibiotics, the results were not statistically significant. Shorter time to organism identification in the postintervention group compared to the preintervention group was the lone statistically significant metric (P < .001).

Discussion

A lack of statistical significance in the primary outcome may have been due to nonadherence to facility de-escalation protocols or a suboptimal BioFire BCID2 result notification system. Additionally, use of rapid BCID at VHI may improve over time as clinicians become more familiar with the technology. Gaps in clinical pharmacy coverage during the night shift may have also contributed to delays in antibiotic optimization, particularly if other clinicians are not equipped with the knowledge or training to appropriately deescalate antibiotics based on microorganisms identified. A 2017 study by Donner et al concluded that physician interpretation of BCID results is suboptimal and should be augmented with clinical decision support tools as new technology becomes available.⁸ Despite the statistically insignificant results of this study, it did highlight potential areas of improvement which can lead to improved patient care.

Previous research has evaluated the impact of rapid BCID technology on antibiotic treatment and clinical outcomes. Chiasson et al found that median time to optimal therapy was 73.8 hours in the pre-BCID arm compared to 34.7 hours in the post- BCID arm (P ≤ .001), emphasizing the importance of combining rapid BCID with clinical decision support tools and pharmacy input.⁶ Senok et al found that BCID2 implementation led to a significant decrease in median time to culture result, which informed optimal antibiotic therapy and decreased 30-day mortality in the intensive care setting.⁹ In contrast, the current study did not stratify patients according to medical ward or illness severity even though clinicians may be less likely to de-escalate antibiotic therapy in critically ill patients.

Bae et al reported findings consistent with the current study and concluded that BCID did not affect the clinical outcomes of overall BSIs; however, it contributed to early administration of effective antibiotics in cases of BSIs caused by multidrug-resistant organisms.¹⁰ Results of this study were not stratified according to multidrug-resistant organisms because the sample size was too small. The current study also included patients with polymicrobial infections, which may have impacted the results due to a less streamlined approach to antibiotic optimization.

Limitations

This single-center, retrospective study had a small sample size, short time frame, and lacked patient diversity, and therefore may not be generalizable to other health care systems. The sample size was limited by shorter date range and smaller patient list between BioFire BCID2 implementation and data collection, which was used to determine the number of charts selected in each group. Some patients received antibiotics prior to blood cultures being drawn, which may falsely decrease time to optimal/ appropriate antibiotics and falsely increase time on broad spectrum/any antibiotics due to early antibiotic administration. The total number of patients on broad-spectrum antibiotics differed from the total number of patients for other outcomes because several patients never received the defined broad spectrum antibiotics.

Conclusions

When combined with a pre-existing ASP without active notification, the implementation of BioFire BCID2 did not return statistically significant data showing a decrease in time to optimal antibiotics, time to appropriate antibiotics, or time on broad-spectrum antibiotics at VHI. To make this program more successful, pharmacist intervention and clinical decision support tools may be needed.

Additional research is required to determine the optimal integration of antimicrobial stewardship, rapid diagnostic technology, and pharmacy services for maximum benefit. Even though the primary outcome was not statistically significant, the results may be clinically significant from a stewardship perspective. Realigning microbiology workflows to mimic other research, which emphasizes the importance of funneling rapid BCID results through the ASP, may improve outcomes. Future studies may be warranted following the implementation of clinical decision support tools to assess their impact on stewardship practices and patient outcomes.

About 530,000 to 628,000 episodes of bloodstream infections (BSI) occur annually in the US.¹ Early identification and treatment of bacteremia are essential to improve patient outcomes because it allows for more timely targeted antibiotic therapy.² Organism identification and susceptibility testing can take 2 to 5 days, prolonging the use of broad-spectrum empiric antibiotics and increasing the risk of adverse events.^3,4 The Infectious Disease Society of America recommends the use of rapid diagnostic testing and antimicrobial stewardship programs (ASPs) to improve rates of antibiotic susceptibilities to targeted antibiotics and optimize resource utilization.³ Rapid blood culture identification (BCID) technologies reduce the duration of empiric antibiotics in patients with contaminated blood cultures, resulting in shorter hospital stays and saving money per each patient tested.⁴

In March 2023, Veteran Health Indiana (VHI) implemented the BioFire FilmArray Blood Culture Identification (BCID2), a BSI panel test that identifies select gram-negative bacteria, gram-positive bacteria, yeast, and antimicrobial resistance genes with an aggregate sensitivity of 99% and a specificity of 99.8%. The BCID2 presents clinically relevant information faster than traditional culture methods, allowing clinicians to make more efficient and educated antibiotic regimen decisions than with previous methods.⁵

It takes 24 to 48 hours from blood collection for culture incubation, positivity, and gram staining to occur at VHI. If the gram stain is positive, the blood culture is placed on the BioFire BCID2 in addition to traditional culture medium. BioFire BCID2 results are ready in 45 to 60 minutes. Results are uploaded into the electronic health record (EHR) ≤ 2 hours after they are obtained and the primary team is notified if the test is positive for certain critical results. Susceptibility testing of an identified organism typically requires an additional 24 to 48 hours for finalization. VHI Infectious Disease created an evidence-based antibiotic recommendation chart for certain medication(s) and alternate therapies based on the reported organism and its interpreted presence of resistance markers (eg, ceftriaxone for Escherichia coli when extended-spectrum beta lactamases are not detected vs meropenem if extended-spectrum beta lactamases marker are present). These charts optimize the antibiotic regimen while awaiting susceptibility finalizations.

Two previous studies describe the impact of rapid diagnostic testing technology at US Department of Veterans Affairs (VA) medical centers.^6,7 In Texas, the ASP reviewed BCID panel results via clinical decision support software for about 1 hour per day.⁶ A Los Angeles study analyzed the impact of Biofire BCID with an interpretation guide centered on unnecessary vancomycin use and determined that shorter duration of the medication may have been the result of more frequent infectious disease consultation.⁷

This study assessed the time to optimal antibiotic de-escalation before and after the implementation of BioFire BCID2 with results reviewed by the ASP without active notification or assistance of any clinical decision support technology. The primary objective was to evaluate difference in time to optimal antibiotics from blood culture draw pre- vs postintervention. Secondary objectives included differences in time to organism identification, difference in time on broad-spectrum antibiotics, and difference in time to appropriate antibiotics.

Methods

This quasi-experimental retrospective chart review assessed the impact of BioFire BCID2 use on timely antibiotic de-escalation for patients who experienced a BSI at VHI between March 1, 2022, and October 1, 2023. Microbiology laboratory records identified eligible patients with positive blood cultures within the study time frame. Data were collected from the VHI EHR.

Patients were included if they had a positive bacterial blood culture and received ≥ 1 antibiotic indicated for bacteremia while receiving inpatient care. Patients were excluded if they died prior to blood culture results, transferred out of VHI, left against medical advice, or had untreated contaminants in blood culture results (ie, never received antibiotics aimed at the contaminated culture).

Patient lists were generated for before and after implementation of BioFire BCID2 (pre- and postintervention) using the VHI EHR and microbiology laboratory record system. The pre- and postinterventions groups were different sizes. As a result, a random sampling of the preintervention group was selected and included patients from March 1, 2022, through March 26, 2023. The postintervention group was smaller due to time constraints between initiation of BioFire BCID2 for data collection and included all patients from March 27, 2023, through October 1, 2023.

Optimal antibiotics were defined as escalation from inappropriate therapy to broader agent(s), de-escalation from broad-spectrum therapy to targeted agent(s), discontinuation of therapy due to an organism being identified as a contaminant, or optimization of a regimen to the preferred antimicrobial agent based on evidence-based consensus guidelines. Broad-spectrum antibiotics included: piperacillin/tazobactam, cefepime, ceftazidime, ceftazidime-avibactam, cefiderocol, carbapenems, fluroquinolones, vancomycin, daptomycin, ceftaroline, linezolid, or aztreonam. Appropriate antibiotics were defined as those with activity toward the final identified organism(s).

Deidentified participant data were entered into Microsoft Excel and kept on a secure VA server to complete statistical analyses. Parametric continuous data, such as age, were analyzed using the t-test, while nonparametric continuous data, such as time to optimal antibiotics, were analyzed using the Mann-Whitney U test. Categorical data, like sex and race, were analyzed using either Fisher exact test for small sample sizes or X² test for a larger sample size. Statistical significance levels was defined as P < .05.

Results

Using patient lists drawn from the EHR and the microbiology laboratory records, 110 electronic charts were randomly selected for review. Fifteen patients were excluded: 8 had untreated contaminants, 4 died, and 3 were transferred out of VHI. Of the 95 patients included, 48 were in the preintervention group and 47 were in the postintervention group (Figure 1).

Baseline characteristics were similar between the 2 groups (Table 1). Most patients were White males aged > 70 years in the EHR. The urinary tract was the most common source of infection, impacting 12 patients in each group (Figure 2). Escherichia coli, Klebsiella, Staphylococcus, and Streptococcus were the most common bloodstream isolates identified.

The median time to optimal antibiotics in the preintervention group was 58.5 hours vs 43.4 hours in the postintervention group (P = .11). The median time to organism identification was 37.8 hours in the preintervention group vs 16.9 hours in the postintervention group (P < .001). The median time on broad-spectrum antibiotics was 45.2 hours in the preintervention group vs 46.6 hours in the postintervention group (P = .99). The median time on appropriate antibiotics in the preintervention group was 2.3 hours vs 1.9 hours in the postintervention group (P = .79). Differences in other measured outcomes between the groups were not statistically significant (Table 2).

Although implementation of rapid diagnostic technology reduced the median time to optimal antibiotics, the results were not statistically significant. Shorter time to organism identification in the postintervention group compared to the preintervention group was the lone statistically significant metric (P < .001).

Discussion

A lack of statistical significance in the primary outcome may have been due to nonadherence to facility de-escalation protocols or a suboptimal BioFire BCID2 result notification system. Additionally, use of rapid BCID at VHI may improve over time as clinicians become more familiar with the technology. Gaps in clinical pharmacy coverage during the night shift may have also contributed to delays in antibiotic optimization, particularly if other clinicians are not equipped with the knowledge or training to appropriately deescalate antibiotics based on microorganisms identified. A 2017 study by Donner et al concluded that physician interpretation of BCID results is suboptimal and should be augmented with clinical decision support tools as new technology becomes available.⁸ Despite the statistically insignificant results of this study, it did highlight potential areas of improvement which can lead to improved patient care.

Previous research has evaluated the impact of rapid BCID technology on antibiotic treatment and clinical outcomes. Chiasson et al found that median time to optimal therapy was 73.8 hours in the pre-BCID arm compared to 34.7 hours in the post- BCID arm (P ≤ .001), emphasizing the importance of combining rapid BCID with clinical decision support tools and pharmacy input.⁶ Senok et al found that BCID2 implementation led to a significant decrease in median time to culture result, which informed optimal antibiotic therapy and decreased 30-day mortality in the intensive care setting.⁹ In contrast, the current study did not stratify patients according to medical ward or illness severity even though clinicians may be less likely to de-escalate antibiotic therapy in critically ill patients.

Bae et al reported findings consistent with the current study and concluded that BCID did not affect the clinical outcomes of overall BSIs; however, it contributed to early administration of effective antibiotics in cases of BSIs caused by multidrug-resistant organisms.¹⁰ Results of this study were not stratified according to multidrug-resistant organisms because the sample size was too small. The current study also included patients with polymicrobial infections, which may have impacted the results due to a less streamlined approach to antibiotic optimization.

Limitations

This single-center, retrospective study had a small sample size, short time frame, and lacked patient diversity, and therefore may not be generalizable to other health care systems. The sample size was limited by shorter date range and smaller patient list between BioFire BCID2 implementation and data collection, which was used to determine the number of charts selected in each group. Some patients received antibiotics prior to blood cultures being drawn, which may falsely decrease time to optimal/ appropriate antibiotics and falsely increase time on broad spectrum/any antibiotics due to early antibiotic administration. The total number of patients on broad-spectrum antibiotics differed from the total number of patients for other outcomes because several patients never received the defined broad spectrum antibiotics.

Conclusions

When combined with a pre-existing ASP without active notification, the implementation of BioFire BCID2 did not return statistically significant data showing a decrease in time to optimal antibiotics, time to appropriate antibiotics, or time on broad-spectrum antibiotics at VHI. To make this program more successful, pharmacist intervention and clinical decision support tools may be needed.

Additional research is required to determine the optimal integration of antimicrobial stewardship, rapid diagnostic technology, and pharmacy services for maximum benefit. Even though the primary outcome was not statistically significant, the results may be clinically significant from a stewardship perspective. Realigning microbiology workflows to mimic other research, which emphasizes the importance of funneling rapid BCID results through the ASP, may improve outcomes. Future studies may be warranted following the implementation of clinical decision support tools to assess their impact on stewardship practices and patient outcomes.

About 530,000 to 628,000 episodes of bloodstream infections (BSI) occur annually in the US.¹ Early identification and treatment of bacteremia are essential to improve patient outcomes because it allows for more timely targeted antibiotic therapy.² Organism identification and susceptibility testing can take 2 to 5 days, prolonging the use of broad-spectrum empiric antibiotics and increasing the risk of adverse events.^3,4 The Infectious Disease Society of America recommends the use of rapid diagnostic testing and antimicrobial stewardship programs (ASPs) to improve rates of antibiotic susceptibilities to targeted antibiotics and optimize resource utilization.³ Rapid blood culture identification (BCID) technologies reduce the duration of empiric antibiotics in patients with contaminated blood cultures, resulting in shorter hospital stays and saving money per each patient tested.⁴

In March 2023, Veteran Health Indiana (VHI) implemented the BioFire FilmArray Blood Culture Identification (BCID2), a BSI panel test that identifies select gram-negative bacteria, gram-positive bacteria, yeast, and antimicrobial resistance genes with an aggregate sensitivity of 99% and a specificity of 99.8%. The BCID2 presents clinically relevant information faster than traditional culture methods, allowing clinicians to make more efficient and educated antibiotic regimen decisions than with previous methods.⁵

It takes 24 to 48 hours from blood collection for culture incubation, positivity, and gram staining to occur at VHI. If the gram stain is positive, the blood culture is placed on the BioFire BCID2 in addition to traditional culture medium. BioFire BCID2 results are ready in 45 to 60 minutes. Results are uploaded into the electronic health record (EHR) ≤ 2 hours after they are obtained and the primary team is notified if the test is positive for certain critical results. Susceptibility testing of an identified organism typically requires an additional 24 to 48 hours for finalization. VHI Infectious Disease created an evidence-based antibiotic recommendation chart for certain medication(s) and alternate therapies based on the reported organism and its interpreted presence of resistance markers (eg, ceftriaxone for Escherichia coli when extended-spectrum beta lactamases are not detected vs meropenem if extended-spectrum beta lactamases marker are present). These charts optimize the antibiotic regimen while awaiting susceptibility finalizations.

Two previous studies describe the impact of rapid diagnostic testing technology at US Department of Veterans Affairs (VA) medical centers.^6,7 In Texas, the ASP reviewed BCID panel results via clinical decision support software for about 1 hour per day.⁶ A Los Angeles study analyzed the impact of Biofire BCID with an interpretation guide centered on unnecessary vancomycin use and determined that shorter duration of the medication may have been the result of more frequent infectious disease consultation.⁷

This study assessed the time to optimal antibiotic de-escalation before and after the implementation of BioFire BCID2 with results reviewed by the ASP without active notification or assistance of any clinical decision support technology. The primary objective was to evaluate difference in time to optimal antibiotics from blood culture draw pre- vs postintervention. Secondary objectives included differences in time to organism identification, difference in time on broad-spectrum antibiotics, and difference in time to appropriate antibiotics.

Methods

This quasi-experimental retrospective chart review assessed the impact of BioFire BCID2 use on timely antibiotic de-escalation for patients who experienced a BSI at VHI between March 1, 2022, and October 1, 2023. Microbiology laboratory records identified eligible patients with positive blood cultures within the study time frame. Data were collected from the VHI EHR.

Patients were included if they had a positive bacterial blood culture and received ≥ 1 antibiotic indicated for bacteremia while receiving inpatient care. Patients were excluded if they died prior to blood culture results, transferred out of VHI, left against medical advice, or had untreated contaminants in blood culture results (ie, never received antibiotics aimed at the contaminated culture).

Patient lists were generated for before and after implementation of BioFire BCID2 (pre- and postintervention) using the VHI EHR and microbiology laboratory record system. The pre- and postinterventions groups were different sizes. As a result, a random sampling of the preintervention group was selected and included patients from March 1, 2022, through March 26, 2023. The postintervention group was smaller due to time constraints between initiation of BioFire BCID2 for data collection and included all patients from March 27, 2023, through October 1, 2023.

Optimal antibiotics were defined as escalation from inappropriate therapy to broader agent(s), de-escalation from broad-spectrum therapy to targeted agent(s), discontinuation of therapy due to an organism being identified as a contaminant, or optimization of a regimen to the preferred antimicrobial agent based on evidence-based consensus guidelines. Broad-spectrum antibiotics included: piperacillin/tazobactam, cefepime, ceftazidime, ceftazidime-avibactam, cefiderocol, carbapenems, fluroquinolones, vancomycin, daptomycin, ceftaroline, linezolid, or aztreonam. Appropriate antibiotics were defined as those with activity toward the final identified organism(s).

Deidentified participant data were entered into Microsoft Excel and kept on a secure VA server to complete statistical analyses. Parametric continuous data, such as age, were analyzed using the t-test, while nonparametric continuous data, such as time to optimal antibiotics, were analyzed using the Mann-Whitney U test. Categorical data, like sex and race, were analyzed using either Fisher exact test for small sample sizes or X² test for a larger sample size. Statistical significance levels was defined as P < .05.

Results

Using patient lists drawn from the EHR and the microbiology laboratory records, 110 electronic charts were randomly selected for review. Fifteen patients were excluded: 8 had untreated contaminants, 4 died, and 3 were transferred out of VHI. Of the 95 patients included, 48 were in the preintervention group and 47 were in the postintervention group (Figure 1).

Baseline characteristics were similar between the 2 groups (Table 1). Most patients were White males aged > 70 years in the EHR. The urinary tract was the most common source of infection, impacting 12 patients in each group (Figure 2). Escherichia coli, Klebsiella, Staphylococcus, and Streptococcus were the most common bloodstream isolates identified.

The median time to optimal antibiotics in the preintervention group was 58.5 hours vs 43.4 hours in the postintervention group (P = .11). The median time to organism identification was 37.8 hours in the preintervention group vs 16.9 hours in the postintervention group (P < .001). The median time on broad-spectrum antibiotics was 45.2 hours in the preintervention group vs 46.6 hours in the postintervention group (P = .99). The median time on appropriate antibiotics in the preintervention group was 2.3 hours vs 1.9 hours in the postintervention group (P = .79). Differences in other measured outcomes between the groups were not statistically significant (Table 2).

Although implementation of rapid diagnostic technology reduced the median time to optimal antibiotics, the results were not statistically significant. Shorter time to organism identification in the postintervention group compared to the preintervention group was the lone statistically significant metric (P < .001).

Discussion

A lack of statistical significance in the primary outcome may have been due to nonadherence to facility de-escalation protocols or a suboptimal BioFire BCID2 result notification system. Additionally, use of rapid BCID at VHI may improve over time as clinicians become more familiar with the technology. Gaps in clinical pharmacy coverage during the night shift may have also contributed to delays in antibiotic optimization, particularly if other clinicians are not equipped with the knowledge or training to appropriately deescalate antibiotics based on microorganisms identified. A 2017 study by Donner et al concluded that physician interpretation of BCID results is suboptimal and should be augmented with clinical decision support tools as new technology becomes available.⁸ Despite the statistically insignificant results of this study, it did highlight potential areas of improvement which can lead to improved patient care.

Previous research has evaluated the impact of rapid BCID technology on antibiotic treatment and clinical outcomes. Chiasson et al found that median time to optimal therapy was 73.8 hours in the pre-BCID arm compared to 34.7 hours in the post- BCID arm (P ≤ .001), emphasizing the importance of combining rapid BCID with clinical decision support tools and pharmacy input.⁶ Senok et al found that BCID2 implementation led to a significant decrease in median time to culture result, which informed optimal antibiotic therapy and decreased 30-day mortality in the intensive care setting.⁹ In contrast, the current study did not stratify patients according to medical ward or illness severity even though clinicians may be less likely to de-escalate antibiotic therapy in critically ill patients.

Bae et al reported findings consistent with the current study and concluded that BCID did not affect the clinical outcomes of overall BSIs; however, it contributed to early administration of effective antibiotics in cases of BSIs caused by multidrug-resistant organisms.¹⁰ Results of this study were not stratified according to multidrug-resistant organisms because the sample size was too small. The current study also included patients with polymicrobial infections, which may have impacted the results due to a less streamlined approach to antibiotic optimization.

Limitations

This single-center, retrospective study had a small sample size, short time frame, and lacked patient diversity, and therefore may not be generalizable to other health care systems. The sample size was limited by shorter date range and smaller patient list between BioFire BCID2 implementation and data collection, which was used to determine the number of charts selected in each group. Some patients received antibiotics prior to blood cultures being drawn, which may falsely decrease time to optimal/ appropriate antibiotics and falsely increase time on broad spectrum/any antibiotics due to early antibiotic administration. The total number of patients on broad-spectrum antibiotics differed from the total number of patients for other outcomes because several patients never received the defined broad spectrum antibiotics.

Conclusions

When combined with a pre-existing ASP without active notification, the implementation of BioFire BCID2 did not return statistically significant data showing a decrease in time to optimal antibiotics, time to appropriate antibiotics, or time on broad-spectrum antibiotics at VHI. To make this program more successful, pharmacist intervention and clinical decision support tools may be needed.

Additional research is required to determine the optimal integration of antimicrobial stewardship, rapid diagnostic technology, and pharmacy services for maximum benefit. Even though the primary outcome was not statistically significant, the results may be clinically significant from a stewardship perspective. Realigning microbiology workflows to mimic other research, which emphasizes the importance of funneling rapid BCID results through the ASP, may improve outcomes. Future studies may be warranted following the implementation of clinical decision support tools to assess their impact on stewardship practices and patient outcomes.

Systemic lupus erythematosus (SLE), or lupus, is a rare autoimmune disease estimated to occur in about 5.1 cases per 100,000 person-years in the United States in 2018.¹ The disease predominantly affects females, with an incidence of 8.7 cases per 100,000 person-years vs 1.2 cases per 100,000 person-years in males, and is most common in patients aged 15 to 44 years.^1,2

Lupus presents with a constellation of clinical signs and symptoms that evolve, along with hallmark laboratory findings indicative of immune dysregulation and polyclonal B-cell activation. Consequently, a wide array of autoantibodies may be produced, although the combination of epitope specificity can vary from patient to patient.3 Nevertheless, > 98% of individuals diagnosed with lupus produce antinuclear antibodies (ANA), making ANA positivity a near-universal serologic feature at the time of diagnosis.

The pathogenesis of lupus is complex. Research from the past 5 decades supports the role of certain viral infections—such as Epstein-Barr virus (EBV) and cytomegalovirus—as potential triggers.⁴ These viruses are thought to initiate disease through mechanisms including activation of interferon pathways, exposure of cryptic intracellular antigens, molecular mimicry, and epitope spreading. Subsequent clonal expansion and autoantibody production occur to varying degrees, influenced by viral load and host susceptibility factors.

During the COVID-19 pandemic, it became evident that SARS-CoV-2 exerts profound effects on immune regulation, influencing infection outcomes through mechanisms such as hyperactivation of innate immunity, especially in the lungs, leading to acute respiratory distress syndrome. Additionally, SARS-CoV-2 has been associated with polyclonal B-cell activation and the generation of autoantibodies. This association gained attention after Bastard et al identified anti–type I interferon antibodies in patients with severe COVID-19, predominantly among males with a genetic predisposition. These autoantibodies were shown to impair antiviral defenses and contribute to life-threatening pneumonia.⁵

Subsequent studies demonstrated the production of a wide spectrum of functional autoantibodies, including ANA, in patients with COVID-19.^6,7 These findings were attributed to the acute expansion of autoreactive clones among naïve-derived immunoglobulin G1 antibody-secreting cells during the early stages of infection, with the degree of expansion correlating with disease severity.^8,9 Although longitudinal data up to 15 months postinfection suggest this serologic abnormality resolves in more than two-thirds of patients, the number of individuals infected globally has raised serious public health concerns regarding the potential long-term sequelae, including the onset of lupus or other autoimmune diseases in COVID-19 survivors.^6-9 A limited number of case reports describing the onset of lupus following SARS-CoV-2 infection support this hypothesis.¹⁰

This surveillance analysis investigates lupus incidence among patients within the Military Health System (MHS), encompassing all TRICARE beneficiaries, from January 2018 to December 2022. The objective of this analysis was to examine lupus incidence trends throughout the COVID-19 pandemic, stratified by sex, age, and active-duty status.

Methods

The MHS provides health care services to about 9.5 million US Department of Defense (DoD) beneficiaries. Outpatient health records and laboratory results for individuals receiving care at military treatment facilities (MTFs) between January 1, 2018, and December 31, 2022, were obtained from the Comprehensive Ambulatory/ Professional Encounter Record and MHS GENESIS. For beneficiaries receiving care in the private sector, data were sourced from the TRICARE Encounter Data—Non-Institutional database.

Laboratory test results, including ANA testing, were available only for individuals receiving care at MTFs. These laboratory data were extracted from the Composite Health Care System Chemistry database and MHS GENESIS laboratory systems for the same time frame. Inpatient data were not included in this analysis. Data from 2017 were used solely as a look-back (or washout) period to identify and exclude prevalent lupus cases diagnosed before 2018 and were not included in the final results.

Lupus cases were identified by the presence of a positive ANA test and appropriate International Classification of Diseases, 10th Revision, Clinical Modification (ICD-10-CM) codes. A positive ANA result was defined as either a qualitative result marked positive or a titer ≥ 1:80. The ICD-10-CM codes considered indicative of lupus included variations of M32, L93, or H01.12.

M32, L93, or H01.12. For cases with a positive ANA test, a lupus diagnosis required the presence of ≥ 2 lupus related ICD-10-CM codes. In the absence of ANA test results, a stricter criterion was applied: ≥ 4 lupus ICD-10-CM diagnosis codes recorded on separate days were required for inclusion.

Beneficiaries were excluded if they had a negative ANA result, only 1 lupus ICD- 10-CM diagnosis code, 1 positive ANA with only 1 corresponding ICD-10-CM code, or if their diagnosis occurred outside the defined study period. Patients and members of the public were not involved in the design, conduct, reporting, or dissemination of this study.

Results

Between January 1, 2017, and December 31, 2022, 99,946 TRICARE beneficiaries had some indication of lupus testing or diagnosis in their health records (Figure 1). Of these beneficiaries, 5335 had a positive ANA result and ≥ 2 ICD-10-CM lupus diagnosis codes. An additional 28,275 beneficiaries had ≥ 4 ICD-10-CM lupus diagnosis codes but no ANA test results. From these groups, the final sample included 10,760 beneficiaries who met the incident case definitions for SLE during the study period (2018 through 2022).

Most cases (85.1%, n = 9157) were diagnosed through TRICARE claims, while 1205 (11.2%) were diagnosed within the MHS. Another 398 (3.7%) had documentation of care both within and outside the MHS. Incident SLE cases declined by an average of 16% annually during the study period (Figure 2). This trend amounted to an overall reduction of 48.2%, from 2866 cases in 2018 to 1399 cases in 2022. This decline occurred despite total medical encounters among DoD beneficiaries remaining relatively stable during the pandemic years, with only a 3.5% change between 2018 and 2022.

The disease was more prevalent among female beneficiaries, with a female to- male ratio of 7:1 (Table 1). Among women, the number of new cases declined from 2519 in 2018 to 1223 in 2022, while the number of cases among men remained consistently < 350 annually. Similar trends were observed across other strata. Incident SLE cases were more common among nonactive-duty beneficiaries than active-duty service members, with a ratio of 18:1. New cases among active-duty members remained < 155 per year. Age-stratified data revealed that SLE was diagnosed predominantly in individuals aged ≥ 18 years, with a ratio of 37:1 compared with individuals aged < 18 years. Among children, the number of new cases remained < 75 per year throughout the study period.

A mean 56,850 ANA tests were conducted annually in centralized laboratories using standardized protocols (Table 2). The mean ANA positivity rate was 17.3%, which remained relatively stable from 2018 through 2022.

Discussion

This study examined the annual incidence of newly diagnosed SLE cases among all TRICARE beneficiaries from January 1, 2018, through December 31, 2022, covering both before and during the peak years of the COVID-19 pandemic. This analysis revealed a steady decline in SLE cases during this period. The reliability of these findings is reinforced by the comprehensiveness of the MHS, one of the largest US health care delivery systems, which maintains near-complete medical data capture for about 9.5 million DoD TRICARE beneficiaries across domestic and international settings.

SLE is a rare autoimmune disorder that presents a diagnostic challenge due to its wide range of nonspecific symptoms, many of which resemble other conditions. To reduce the likelihood of false-positive results and ensure diagnostic accuracy, this study adopted a stringent case definition. Incident cases were identified by the presence of ANA testing in conjunction with lupus-specific ICD-10-CM codes and required ≥ 4 lupus related diagnostic entries. This criterion was necessary due to the absence of ANA test results in data from private sector care settings. Our case definition aligns with established literature. For example, a Vanderbilt University chart review study demonstrated that combining ANA positivity with ≥ 4 lupus related ICD-10-CM codes achieves a positive predictive value of 100%, albeit with a sensitivity of 45%.¹¹ Other studies similarly affirm the diagnostic validity of using recurrent ICD-10-CM codes to improve specificity in identifying lupus cases.^12,13

The primary objective of this study was to examine the temporal trend in newly diagnosed lupus cases, rather than derive precise incidence rates. Although the TRICARE system includes about 9.5 million beneficiaries, this number represents a dynamic population with continual inflow and outflow. Accurate incidence rate calculation would require access to detailed denominator data, which were not readily available. In comparison with our findings, a study limited to active-duty service members reported fewer lupus cases. This discrepancy likely reflects differences in case definitions—specifically, the absence of laboratory data, the restricted range of diagnostic codes, and the requirement that diagnoses be rendered by specialists.¹⁴ Despite these differences, demographic patterns were consistent, with higher incidence observed in females and individuals aged ≥ 20 years.

A Centers for Disease Control and Prevention (CDC) study of lupus incidence in the general population also reported lower case counts.¹ However, the CDC estimates were based on 5 state-level registries, which rely on clinician-reported cases and therefore may underestimate true disease burden. Moreover, the DoD beneficiary population differs markedly from the general population: it includes a large cohort of retirees, ensuring an older demographic; all members have comprehensive health care access; and active-duty personnel are subject to pre-enlistment medical screening. Taken together, these factors suggest this study may offer a more complete and systematically captured profile of lupus incidence.

We observed a marked decline of newly diagnosed SLE cases during the study period, which coincided with the widespread circulation of COVID-19. This decrease is unlikely to be attributable to reduced access to care during the pandemic. The MHS operates under a single-payer model, and the total number of patient encounters remained relatively stable throughout the pandemic.

To our knowledge, this is the only study to monitor lupus incidence in a large US population over the 5-year period encompassing before and during the COVID-19 pandemic. To date, only 4 large-scale surveillance studies have addressed similar questions. ^14-17 Our findings are consistent with the most recent of these reports: an analysis limited to active-duty members of the US Armed Forces identified 1127 patients with newly diagnosed lupus between 2000 and 2022 and reported stable incidence trends throughout the pandemic.¹⁴ The other 3 studies adopted a different approach, comparing the emergence of autoimmune diseases, including lupus, between individuals with confirmed SARS-CoV-2 infection and those without. Each of these trials concluded that COVID-19 increases the risk of various autoimmune conditions, although the findings specific to lupus were inconsistent.^15-17

Chang et al reported a significant increase in new lupus diagnoses (n = 2,926,016), with an adjusted hazard ratio (aHR) of 2.99 (95% CI, 2.68-3.34), spanning all ages and both sexes. The highest incidence was observed in individuals of Asian descent.¹⁵ Using German routine health care data from 2020, Tesch et al identified a heightened risk of autoimmune diseases, including lupus, among patients with a history of SARS-CoV-2 infection (n = 641,407; 9.4% children, 57.3% female, 6.4% hospitalized), compared with matched infection-naïve controls (n = 1,560,357).¹⁶ Both studies excluded vaccinated individuals.

These 2 studies diverged in their assessment of the relationship between COVID-19 severity and subsequent autoimmune risk. Chang et al found a higher incidence among nonhospitalized ambulatory patients, while Tesch et al reported that increased risk was associated with patients requiring intensive care unit admission.^15,16

In contrast, based on a cohort of 4,197,188 individuals, Peng et al found no significant difference in lupus incidence among patients with SARS-CoV-2 infection (aHR, 1.05; 95% CI, 0.79-1.39).¹⁷ Notably, within the infected group, the incidence of SLE was significantly lower among vaccinated individuals compared with the unvaccinated group (aHR, 0.29; 95% CI, 0.18-0.47). Similar protective associations were observed for other antibody-mediated autoimmune disorders, including pemphigoid, Graves’ disease, and antiphospholipid antibody syndrome.

Limitations

Due to fundamental differences in study design, we were unable to directly reconcile our findings with those reported in the literature. This study lacked access to reliable documentation of COVID-19 infection status, primarily due to the widespread use of home testing among TRICARE beneficiaries. Additionally, the dataset did not include inpatient records and therefore did not permit evaluation of disease severity. Despite these constraints, it is plausible that the overall burden of COVID-19 infection within the study population was lower than that observed in the general US population.

As of December 2022, the DoD had reported about 750,000 confirmed COVID-19 cases among military personnel, civilian employees, dependents, and DoD contractors.¹⁸ Given that TRICARE beneficiaries represent about 2.8% of the total US population—and that > 90 million US individuals were infected between 2020 and 2022—the implied infection rate in our cohort appears to be about one-third of what might be expected.¹⁹ This discrepancy may be due to higher adherence to mitigation measures, such as social distancing and mask usage, among DoD-affiliated populations. COVID-19 vaccination was mandated for all active-duty service members, who constitute 5.4% of the study population. The remaining TRICARE beneficiaries also had access to guaranteed health care and vaccination coverage, likely contributing to high overall vaccination rates.

Because > 80% of the study population was composed of individuals from diverse civilian backgrounds, we expect the distribution of infection severity within the DoD beneficiary population to approximate that of the general US population.

Future Directions

The findings of this study offer circumstantial yet real-time evidence of the complexity underlying immune dysregulation at the intersection of host susceptibility and environmental exposures. The stability in ANA positivity rates during the study period mitigates concerns regarding undiagnosed subclinical lupus and may suggest that, overall, immune homeostasis was preserved among DoD beneficiaries.

It is noteworthy that during the COVID-19 pandemic, the incidence of several common infections—such as influenza and EBV—declined markedly, likely as a result of widespread social distancing and other public health interventions.²⁰ Mitigation strategies implemented within the military may have conferred protection not only against COVID-19 but also against other community-acquired pathogens.

In light of these observations, we hypothesize that for COVID-19 to act as a trigger for SLE, a prolonged or repeated disruption of immune equilibrium may be required—potentially mediated by recurrent infections or sustained inflammatory states. The association between viral infections and autoimmunity is well established. Immune dysregulation leading to autoantibody production has been observed not only in the context of SARS-CoV-2 but also following infections with EBV, cytomegalovirus, enteroviruses, hepatitis B and C viruses, HIV, and parvovirus B19.²¹

This dysregulation is often transient, accompanied by compensatory immune regulatory responses. However, in individuals subjected to successive or overlapping infections, these regulatory mechanisms may become compromised or overwhelmed, due to emergent patterns of immune interference of varying severity. In such cases, a transient immune perturbation may progress into a bona fide autoimmune disease, contingent upon individual risk factors such as genetic predisposition, preexisting immune memory, and regenerative capacity.²¹

Therefore, we believe the significance of this study is 2-fold. First, lupus is known to develop gradually and may require 3 to 5 years to clinically manifest after the initial break in immunological tolerance.³ Continued public health surveillance represents a more pragmatic strategy than retrospective cohort construction, especially as histories of COVID-19 infection become increasingly complete and definitive. Our findings provide a valuable baseline reference point for future longitudinal studies.

The interpretation of surveillance outcomes—whether indicating an upward trend, a stable baseline, or a downward trend—offers distinct analytical value. Within this study population, we observed neither an upward trajectory that might suggest a direct causal link, nor a flat trend that would imply absence of association between COVID-19 and lupus pathogenesis. Instead, the observation of a downward trend invites consideration of nonlinear or protective influences. From this perspective, we recommend that future investigations adopt a holistic framework when assessing environmental contributions to immune dysregulation—particularly when evaluating the long-term immunopathological consequences of the COVID-19 pandemic on lupus and related autoimmune conditions.

Conclusions

This study identified a declining trend in incident lupus cases during the COVID-19 pandemic among the DoD beneficiary population. Further investigation is warranted to elucidate the underlying factors contributing to this decline. Conducting longitudinal epidemiologic studies and applying multivariable regression analyses will be essential to determine whether incidence rates revert to prepandemic baselines and how these trends may be influenced by evolving environmental factors within the general population.

Systemic lupus erythematosus (SLE), or lupus, is a rare autoimmune disease estimated to occur in about 5.1 cases per 100,000 person-years in the United States in 2018.¹ The disease predominantly affects females, with an incidence of 8.7 cases per 100,000 person-years vs 1.2 cases per 100,000 person-years in males, and is most common in patients aged 15 to 44 years.^1,2

Lupus presents with a constellation of clinical signs and symptoms that evolve, along with hallmark laboratory findings indicative of immune dysregulation and polyclonal B-cell activation. Consequently, a wide array of autoantibodies may be produced, although the combination of epitope specificity can vary from patient to patient.3 Nevertheless, > 98% of individuals diagnosed with lupus produce antinuclear antibodies (ANA), making ANA positivity a near-universal serologic feature at the time of diagnosis.

The pathogenesis of lupus is complex. Research from the past 5 decades supports the role of certain viral infections—such as Epstein-Barr virus (EBV) and cytomegalovirus—as potential triggers.⁴ These viruses are thought to initiate disease through mechanisms including activation of interferon pathways, exposure of cryptic intracellular antigens, molecular mimicry, and epitope spreading. Subsequent clonal expansion and autoantibody production occur to varying degrees, influenced by viral load and host susceptibility factors.

During the COVID-19 pandemic, it became evident that SARS-CoV-2 exerts profound effects on immune regulation, influencing infection outcomes through mechanisms such as hyperactivation of innate immunity, especially in the lungs, leading to acute respiratory distress syndrome. Additionally, SARS-CoV-2 has been associated with polyclonal B-cell activation and the generation of autoantibodies. This association gained attention after Bastard et al identified anti–type I interferon antibodies in patients with severe COVID-19, predominantly among males with a genetic predisposition. These autoantibodies were shown to impair antiviral defenses and contribute to life-threatening pneumonia.⁵

Subsequent studies demonstrated the production of a wide spectrum of functional autoantibodies, including ANA, in patients with COVID-19.^6,7 These findings were attributed to the acute expansion of autoreactive clones among naïve-derived immunoglobulin G1 antibody-secreting cells during the early stages of infection, with the degree of expansion correlating with disease severity.^8,9 Although longitudinal data up to 15 months postinfection suggest this serologic abnormality resolves in more than two-thirds of patients, the number of individuals infected globally has raised serious public health concerns regarding the potential long-term sequelae, including the onset of lupus or other autoimmune diseases in COVID-19 survivors.^6-9 A limited number of case reports describing the onset of lupus following SARS-CoV-2 infection support this hypothesis.¹⁰

This surveillance analysis investigates lupus incidence among patients within the Military Health System (MHS), encompassing all TRICARE beneficiaries, from January 2018 to December 2022. The objective of this analysis was to examine lupus incidence trends throughout the COVID-19 pandemic, stratified by sex, age, and active-duty status.

Methods

The MHS provides health care services to about 9.5 million US Department of Defense (DoD) beneficiaries. Outpatient health records and laboratory results for individuals receiving care at military treatment facilities (MTFs) between January 1, 2018, and December 31, 2022, were obtained from the Comprehensive Ambulatory/ Professional Encounter Record and MHS GENESIS. For beneficiaries receiving care in the private sector, data were sourced from the TRICARE Encounter Data—Non-Institutional database.

Laboratory test results, including ANA testing, were available only for individuals receiving care at MTFs. These laboratory data were extracted from the Composite Health Care System Chemistry database and MHS GENESIS laboratory systems for the same time frame. Inpatient data were not included in this analysis. Data from 2017 were used solely as a look-back (or washout) period to identify and exclude prevalent lupus cases diagnosed before 2018 and were not included in the final results.

Lupus cases were identified by the presence of a positive ANA test and appropriate International Classification of Diseases, 10th Revision, Clinical Modification (ICD-10-CM) codes. A positive ANA result was defined as either a qualitative result marked positive or a titer ≥ 1:80. The ICD-10-CM codes considered indicative of lupus included variations of M32, L93, or H01.12.

M32, L93, or H01.12. For cases with a positive ANA test, a lupus diagnosis required the presence of ≥ 2 lupus related ICD-10-CM codes. In the absence of ANA test results, a stricter criterion was applied: ≥ 4 lupus ICD-10-CM diagnosis codes recorded on separate days were required for inclusion.

Beneficiaries were excluded if they had a negative ANA result, only 1 lupus ICD- 10-CM diagnosis code, 1 positive ANA with only 1 corresponding ICD-10-CM code, or if their diagnosis occurred outside the defined study period. Patients and members of the public were not involved in the design, conduct, reporting, or dissemination of this study.

Results

Between January 1, 2017, and December 31, 2022, 99,946 TRICARE beneficiaries had some indication of lupus testing or diagnosis in their health records (Figure 1). Of these beneficiaries, 5335 had a positive ANA result and ≥ 2 ICD-10-CM lupus diagnosis codes. An additional 28,275 beneficiaries had ≥ 4 ICD-10-CM lupus diagnosis codes but no ANA test results. From these groups, the final sample included 10,760 beneficiaries who met the incident case definitions for SLE during the study period (2018 through 2022).

Most cases (85.1%, n = 9157) were diagnosed through TRICARE claims, while 1205 (11.2%) were diagnosed within the MHS. Another 398 (3.7%) had documentation of care both within and outside the MHS. Incident SLE cases declined by an average of 16% annually during the study period (Figure 2). This trend amounted to an overall reduction of 48.2%, from 2866 cases in 2018 to 1399 cases in 2022. This decline occurred despite total medical encounters among DoD beneficiaries remaining relatively stable during the pandemic years, with only a 3.5% change between 2018 and 2022.

The disease was more prevalent among female beneficiaries, with a female to- male ratio of 7:1 (Table 1). Among women, the number of new cases declined from 2519 in 2018 to 1223 in 2022, while the number of cases among men remained consistently < 350 annually. Similar trends were observed across other strata. Incident SLE cases were more common among nonactive-duty beneficiaries than active-duty service members, with a ratio of 18:1. New cases among active-duty members remained < 155 per year. Age-stratified data revealed that SLE was diagnosed predominantly in individuals aged ≥ 18 years, with a ratio of 37:1 compared with individuals aged < 18 years. Among children, the number of new cases remained < 75 per year throughout the study period.

A mean 56,850 ANA tests were conducted annually in centralized laboratories using standardized protocols (Table 2). The mean ANA positivity rate was 17.3%, which remained relatively stable from 2018 through 2022.

Discussion

This study examined the annual incidence of newly diagnosed SLE cases among all TRICARE beneficiaries from January 1, 2018, through December 31, 2022, covering both before and during the peak years of the COVID-19 pandemic. This analysis revealed a steady decline in SLE cases during this period. The reliability of these findings is reinforced by the comprehensiveness of the MHS, one of the largest US health care delivery systems, which maintains near-complete medical data capture for about 9.5 million DoD TRICARE beneficiaries across domestic and international settings.

SLE is a rare autoimmune disorder that presents a diagnostic challenge due to its wide range of nonspecific symptoms, many of which resemble other conditions. To reduce the likelihood of false-positive results and ensure diagnostic accuracy, this study adopted a stringent case definition. Incident cases were identified by the presence of ANA testing in conjunction with lupus-specific ICD-10-CM codes and required ≥ 4 lupus related diagnostic entries. This criterion was necessary due to the absence of ANA test results in data from private sector care settings. Our case definition aligns with established literature. For example, a Vanderbilt University chart review study demonstrated that combining ANA positivity with ≥ 4 lupus related ICD-10-CM codes achieves a positive predictive value of 100%, albeit with a sensitivity of 45%.¹¹ Other studies similarly affirm the diagnostic validity of using recurrent ICD-10-CM codes to improve specificity in identifying lupus cases.^12,13

The primary objective of this study was to examine the temporal trend in newly diagnosed lupus cases, rather than derive precise incidence rates. Although the TRICARE system includes about 9.5 million beneficiaries, this number represents a dynamic population with continual inflow and outflow. Accurate incidence rate calculation would require access to detailed denominator data, which were not readily available. In comparison with our findings, a study limited to active-duty service members reported fewer lupus cases. This discrepancy likely reflects differences in case definitions—specifically, the absence of laboratory data, the restricted range of diagnostic codes, and the requirement that diagnoses be rendered by specialists.¹⁴ Despite these differences, demographic patterns were consistent, with higher incidence observed in females and individuals aged ≥ 20 years.

A Centers for Disease Control and Prevention (CDC) study of lupus incidence in the general population also reported lower case counts.¹ However, the CDC estimates were based on 5 state-level registries, which rely on clinician-reported cases and therefore may underestimate true disease burden. Moreover, the DoD beneficiary population differs markedly from the general population: it includes a large cohort of retirees, ensuring an older demographic; all members have comprehensive health care access; and active-duty personnel are subject to pre-enlistment medical screening. Taken together, these factors suggest this study may offer a more complete and systematically captured profile of lupus incidence.

We observed a marked decline of newly diagnosed SLE cases during the study period, which coincided with the widespread circulation of COVID-19. This decrease is unlikely to be attributable to reduced access to care during the pandemic. The MHS operates under a single-payer model, and the total number of patient encounters remained relatively stable throughout the pandemic.

To our knowledge, this is the only study to monitor lupus incidence in a large US population over the 5-year period encompassing before and during the COVID-19 pandemic. To date, only 4 large-scale surveillance studies have addressed similar questions. ^14-17 Our findings are consistent with the most recent of these reports: an analysis limited to active-duty members of the US Armed Forces identified 1127 patients with newly diagnosed lupus between 2000 and 2022 and reported stable incidence trends throughout the pandemic.¹⁴ The other 3 studies adopted a different approach, comparing the emergence of autoimmune diseases, including lupus, between individuals with confirmed SARS-CoV-2 infection and those without. Each of these trials concluded that COVID-19 increases the risk of various autoimmune conditions, although the findings specific to lupus were inconsistent.^15-17

Chang et al reported a significant increase in new lupus diagnoses (n = 2,926,016), with an adjusted hazard ratio (aHR) of 2.99 (95% CI, 2.68-3.34), spanning all ages and both sexes. The highest incidence was observed in individuals of Asian descent.¹⁵ Using German routine health care data from 2020, Tesch et al identified a heightened risk of autoimmune diseases, including lupus, among patients with a history of SARS-CoV-2 infection (n = 641,407; 9.4% children, 57.3% female, 6.4% hospitalized), compared with matched infection-naïve controls (n = 1,560,357).¹⁶ Both studies excluded vaccinated individuals.

These 2 studies diverged in their assessment of the relationship between COVID-19 severity and subsequent autoimmune risk. Chang et al found a higher incidence among nonhospitalized ambulatory patients, while Tesch et al reported that increased risk was associated with patients requiring intensive care unit admission.^15,16

In contrast, based on a cohort of 4,197,188 individuals, Peng et al found no significant difference in lupus incidence among patients with SARS-CoV-2 infection (aHR, 1.05; 95% CI, 0.79-1.39).¹⁷ Notably, within the infected group, the incidence of SLE was significantly lower among vaccinated individuals compared with the unvaccinated group (aHR, 0.29; 95% CI, 0.18-0.47). Similar protective associations were observed for other antibody-mediated autoimmune disorders, including pemphigoid, Graves’ disease, and antiphospholipid antibody syndrome.

Limitations

Due to fundamental differences in study design, we were unable to directly reconcile our findings with those reported in the literature. This study lacked access to reliable documentation of COVID-19 infection status, primarily due to the widespread use of home testing among TRICARE beneficiaries. Additionally, the dataset did not include inpatient records and therefore did not permit evaluation of disease severity. Despite these constraints, it is plausible that the overall burden of COVID-19 infection within the study population was lower than that observed in the general US population.

As of December 2022, the DoD had reported about 750,000 confirmed COVID-19 cases among military personnel, civilian employees, dependents, and DoD contractors.¹⁸ Given that TRICARE beneficiaries represent about 2.8% of the total US population—and that > 90 million US individuals were infected between 2020 and 2022—the implied infection rate in our cohort appears to be about one-third of what might be expected.¹⁹ This discrepancy may be due to higher adherence to mitigation measures, such as social distancing and mask usage, among DoD-affiliated populations. COVID-19 vaccination was mandated for all active-duty service members, who constitute 5.4% of the study population. The remaining TRICARE beneficiaries also had access to guaranteed health care and vaccination coverage, likely contributing to high overall vaccination rates.

Because > 80% of the study population was composed of individuals from diverse civilian backgrounds, we expect the distribution of infection severity within the DoD beneficiary population to approximate that of the general US population.

Future Directions

The findings of this study offer circumstantial yet real-time evidence of the complexity underlying immune dysregulation at the intersection of host susceptibility and environmental exposures. The stability in ANA positivity rates during the study period mitigates concerns regarding undiagnosed subclinical lupus and may suggest that, overall, immune homeostasis was preserved among DoD beneficiaries.

It is noteworthy that during the COVID-19 pandemic, the incidence of several common infections—such as influenza and EBV—declined markedly, likely as a result of widespread social distancing and other public health interventions.²⁰ Mitigation strategies implemented within the military may have conferred protection not only against COVID-19 but also against other community-acquired pathogens.

In light of these observations, we hypothesize that for COVID-19 to act as a trigger for SLE, a prolonged or repeated disruption of immune equilibrium may be required—potentially mediated by recurrent infections or sustained inflammatory states. The association between viral infections and autoimmunity is well established. Immune dysregulation leading to autoantibody production has been observed not only in the context of SARS-CoV-2 but also following infections with EBV, cytomegalovirus, enteroviruses, hepatitis B and C viruses, HIV, and parvovirus B19.²¹

This dysregulation is often transient, accompanied by compensatory immune regulatory responses. However, in individuals subjected to successive or overlapping infections, these regulatory mechanisms may become compromised or overwhelmed, due to emergent patterns of immune interference of varying severity. In such cases, a transient immune perturbation may progress into a bona fide autoimmune disease, contingent upon individual risk factors such as genetic predisposition, preexisting immune memory, and regenerative capacity.²¹

Therefore, we believe the significance of this study is 2-fold. First, lupus is known to develop gradually and may require 3 to 5 years to clinically manifest after the initial break in immunological tolerance.³ Continued public health surveillance represents a more pragmatic strategy than retrospective cohort construction, especially as histories of COVID-19 infection become increasingly complete and definitive. Our findings provide a valuable baseline reference point for future longitudinal studies.

The interpretation of surveillance outcomes—whether indicating an upward trend, a stable baseline, or a downward trend—offers distinct analytical value. Within this study population, we observed neither an upward trajectory that might suggest a direct causal link, nor a flat trend that would imply absence of association between COVID-19 and lupus pathogenesis. Instead, the observation of a downward trend invites consideration of nonlinear or protective influences. From this perspective, we recommend that future investigations adopt a holistic framework when assessing environmental contributions to immune dysregulation—particularly when evaluating the long-term immunopathological consequences of the COVID-19 pandemic on lupus and related autoimmune conditions.

Conclusions

This study identified a declining trend in incident lupus cases during the COVID-19 pandemic among the DoD beneficiary population. Further investigation is warranted to elucidate the underlying factors contributing to this decline. Conducting longitudinal epidemiologic studies and applying multivariable regression analyses will be essential to determine whether incidence rates revert to prepandemic baselines and how these trends may be influenced by evolving environmental factors within the general population.

Systemic lupus erythematosus (SLE), or lupus, is a rare autoimmune disease estimated to occur in about 5.1 cases per 100,000 person-years in the United States in 2018.¹ The disease predominantly affects females, with an incidence of 8.7 cases per 100,000 person-years vs 1.2 cases per 100,000 person-years in males, and is most common in patients aged 15 to 44 years.^1,2

Lupus presents with a constellation of clinical signs and symptoms that evolve, along with hallmark laboratory findings indicative of immune dysregulation and polyclonal B-cell activation. Consequently, a wide array of autoantibodies may be produced, although the combination of epitope specificity can vary from patient to patient.3 Nevertheless, > 98% of individuals diagnosed with lupus produce antinuclear antibodies (ANA), making ANA positivity a near-universal serologic feature at the time of diagnosis.

The pathogenesis of lupus is complex. Research from the past 5 decades supports the role of certain viral infections—such as Epstein-Barr virus (EBV) and cytomegalovirus—as potential triggers.⁴ These viruses are thought to initiate disease through mechanisms including activation of interferon pathways, exposure of cryptic intracellular antigens, molecular mimicry, and epitope spreading. Subsequent clonal expansion and autoantibody production occur to varying degrees, influenced by viral load and host susceptibility factors.

During the COVID-19 pandemic, it became evident that SARS-CoV-2 exerts profound effects on immune regulation, influencing infection outcomes through mechanisms such as hyperactivation of innate immunity, especially in the lungs, leading to acute respiratory distress syndrome. Additionally, SARS-CoV-2 has been associated with polyclonal B-cell activation and the generation of autoantibodies. This association gained attention after Bastard et al identified anti–type I interferon antibodies in patients with severe COVID-19, predominantly among males with a genetic predisposition. These autoantibodies were shown to impair antiviral defenses and contribute to life-threatening pneumonia.⁵

Subsequent studies demonstrated the production of a wide spectrum of functional autoantibodies, including ANA, in patients with COVID-19.^6,7 These findings were attributed to the acute expansion of autoreactive clones among naïve-derived immunoglobulin G1 antibody-secreting cells during the early stages of infection, with the degree of expansion correlating with disease severity.^8,9 Although longitudinal data up to 15 months postinfection suggest this serologic abnormality resolves in more than two-thirds of patients, the number of individuals infected globally has raised serious public health concerns regarding the potential long-term sequelae, including the onset of lupus or other autoimmune diseases in COVID-19 survivors.^6-9 A limited number of case reports describing the onset of lupus following SARS-CoV-2 infection support this hypothesis.¹⁰

This surveillance analysis investigates lupus incidence among patients within the Military Health System (MHS), encompassing all TRICARE beneficiaries, from January 2018 to December 2022. The objective of this analysis was to examine lupus incidence trends throughout the COVID-19 pandemic, stratified by sex, age, and active-duty status.

Methods

The MHS provides health care services to about 9.5 million US Department of Defense (DoD) beneficiaries. Outpatient health records and laboratory results for individuals receiving care at military treatment facilities (MTFs) between January 1, 2018, and December 31, 2022, were obtained from the Comprehensive Ambulatory/ Professional Encounter Record and MHS GENESIS. For beneficiaries receiving care in the private sector, data were sourced from the TRICARE Encounter Data—Non-Institutional database.

Laboratory test results, including ANA testing, were available only for individuals receiving care at MTFs. These laboratory data were extracted from the Composite Health Care System Chemistry database and MHS GENESIS laboratory systems for the same time frame. Inpatient data were not included in this analysis. Data from 2017 were used solely as a look-back (or washout) period to identify and exclude prevalent lupus cases diagnosed before 2018 and were not included in the final results.

Lupus cases were identified by the presence of a positive ANA test and appropriate International Classification of Diseases, 10th Revision, Clinical Modification (ICD-10-CM) codes. A positive ANA result was defined as either a qualitative result marked positive or a titer ≥ 1:80. The ICD-10-CM codes considered indicative of lupus included variations of M32, L93, or H01.12.

M32, L93, or H01.12. For cases with a positive ANA test, a lupus diagnosis required the presence of ≥ 2 lupus related ICD-10-CM codes. In the absence of ANA test results, a stricter criterion was applied: ≥ 4 lupus ICD-10-CM diagnosis codes recorded on separate days were required for inclusion.

Beneficiaries were excluded if they had a negative ANA result, only 1 lupus ICD- 10-CM diagnosis code, 1 positive ANA with only 1 corresponding ICD-10-CM code, or if their diagnosis occurred outside the defined study period. Patients and members of the public were not involved in the design, conduct, reporting, or dissemination of this study.

Results

Between January 1, 2017, and December 31, 2022, 99,946 TRICARE beneficiaries had some indication of lupus testing or diagnosis in their health records (Figure 1). Of these beneficiaries, 5335 had a positive ANA result and ≥ 2 ICD-10-CM lupus diagnosis codes. An additional 28,275 beneficiaries had ≥ 4 ICD-10-CM lupus diagnosis codes but no ANA test results. From these groups, the final sample included 10,760 beneficiaries who met the incident case definitions for SLE during the study period (2018 through 2022).

Most cases (85.1%, n = 9157) were diagnosed through TRICARE claims, while 1205 (11.2%) were diagnosed within the MHS. Another 398 (3.7%) had documentation of care both within and outside the MHS. Incident SLE cases declined by an average of 16% annually during the study period (Figure 2). This trend amounted to an overall reduction of 48.2%, from 2866 cases in 2018 to 1399 cases in 2022. This decline occurred despite total medical encounters among DoD beneficiaries remaining relatively stable during the pandemic years, with only a 3.5% change between 2018 and 2022.

The disease was more prevalent among female beneficiaries, with a female to- male ratio of 7:1 (Table 1). Among women, the number of new cases declined from 2519 in 2018 to 1223 in 2022, while the number of cases among men remained consistently < 350 annually. Similar trends were observed across other strata. Incident SLE cases were more common among nonactive-duty beneficiaries than active-duty service members, with a ratio of 18:1. New cases among active-duty members remained < 155 per year. Age-stratified data revealed that SLE was diagnosed predominantly in individuals aged ≥ 18 years, with a ratio of 37:1 compared with individuals aged < 18 years. Among children, the number of new cases remained < 75 per year throughout the study period.

A mean 56,850 ANA tests were conducted annually in centralized laboratories using standardized protocols (Table 2). The mean ANA positivity rate was 17.3%, which remained relatively stable from 2018 through 2022.

Discussion

This study examined the annual incidence of newly diagnosed SLE cases among all TRICARE beneficiaries from January 1, 2018, through December 31, 2022, covering both before and during the peak years of the COVID-19 pandemic. This analysis revealed a steady decline in SLE cases during this period. The reliability of these findings is reinforced by the comprehensiveness of the MHS, one of the largest US health care delivery systems, which maintains near-complete medical data capture for about 9.5 million DoD TRICARE beneficiaries across domestic and international settings.

SLE is a rare autoimmune disorder that presents a diagnostic challenge due to its wide range of nonspecific symptoms, many of which resemble other conditions. To reduce the likelihood of false-positive results and ensure diagnostic accuracy, this study adopted a stringent case definition. Incident cases were identified by the presence of ANA testing in conjunction with lupus-specific ICD-10-CM codes and required ≥ 4 lupus related diagnostic entries. This criterion was necessary due to the absence of ANA test results in data from private sector care settings. Our case definition aligns with established literature. For example, a Vanderbilt University chart review study demonstrated that combining ANA positivity with ≥ 4 lupus related ICD-10-CM codes achieves a positive predictive value of 100%, albeit with a sensitivity of 45%.¹¹ Other studies similarly affirm the diagnostic validity of using recurrent ICD-10-CM codes to improve specificity in identifying lupus cases.^12,13

The primary objective of this study was to examine the temporal trend in newly diagnosed lupus cases, rather than derive precise incidence rates. Although the TRICARE system includes about 9.5 million beneficiaries, this number represents a dynamic population with continual inflow and outflow. Accurate incidence rate calculation would require access to detailed denominator data, which were not readily available. In comparison with our findings, a study limited to active-duty service members reported fewer lupus cases. This discrepancy likely reflects differences in case definitions—specifically, the absence of laboratory data, the restricted range of diagnostic codes, and the requirement that diagnoses be rendered by specialists.¹⁴ Despite these differences, demographic patterns were consistent, with higher incidence observed in females and individuals aged ≥ 20 years.

A Centers for Disease Control and Prevention (CDC) study of lupus incidence in the general population also reported lower case counts.¹ However, the CDC estimates were based on 5 state-level registries, which rely on clinician-reported cases and therefore may underestimate true disease burden. Moreover, the DoD beneficiary population differs markedly from the general population: it includes a large cohort of retirees, ensuring an older demographic; all members have comprehensive health care access; and active-duty personnel are subject to pre-enlistment medical screening. Taken together, these factors suggest this study may offer a more complete and systematically captured profile of lupus incidence.

We observed a marked decline of newly diagnosed SLE cases during the study period, which coincided with the widespread circulation of COVID-19. This decrease is unlikely to be attributable to reduced access to care during the pandemic. The MHS operates under a single-payer model, and the total number of patient encounters remained relatively stable throughout the pandemic.

To our knowledge, this is the only study to monitor lupus incidence in a large US population over the 5-year period encompassing before and during the COVID-19 pandemic. To date, only 4 large-scale surveillance studies have addressed similar questions. ^14-17 Our findings are consistent with the most recent of these reports: an analysis limited to active-duty members of the US Armed Forces identified 1127 patients with newly diagnosed lupus between 2000 and 2022 and reported stable incidence trends throughout the pandemic.¹⁴ The other 3 studies adopted a different approach, comparing the emergence of autoimmune diseases, including lupus, between individuals with confirmed SARS-CoV-2 infection and those without. Each of these trials concluded that COVID-19 increases the risk of various autoimmune conditions, although the findings specific to lupus were inconsistent.^15-17

Chang et al reported a significant increase in new lupus diagnoses (n = 2,926,016), with an adjusted hazard ratio (aHR) of 2.99 (95% CI, 2.68-3.34), spanning all ages and both sexes. The highest incidence was observed in individuals of Asian descent.¹⁵ Using German routine health care data from 2020, Tesch et al identified a heightened risk of autoimmune diseases, including lupus, among patients with a history of SARS-CoV-2 infection (n = 641,407; 9.4% children, 57.3% female, 6.4% hospitalized), compared with matched infection-naïve controls (n = 1,560,357).¹⁶ Both studies excluded vaccinated individuals.

These 2 studies diverged in their assessment of the relationship between COVID-19 severity and subsequent autoimmune risk. Chang et al found a higher incidence among nonhospitalized ambulatory patients, while Tesch et al reported that increased risk was associated with patients requiring intensive care unit admission.^15,16

In contrast, based on a cohort of 4,197,188 individuals, Peng et al found no significant difference in lupus incidence among patients with SARS-CoV-2 infection (aHR, 1.05; 95% CI, 0.79-1.39).¹⁷ Notably, within the infected group, the incidence of SLE was significantly lower among vaccinated individuals compared with the unvaccinated group (aHR, 0.29; 95% CI, 0.18-0.47). Similar protective associations were observed for other antibody-mediated autoimmune disorders, including pemphigoid, Graves’ disease, and antiphospholipid antibody syndrome.

Limitations

Due to fundamental differences in study design, we were unable to directly reconcile our findings with those reported in the literature. This study lacked access to reliable documentation of COVID-19 infection status, primarily due to the widespread use of home testing among TRICARE beneficiaries. Additionally, the dataset did not include inpatient records and therefore did not permit evaluation of disease severity. Despite these constraints, it is plausible that the overall burden of COVID-19 infection within the study population was lower than that observed in the general US population.

As of December 2022, the DoD had reported about 750,000 confirmed COVID-19 cases among military personnel, civilian employees, dependents, and DoD contractors.¹⁸ Given that TRICARE beneficiaries represent about 2.8% of the total US population—and that > 90 million US individuals were infected between 2020 and 2022—the implied infection rate in our cohort appears to be about one-third of what might be expected.¹⁹ This discrepancy may be due to higher adherence to mitigation measures, such as social distancing and mask usage, among DoD-affiliated populations. COVID-19 vaccination was mandated for all active-duty service members, who constitute 5.4% of the study population. The remaining TRICARE beneficiaries also had access to guaranteed health care and vaccination coverage, likely contributing to high overall vaccination rates.

Because > 80% of the study population was composed of individuals from diverse civilian backgrounds, we expect the distribution of infection severity within the DoD beneficiary population to approximate that of the general US population.

Future Directions

The findings of this study offer circumstantial yet real-time evidence of the complexity underlying immune dysregulation at the intersection of host susceptibility and environmental exposures. The stability in ANA positivity rates during the study period mitigates concerns regarding undiagnosed subclinical lupus and may suggest that, overall, immune homeostasis was preserved among DoD beneficiaries.

It is noteworthy that during the COVID-19 pandemic, the incidence of several common infections—such as influenza and EBV—declined markedly, likely as a result of widespread social distancing and other public health interventions.²⁰ Mitigation strategies implemented within the military may have conferred protection not only against COVID-19 but also against other community-acquired pathogens.

In light of these observations, we hypothesize that for COVID-19 to act as a trigger for SLE, a prolonged or repeated disruption of immune equilibrium may be required—potentially mediated by recurrent infections or sustained inflammatory states. The association between viral infections and autoimmunity is well established. Immune dysregulation leading to autoantibody production has been observed not only in the context of SARS-CoV-2 but also following infections with EBV, cytomegalovirus, enteroviruses, hepatitis B and C viruses, HIV, and parvovirus B19.²¹

This dysregulation is often transient, accompanied by compensatory immune regulatory responses. However, in individuals subjected to successive or overlapping infections, these regulatory mechanisms may become compromised or overwhelmed, due to emergent patterns of immune interference of varying severity. In such cases, a transient immune perturbation may progress into a bona fide autoimmune disease, contingent upon individual risk factors such as genetic predisposition, preexisting immune memory, and regenerative capacity.²¹

Therefore, we believe the significance of this study is 2-fold. First, lupus is known to develop gradually and may require 3 to 5 years to clinically manifest after the initial break in immunological tolerance.³ Continued public health surveillance represents a more pragmatic strategy than retrospective cohort construction, especially as histories of COVID-19 infection become increasingly complete and definitive. Our findings provide a valuable baseline reference point for future longitudinal studies.

The interpretation of surveillance outcomes—whether indicating an upward trend, a stable baseline, or a downward trend—offers distinct analytical value. Within this study population, we observed neither an upward trajectory that might suggest a direct causal link, nor a flat trend that would imply absence of association between COVID-19 and lupus pathogenesis. Instead, the observation of a downward trend invites consideration of nonlinear or protective influences. From this perspective, we recommend that future investigations adopt a holistic framework when assessing environmental contributions to immune dysregulation—particularly when evaluating the long-term immunopathological consequences of the COVID-19 pandemic on lupus and related autoimmune conditions.

Conclusions

This study identified a declining trend in incident lupus cases during the COVID-19 pandemic among the DoD beneficiary population. Further investigation is warranted to elucidate the underlying factors contributing to this decline. Conducting longitudinal epidemiologic studies and applying multivariable regression analyses will be essential to determine whether incidence rates revert to prepandemic baselines and how these trends may be influenced by evolving environmental factors within the general population.