Matthew M. Davis, MD, MAPP

Population

All discharges from January 1, 2006 through December 31, 2012 from C. S. Mott Children's Hospital were initially eligible. All hospitalizations were considered potential index admissions; therefore, children may contribute more than 1 hospitalization to the dataset. We excluded hospitalizations in which the patient died, was transferred to another institution, was discharged against medical advice, or was discharged to hospice. Newborns admitted to a normal newborn service were also excluded, as they do not represent a typical hospitalization for illness. Among newborns admitted to a higher‐intensity clinical service (eg, special care nursery or neonatal intensive care), we also excluded newborns with a length of stay <5 days, given the typical length of stay of up to 4 days for uncomplicated delivery via Cesarean section that would indicate infants for whom precautionary measures had been taken but there was low estimated health risk. We used International Classification of Diseases, Ninth Revision codes to identify children with complex chronic conditions (CCCs) and technology dependency.[8]

Outcome

We examined unplanned readmission within 30 days of discharge. We defined unplanned readmission as a readmission that was not entered into the hospital registration system at least 24 hours before discharge.[9] Additionally, we performed sensitive analyses examining any 30‐day readmissions.

Statistical Analysis

We fit multivariable logistic regression models for 30‐day unplanned readmission, with the primary predictor of either weekend (Saturday or Sunday) admission or weekend discharge (in separate models). We adjusted for patient age, gender, race/ethnicity, source of admission, insurance, and length of stay. We also adjusted for patient chronic illness complexity using the number of CCCs and technology dependency (yes/no). Variance in all analyses was clustered on individual patients.

Weekend Admission

Overall, 82% of admissions occurred during the week, with Tuesday as the highest admitting volume day (Figure 1). Children admitted on the weekend had higher odds of unplanned readmission compared to children admitted on weekdays (unadjusted odds ratio [OR]=1.15 [95% confidence interval {CI}: 1.07‐1.24]). Adjusting the analysis for age, gender, race/ethnicity, insurance, length of stay, CCCs, and technology dependency, higher odds of readmission remains significantly higher than weekday admission (adjusted OR=1.09 [95% CI: 1.004‐1.18]) (Table 1). Age, admission source, payer, length of stay, number of complex chronic conditions, and technology dependency were also significantly associated with readmission in the weekend admission model (see the Supporting Appendix Table in the online version of this article). A sensitivity analysis examining the association of weekend admission and readmission within different subpopulations of children with varying numbers of CCCs (ie, among children without CCCs, with 1 CCC, 2 CCCs, and 3+ CCCs) showed that the association remains the same in each subgroup. Further, a sensitivity analysis examining odds of any 30‐day readmission was similar to the primary analysis with higher odds of readmission in adjusted analysis (adjusted OR=1.09 [95% CI: 1.02‐1.18]).

Figure 1

Weekend Discharge

Weekend discharges accounted for 34% of all discharges. Fridays had the highest discharge volumes, with lowest discharge volumes on Sunday (Figure 1). Children discharged on the weekend had lower odds of unplanned readmission compared to children discharged on weekdays in bivariate analysis (unadjusted OR=0.91 [95% CI: 0.85‐0.97]). However, when adjusting for important confounders, the relationship was no longer statistically significant (adjusted OR=0.97 [95% CI: 0.91‐1.03]) (Table 1). Age, admission source, payer, length of stay, and number of complex chronic conditions were associated with readmission in the weekend discharge model (see the Supporting Appendix Table in the online version of this article). In a sensitivity analysis examining any 30‐day readmission, weekend discharge was not associated with readmission in adjusted analysis.

Population

All discharges from January 1, 2006 through December 31, 2012 from C. S. Mott Children's Hospital were initially eligible. All hospitalizations were considered potential index admissions; therefore, children may contribute more than 1 hospitalization to the dataset. We excluded hospitalizations in which the patient died, was transferred to another institution, was discharged against medical advice, or was discharged to hospice. Newborns admitted to a normal newborn service were also excluded, as they do not represent a typical hospitalization for illness. Among newborns admitted to a higher‐intensity clinical service (eg, special care nursery or neonatal intensive care), we also excluded newborns with a length of stay <5 days, given the typical length of stay of up to 4 days for uncomplicated delivery via Cesarean section that would indicate infants for whom precautionary measures had been taken but there was low estimated health risk. We used International Classification of Diseases, Ninth Revision codes to identify children with complex chronic conditions (CCCs) and technology dependency.[8]

Outcome

We examined unplanned readmission within 30 days of discharge. We defined unplanned readmission as a readmission that was not entered into the hospital registration system at least 24 hours before discharge.[9] Additionally, we performed sensitive analyses examining any 30‐day readmissions.

Statistical Analysis

We fit multivariable logistic regression models for 30‐day unplanned readmission, with the primary predictor of either weekend (Saturday or Sunday) admission or weekend discharge (in separate models). We adjusted for patient age, gender, race/ethnicity, source of admission, insurance, and length of stay. We also adjusted for patient chronic illness complexity using the number of CCCs and technology dependency (yes/no). Variance in all analyses was clustered on individual patients.

Weekend Admission

Overall, 82% of admissions occurred during the week, with Tuesday as the highest admitting volume day (Figure 1). Children admitted on the weekend had higher odds of unplanned readmission compared to children admitted on weekdays (unadjusted odds ratio [OR]=1.15 [95% confidence interval {CI}: 1.07‐1.24]). Adjusting the analysis for age, gender, race/ethnicity, insurance, length of stay, CCCs, and technology dependency, higher odds of readmission remains significantly higher than weekday admission (adjusted OR=1.09 [95% CI: 1.004‐1.18]) (Table 1). Age, admission source, payer, length of stay, number of complex chronic conditions, and technology dependency were also significantly associated with readmission in the weekend admission model (see the Supporting Appendix Table in the online version of this article). A sensitivity analysis examining the association of weekend admission and readmission within different subpopulations of children with varying numbers of CCCs (ie, among children without CCCs, with 1 CCC, 2 CCCs, and 3+ CCCs) showed that the association remains the same in each subgroup. Further, a sensitivity analysis examining odds of any 30‐day readmission was similar to the primary analysis with higher odds of readmission in adjusted analysis (adjusted OR=1.09 [95% CI: 1.02‐1.18]).

Figure 1

Weekend Discharge

Weekend discharges accounted for 34% of all discharges. Fridays had the highest discharge volumes, with lowest discharge volumes on Sunday (Figure 1). Children discharged on the weekend had lower odds of unplanned readmission compared to children discharged on weekdays in bivariate analysis (unadjusted OR=0.91 [95% CI: 0.85‐0.97]). However, when adjusting for important confounders, the relationship was no longer statistically significant (adjusted OR=0.97 [95% CI: 0.91‐1.03]) (Table 1). Age, admission source, payer, length of stay, and number of complex chronic conditions were associated with readmission in the weekend discharge model (see the Supporting Appendix Table in the online version of this article). In a sensitivity analysis examining any 30‐day readmission, weekend discharge was not associated with readmission in adjusted analysis.

Population

All discharges from January 1, 2006 through December 31, 2012 from C. S. Mott Children's Hospital were initially eligible. All hospitalizations were considered potential index admissions; therefore, children may contribute more than 1 hospitalization to the dataset. We excluded hospitalizations in which the patient died, was transferred to another institution, was discharged against medical advice, or was discharged to hospice. Newborns admitted to a normal newborn service were also excluded, as they do not represent a typical hospitalization for illness. Among newborns admitted to a higher‐intensity clinical service (eg, special care nursery or neonatal intensive care), we also excluded newborns with a length of stay <5 days, given the typical length of stay of up to 4 days for uncomplicated delivery via Cesarean section that would indicate infants for whom precautionary measures had been taken but there was low estimated health risk. We used International Classification of Diseases, Ninth Revision codes to identify children with complex chronic conditions (CCCs) and technology dependency.[8]

Outcome

We examined unplanned readmission within 30 days of discharge. We defined unplanned readmission as a readmission that was not entered into the hospital registration system at least 24 hours before discharge.[9] Additionally, we performed sensitive analyses examining any 30‐day readmissions.

Statistical Analysis

We fit multivariable logistic regression models for 30‐day unplanned readmission, with the primary predictor of either weekend (Saturday or Sunday) admission or weekend discharge (in separate models). We adjusted for patient age, gender, race/ethnicity, source of admission, insurance, and length of stay. We also adjusted for patient chronic illness complexity using the number of CCCs and technology dependency (yes/no). Variance in all analyses was clustered on individual patients.

Weekend Admission

Overall, 82% of admissions occurred during the week, with Tuesday as the highest admitting volume day (Figure 1). Children admitted on the weekend had higher odds of unplanned readmission compared to children admitted on weekdays (unadjusted odds ratio [OR]=1.15 [95% confidence interval {CI}: 1.07‐1.24]). Adjusting the analysis for age, gender, race/ethnicity, insurance, length of stay, CCCs, and technology dependency, higher odds of readmission remains significantly higher than weekday admission (adjusted OR=1.09 [95% CI: 1.004‐1.18]) (Table 1). Age, admission source, payer, length of stay, number of complex chronic conditions, and technology dependency were also significantly associated with readmission in the weekend admission model (see the Supporting Appendix Table in the online version of this article). A sensitivity analysis examining the association of weekend admission and readmission within different subpopulations of children with varying numbers of CCCs (ie, among children without CCCs, with 1 CCC, 2 CCCs, and 3+ CCCs) showed that the association remains the same in each subgroup. Further, a sensitivity analysis examining odds of any 30‐day readmission was similar to the primary analysis with higher odds of readmission in adjusted analysis (adjusted OR=1.09 [95% CI: 1.02‐1.18]).

Figure 1

Weekend Discharge

Weekend discharges accounted for 34% of all discharges. Fridays had the highest discharge volumes, with lowest discharge volumes on Sunday (Figure 1). Children discharged on the weekend had lower odds of unplanned readmission compared to children discharged on weekdays in bivariate analysis (unadjusted OR=0.91 [95% CI: 0.85‐0.97]). However, when adjusting for important confounders, the relationship was no longer statistically significant (adjusted OR=0.97 [95% CI: 0.91‐1.03]) (Table 1). Age, admission source, payer, length of stay, and number of complex chronic conditions were associated with readmission in the weekend discharge model (see the Supporting Appendix Table in the online version of this article). In a sensitivity analysis examining any 30‐day readmission, weekend discharge was not associated with readmission in adjusted analysis.

Search Strategy

With the assistance of a research librarian, we searched MEDLINE and CINAHL (Cumulative Index to Nursing and Allied Health Literature) from the inception of these databases through to March 28, 2012 (for search strategies, see the Supporting Information, Appendix, Part 1, in the online version of this article).

Study Selection

Two authors (K.A. and C.K.) independently reviewed abstracts identified by the initial search, as well as abstracts of references of included articles. Eligibility criteria for inclusion in full review included: (1) discharge‐oriented process or intervention initiated in the inpatient setting, (2) study outcomes related to subsequent utilization including hospital readmission or emergency department visit after hospitalization, (3) child‐ or adolescent‐focused or child‐specific results presented separately, and (4) written or available in English. If abstract review did not sufficiently clarify whether all eligibility criteria were met, the article was included in the full review. Two authors (K.A. and C.K.) independently reviewed articles meeting criteria for full review to determine eligibility. Disagreements regarding inclusion in the final analysis were discussed with all 4 authors. We excluded studies in countries with low or lower‐middle incomes,[15] as discharge interventions in these countries may not be broadly applicable.

Data Abstraction, Quality Assessment, and Data Synthesis

Two authors (K.A. and C.K.) independently abstracted data using a modified Cochrane Collaboration data collection form.[16] We independently scored the included studies using the Downs and Black checklist, which assesses the risk of bias and the quality of both randomized and nonrandomized studies.[17] This checklist yields a composite score of 0 to 28 points, excluding the item assessing power. As many studies either lacked power calculations or included power calculations based on outcomes not included in our review, we performed calculations to determine the sample size needed to detect a decrease in readmission or ED utilization by 20% from baseline or control rates. Due to the heterogeneous nature of included studies in terms of population, interventions, study design, and outcomes, meta‐analysis was not performed.

Patient Populations and Intervention Timing and Components

Studies varied regarding the specific medical conditions they evaluated. Eight of the papers reported discharge interventions for children with asthma, 5 papers focused on discharge from the neonatal intensive care unit (NICU), and a final study discussed a discharge intervention for children with cancer (Table 1). Although our primary goal was to synthesize discharge interventions across pediatric conditions, we provide a summary of discharge interventions by condition (see Supporting Information, Appendix, Part 3, in the online version of this article).

Studies varied regarding the timing and nature of the intervention components. Eight discharge interventions included a major inpatient component, in addition to outpatient support or follow‐up.[21, 23, 24, 25, 26, 29, 32, 34] Two studies included an inpatient education component only.[22, 27] The remainder were initiated during index hospitalization but focused primarily on home visitation, enhanced follow‐up, and support after discharge (Figure 2).[28, 30, 31, 33]

Figure 2

Outcome Assessment Methods

Readmission and subsequent ED utilization events were identified using multiple techniques. Some authors accessed claims records to capture all outcomes.[30, 33] Others relied on chart review.[21, 25, 26, 27, 28, 31, 32] One study supplemented hospital records with outpatient records.[24] Some investigators used parental reports.[22, 23, 31] Two studies did not describe methods for identifying postdischarge events.[29, 34]

Study Quality

The quality of the included studies varied (Table 2). Many of the studies had inadequate sample size to detect a difference in either readmission or ED visit subsequent to discharge. Eight studies found differences in either subsequent ED utilization, hospitalization, or both and were considered adequately powered for these specific outcomes.[21, 23, 25, 26, 28, 30, 31, 32] In contrast, among studies with readmission as an outcome, 6 were not adequately powered to detect a difference in this particular outcome.[24, 30, 31, 32, 33, 34] In these 6 studies, all except 1 study³⁰ had <10% of the sample size required to detect differences in readmission. Further, 2 studies that examined ED utilization were underpowered to detect differences between intervention and control groups.[24, 26] We were unable to perform power calculations for 3 studies,[22, 27, 29] as the authors presented the number of events without clear denominators.

Excluding the assessment of statistical power, Downs and Black scores ranged from 10 to 23 (maximum 28 possible points) indicating varying quality. As would be expected with discharge interventions, studies did not blind participants; 2 studies did, however, appropriately blind the outcome evaluators to intervention assignment.[22, 30] Even though 10 out of the 14 studies were randomized controlled trials, randomization may not have been completely effective due to sample size being too small for effective randomization,[31] large numbers of excluded subjects after randomization,[30] and unclear randomization process.[34] Several studies had varying follow‐up periods for patients within a given study. For example, 3 NICU studies assessed readmission at 1‐year corrected age,[30, 31, 33] creating the analytic difficulty that the amount of time a given patient was at risk for readmission was dependent on when the patient was discharged, yet this was not accounted for in the analyses. Only 2 studies demonstrated low rates of loss to follow‐up (<10%).[30, 33] The remainder of the studies either had high incompletion/loss to follow‐up rates (>10%)[22, 24, 31] or did not report rates.[21, 23, 25, 26, 27, 28, 29, 32, 34] Finally, 3 studies recruited patients from multiple sites,[24, 31, 33] and none adjusted for potential differences in effect based on enrollment site.

Findings Across Patient Populations Regarding Readmission

Of the 4 studies that demonstrated change in overall readmission,[23, 25, 26, 28] all were asthma focused; 3 demonstrated a decrease in readmissions,[23, 25, 26] and 1 an increase in readmissions.[28] The 3 effective interventions included 1‐on‐1 inpatient education delivered by an asthma nurse, in addition to postdischarge follow‐up support, either by telephone or clinic visit. Two of these interventions provided rescue oral steroids to some patients on discharge.[25, 26] In contrast, a study from New Zealand evaluated a series of postdischarge visits using an existing public health nurse infrastructure and demonstrated an increase in readmission between 6 to 18 months after admission in European children.[28] An additional study focused on outpatient support after discharge from the NICU, and demonstrated a lower frequency of readmission to the intensive care unit without overall reduction of hospital readmission (Tables 1 and 2).[30]

Findings Across Patient Populations Regarding Subsequent ED Visits

Of all the discharge interventions, 6 demonstrated differences in return to the ED after discharge. Five studies described a decrease in ED visits after hospitalization,[23, 25, 30, 31, 32] and 1 showed an increase.[21] Three studies in the NICU population demonstrated decreased ED utilization through a combination of augmented family engagement during hospitalization and/or enhanced support after discharge. Two inpatient asthma education interventions with structured postdischarge follow‐up decreased return visitation to the ED.[23, 26] The intervention that worsened subsequent ED utilization (ie, increased ED visit hazard compared to matched controls) provided enhanced inpatient education to a nonrandom group of children hospitalized with asthma and provided a follow‐up phone call 3 weeks after discharge (Tables 1 and 2).[21]

Current Conceptual Framework

There are a number of existing discharge transitional care frameworks from prior studies[35, 36] and professional societies.[37] The Stepping Up to the Plate (SUTTP) alliance, a collaborative of 9 professional organizations, including the American Academy of Pediatrics, introduced 1 such framework in 2007. SUTTP sought to enhance care transitions by outlining principles of discharge transitional care including: (1) enhanced accountability, (2) creation of a central coordination hub charged with communicating expectations for care, (3) clear and direct communication of treatment plans and follow‐up, (4) timely feedback/feed‐forward of relevant information, and (5) involvement of family member at every stage.[38] In the context of the SUTTP framework, we present 3 hypotheses based on our findings to guide future work.

Persistent Literature Gaps

Follow‐up with a primary care provider after discharge is another intervention that might decrease postdischarge utilization. We did not identify any studies that specifically examined primary care follow‐up. However, 2 studies[43, 44] that did not meet our inclusion criteria (because they included adults and did not stratify by age group in the analysis) examined any outpatient follow‐up after discharge using state‐specific Medicaid claims. One study found that outpatient follow‐up after sickle cell hospitalization was associated with lower rates of readmission.[43] The other found no difference in readmission across multiple conditions.[44] One recent review of outpatient follow‐up from the ED for asthma found that even when increases in follow‐up were achieved, no reduction in the subsequent utilization was observed.[45]

Additional important questions remain underexplored. First, are condition‐specific interventions superior to those that span conditions? All of the interventions that demonstrated reductions in readmission were condition‐specific, yet no generic interventions met our inclusion criteria. Importantly, only 1 study[29] in our review examined discharge processes from 1 of the pediatric conditions with the most variation[8] in readmission. Further, no studies focused on children with complex medical conditions, who are known to be at increased risk of readmission,[46] indicating a sizable knowledge gap persists in understanding how to prevent readmissions in the most vulnerable pediatric populations.

Lastly, who are the most appropriate effector individuals for discharge‐focused transitional care interventions? Demographically matched effector individuals have shown promise in improving care using community health workers.[47, 48] The degree to which the identity of the intervener mediates subsequent ED and hospital utilization warrants further investigation.

Limitations of This Systematic Review

The studies included in this review assessed different outcomes at different intervals, precluding meta‐analysis. With greater consistency in the collection of data on the quality of discharge processes and their subsequent outcomes, future studies may offer further clarity as to which discharge‐oriented practices are more effective than others. Because we only identified literature in 3 pediatric conditions, generalizability beyond these conditions may be limited. The settings of the interventions also occurred in multiple countries; we excluded countries from low or low‐middle incomes to facilitate generalizability. As many of the discharge processes contained multiple interventions, it is not possible to ascertain which, if any, singular action may decrease posthospitalization utilization. Additionally, some of the included interventions are older, and it is plausible that discharge processes have evolved with the expansion of the hospitalist model.

Methods of data collection influence the quality of results in the included studies. Most of the studies included in this review used either medical record review or parental self‐report of utilization. Parental report may be sufficient for hospitalizations and ED utilization; however, it is subject to recall bias. Chart review likely underestimates the number of postdischarge events, depending on the individual institution's proportion of the market and the tendency of individuals to seek care at multiple institutions. Claims data may offer the most accurate assessments of ED and hospital utilization and cost, but can be more difficult to obtain and do not provide the same potential for granularity as parent report or medical records review.

Finally, subsequent ED visits, readmissions, and cost may not be the best measures of the quality of discharge transitional care. A number of tools have been developed to more specifically evaluate the quality of transitional care in adults,[49, 50] including a validated instrument that consists of only 3 items,[50] which primarily assesses the extent to which patients are prepared for self‐care upon discharge. For pediatric populations, validated tools assessing caregiver experience with discharge[51] and discharge readiness[52] are also available. These instruments may assist those interested in assessing process‐related outcomes that specifically assess discharge transitional care elements and may mediate subsequent ED visits or hospitalizations.

Search Strategy

With the assistance of a research librarian, we searched MEDLINE and CINAHL (Cumulative Index to Nursing and Allied Health Literature) from the inception of these databases through to March 28, 2012 (for search strategies, see the Supporting Information, Appendix, Part 1, in the online version of this article).

Study Selection

Two authors (K.A. and C.K.) independently reviewed abstracts identified by the initial search, as well as abstracts of references of included articles. Eligibility criteria for inclusion in full review included: (1) discharge‐oriented process or intervention initiated in the inpatient setting, (2) study outcomes related to subsequent utilization including hospital readmission or emergency department visit after hospitalization, (3) child‐ or adolescent‐focused or child‐specific results presented separately, and (4) written or available in English. If abstract review did not sufficiently clarify whether all eligibility criteria were met, the article was included in the full review. Two authors (K.A. and C.K.) independently reviewed articles meeting criteria for full review to determine eligibility. Disagreements regarding inclusion in the final analysis were discussed with all 4 authors. We excluded studies in countries with low or lower‐middle incomes,[15] as discharge interventions in these countries may not be broadly applicable.

Data Abstraction, Quality Assessment, and Data Synthesis

Two authors (K.A. and C.K.) independently abstracted data using a modified Cochrane Collaboration data collection form.[16] We independently scored the included studies using the Downs and Black checklist, which assesses the risk of bias and the quality of both randomized and nonrandomized studies.[17] This checklist yields a composite score of 0 to 28 points, excluding the item assessing power. As many studies either lacked power calculations or included power calculations based on outcomes not included in our review, we performed calculations to determine the sample size needed to detect a decrease in readmission or ED utilization by 20% from baseline or control rates. Due to the heterogeneous nature of included studies in terms of population, interventions, study design, and outcomes, meta‐analysis was not performed.

Patient Populations and Intervention Timing and Components

Studies varied regarding the specific medical conditions they evaluated. Eight of the papers reported discharge interventions for children with asthma, 5 papers focused on discharge from the neonatal intensive care unit (NICU), and a final study discussed a discharge intervention for children with cancer (Table 1). Although our primary goal was to synthesize discharge interventions across pediatric conditions, we provide a summary of discharge interventions by condition (see Supporting Information, Appendix, Part 3, in the online version of this article).

Studies varied regarding the timing and nature of the intervention components. Eight discharge interventions included a major inpatient component, in addition to outpatient support or follow‐up.[21, 23, 24, 25, 26, 29, 32, 34] Two studies included an inpatient education component only.[22, 27] The remainder were initiated during index hospitalization but focused primarily on home visitation, enhanced follow‐up, and support after discharge (Figure 2).[28, 30, 31, 33]

Figure 2

Outcome Assessment Methods

Readmission and subsequent ED utilization events were identified using multiple techniques. Some authors accessed claims records to capture all outcomes.[30, 33] Others relied on chart review.[21, 25, 26, 27, 28, 31, 32] One study supplemented hospital records with outpatient records.[24] Some investigators used parental reports.[22, 23, 31] Two studies did not describe methods for identifying postdischarge events.[29, 34]

Study Quality

The quality of the included studies varied (Table 2). Many of the studies had inadequate sample size to detect a difference in either readmission or ED visit subsequent to discharge. Eight studies found differences in either subsequent ED utilization, hospitalization, or both and were considered adequately powered for these specific outcomes.[21, 23, 25, 26, 28, 30, 31, 32] In contrast, among studies with readmission as an outcome, 6 were not adequately powered to detect a difference in this particular outcome.[24, 30, 31, 32, 33, 34] In these 6 studies, all except 1 study³⁰ had <10% of the sample size required to detect differences in readmission. Further, 2 studies that examined ED utilization were underpowered to detect differences between intervention and control groups.[24, 26] We were unable to perform power calculations for 3 studies,[22, 27, 29] as the authors presented the number of events without clear denominators.

Excluding the assessment of statistical power, Downs and Black scores ranged from 10 to 23 (maximum 28 possible points) indicating varying quality. As would be expected with discharge interventions, studies did not blind participants; 2 studies did, however, appropriately blind the outcome evaluators to intervention assignment.[22, 30] Even though 10 out of the 14 studies were randomized controlled trials, randomization may not have been completely effective due to sample size being too small for effective randomization,[31] large numbers of excluded subjects after randomization,[30] and unclear randomization process.[34] Several studies had varying follow‐up periods for patients within a given study. For example, 3 NICU studies assessed readmission at 1‐year corrected age,[30, 31, 33] creating the analytic difficulty that the amount of time a given patient was at risk for readmission was dependent on when the patient was discharged, yet this was not accounted for in the analyses. Only 2 studies demonstrated low rates of loss to follow‐up (<10%).[30, 33] The remainder of the studies either had high incompletion/loss to follow‐up rates (>10%)[22, 24, 31] or did not report rates.[21, 23, 25, 26, 27, 28, 29, 32, 34] Finally, 3 studies recruited patients from multiple sites,[24, 31, 33] and none adjusted for potential differences in effect based on enrollment site.

Findings Across Patient Populations Regarding Readmission

Of the 4 studies that demonstrated change in overall readmission,[23, 25, 26, 28] all were asthma focused; 3 demonstrated a decrease in readmissions,[23, 25, 26] and 1 an increase in readmissions.[28] The 3 effective interventions included 1‐on‐1 inpatient education delivered by an asthma nurse, in addition to postdischarge follow‐up support, either by telephone or clinic visit. Two of these interventions provided rescue oral steroids to some patients on discharge.[25, 26] In contrast, a study from New Zealand evaluated a series of postdischarge visits using an existing public health nurse infrastructure and demonstrated an increase in readmission between 6 to 18 months after admission in European children.[28] An additional study focused on outpatient support after discharge from the NICU, and demonstrated a lower frequency of readmission to the intensive care unit without overall reduction of hospital readmission (Tables 1 and 2).[30]

Findings Across Patient Populations Regarding Subsequent ED Visits

Of all the discharge interventions, 6 demonstrated differences in return to the ED after discharge. Five studies described a decrease in ED visits after hospitalization,[23, 25, 30, 31, 32] and 1 showed an increase.[21] Three studies in the NICU population demonstrated decreased ED utilization through a combination of augmented family engagement during hospitalization and/or enhanced support after discharge. Two inpatient asthma education interventions with structured postdischarge follow‐up decreased return visitation to the ED.[23, 26] The intervention that worsened subsequent ED utilization (ie, increased ED visit hazard compared to matched controls) provided enhanced inpatient education to a nonrandom group of children hospitalized with asthma and provided a follow‐up phone call 3 weeks after discharge (Tables 1 and 2).[21]

Current Conceptual Framework

There are a number of existing discharge transitional care frameworks from prior studies[35, 36] and professional societies.[37] The Stepping Up to the Plate (SUTTP) alliance, a collaborative of 9 professional organizations, including the American Academy of Pediatrics, introduced 1 such framework in 2007. SUTTP sought to enhance care transitions by outlining principles of discharge transitional care including: (1) enhanced accountability, (2) creation of a central coordination hub charged with communicating expectations for care, (3) clear and direct communication of treatment plans and follow‐up, (4) timely feedback/feed‐forward of relevant information, and (5) involvement of family member at every stage.[38] In the context of the SUTTP framework, we present 3 hypotheses based on our findings to guide future work.

Persistent Literature Gaps

Follow‐up with a primary care provider after discharge is another intervention that might decrease postdischarge utilization. We did not identify any studies that specifically examined primary care follow‐up. However, 2 studies[43, 44] that did not meet our inclusion criteria (because they included adults and did not stratify by age group in the analysis) examined any outpatient follow‐up after discharge using state‐specific Medicaid claims. One study found that outpatient follow‐up after sickle cell hospitalization was associated with lower rates of readmission.[43] The other found no difference in readmission across multiple conditions.[44] One recent review of outpatient follow‐up from the ED for asthma found that even when increases in follow‐up were achieved, no reduction in the subsequent utilization was observed.[45]

Additional important questions remain underexplored. First, are condition‐specific interventions superior to those that span conditions? All of the interventions that demonstrated reductions in readmission were condition‐specific, yet no generic interventions met our inclusion criteria. Importantly, only 1 study[29] in our review examined discharge processes from 1 of the pediatric conditions with the most variation[8] in readmission. Further, no studies focused on children with complex medical conditions, who are known to be at increased risk of readmission,[46] indicating a sizable knowledge gap persists in understanding how to prevent readmissions in the most vulnerable pediatric populations.

Lastly, who are the most appropriate effector individuals for discharge‐focused transitional care interventions? Demographically matched effector individuals have shown promise in improving care using community health workers.[47, 48] The degree to which the identity of the intervener mediates subsequent ED and hospital utilization warrants further investigation.

Limitations of This Systematic Review

The studies included in this review assessed different outcomes at different intervals, precluding meta‐analysis. With greater consistency in the collection of data on the quality of discharge processes and their subsequent outcomes, future studies may offer further clarity as to which discharge‐oriented practices are more effective than others. Because we only identified literature in 3 pediatric conditions, generalizability beyond these conditions may be limited. The settings of the interventions also occurred in multiple countries; we excluded countries from low or low‐middle incomes to facilitate generalizability. As many of the discharge processes contained multiple interventions, it is not possible to ascertain which, if any, singular action may decrease posthospitalization utilization. Additionally, some of the included interventions are older, and it is plausible that discharge processes have evolved with the expansion of the hospitalist model.

Methods of data collection influence the quality of results in the included studies. Most of the studies included in this review used either medical record review or parental self‐report of utilization. Parental report may be sufficient for hospitalizations and ED utilization; however, it is subject to recall bias. Chart review likely underestimates the number of postdischarge events, depending on the individual institution's proportion of the market and the tendency of individuals to seek care at multiple institutions. Claims data may offer the most accurate assessments of ED and hospital utilization and cost, but can be more difficult to obtain and do not provide the same potential for granularity as parent report or medical records review.

Finally, subsequent ED visits, readmissions, and cost may not be the best measures of the quality of discharge transitional care. A number of tools have been developed to more specifically evaluate the quality of transitional care in adults,[49, 50] including a validated instrument that consists of only 3 items,[50] which primarily assesses the extent to which patients are prepared for self‐care upon discharge. For pediatric populations, validated tools assessing caregiver experience with discharge[51] and discharge readiness[52] are also available. These instruments may assist those interested in assessing process‐related outcomes that specifically assess discharge transitional care elements and may mediate subsequent ED visits or hospitalizations.

Search Strategy

With the assistance of a research librarian, we searched MEDLINE and CINAHL (Cumulative Index to Nursing and Allied Health Literature) from the inception of these databases through to March 28, 2012 (for search strategies, see the Supporting Information, Appendix, Part 1, in the online version of this article).

Study Selection

Two authors (K.A. and C.K.) independently reviewed abstracts identified by the initial search, as well as abstracts of references of included articles. Eligibility criteria for inclusion in full review included: (1) discharge‐oriented process or intervention initiated in the inpatient setting, (2) study outcomes related to subsequent utilization including hospital readmission or emergency department visit after hospitalization, (3) child‐ or adolescent‐focused or child‐specific results presented separately, and (4) written or available in English. If abstract review did not sufficiently clarify whether all eligibility criteria were met, the article was included in the full review. Two authors (K.A. and C.K.) independently reviewed articles meeting criteria for full review to determine eligibility. Disagreements regarding inclusion in the final analysis were discussed with all 4 authors. We excluded studies in countries with low or lower‐middle incomes,[15] as discharge interventions in these countries may not be broadly applicable.

Data Abstraction, Quality Assessment, and Data Synthesis

Two authors (K.A. and C.K.) independently abstracted data using a modified Cochrane Collaboration data collection form.[16] We independently scored the included studies using the Downs and Black checklist, which assesses the risk of bias and the quality of both randomized and nonrandomized studies.[17] This checklist yields a composite score of 0 to 28 points, excluding the item assessing power. As many studies either lacked power calculations or included power calculations based on outcomes not included in our review, we performed calculations to determine the sample size needed to detect a decrease in readmission or ED utilization by 20% from baseline or control rates. Due to the heterogeneous nature of included studies in terms of population, interventions, study design, and outcomes, meta‐analysis was not performed.

Patient Populations and Intervention Timing and Components

Studies varied regarding the specific medical conditions they evaluated. Eight of the papers reported discharge interventions for children with asthma, 5 papers focused on discharge from the neonatal intensive care unit (NICU), and a final study discussed a discharge intervention for children with cancer (Table 1). Although our primary goal was to synthesize discharge interventions across pediatric conditions, we provide a summary of discharge interventions by condition (see Supporting Information, Appendix, Part 3, in the online version of this article).

Studies varied regarding the timing and nature of the intervention components. Eight discharge interventions included a major inpatient component, in addition to outpatient support or follow‐up.[21, 23, 24, 25, 26, 29, 32, 34] Two studies included an inpatient education component only.[22, 27] The remainder were initiated during index hospitalization but focused primarily on home visitation, enhanced follow‐up, and support after discharge (Figure 2).[28, 30, 31, 33]

Figure 2

Outcome Assessment Methods

Readmission and subsequent ED utilization events were identified using multiple techniques. Some authors accessed claims records to capture all outcomes.[30, 33] Others relied on chart review.[21, 25, 26, 27, 28, 31, 32] One study supplemented hospital records with outpatient records.[24] Some investigators used parental reports.[22, 23, 31] Two studies did not describe methods for identifying postdischarge events.[29, 34]

Study Quality

The quality of the included studies varied (Table 2). Many of the studies had inadequate sample size to detect a difference in either readmission or ED visit subsequent to discharge. Eight studies found differences in either subsequent ED utilization, hospitalization, or both and were considered adequately powered for these specific outcomes.[21, 23, 25, 26, 28, 30, 31, 32] In contrast, among studies with readmission as an outcome, 6 were not adequately powered to detect a difference in this particular outcome.[24, 30, 31, 32, 33, 34] In these 6 studies, all except 1 study³⁰ had <10% of the sample size required to detect differences in readmission. Further, 2 studies that examined ED utilization were underpowered to detect differences between intervention and control groups.[24, 26] We were unable to perform power calculations for 3 studies,[22, 27, 29] as the authors presented the number of events without clear denominators.

Excluding the assessment of statistical power, Downs and Black scores ranged from 10 to 23 (maximum 28 possible points) indicating varying quality. As would be expected with discharge interventions, studies did not blind participants; 2 studies did, however, appropriately blind the outcome evaluators to intervention assignment.[22, 30] Even though 10 out of the 14 studies were randomized controlled trials, randomization may not have been completely effective due to sample size being too small for effective randomization,[31] large numbers of excluded subjects after randomization,[30] and unclear randomization process.[34] Several studies had varying follow‐up periods for patients within a given study. For example, 3 NICU studies assessed readmission at 1‐year corrected age,[30, 31, 33] creating the analytic difficulty that the amount of time a given patient was at risk for readmission was dependent on when the patient was discharged, yet this was not accounted for in the analyses. Only 2 studies demonstrated low rates of loss to follow‐up (<10%).[30, 33] The remainder of the studies either had high incompletion/loss to follow‐up rates (>10%)[22, 24, 31] or did not report rates.[21, 23, 25, 26, 27, 28, 29, 32, 34] Finally, 3 studies recruited patients from multiple sites,[24, 31, 33] and none adjusted for potential differences in effect based on enrollment site.

Findings Across Patient Populations Regarding Readmission

Of the 4 studies that demonstrated change in overall readmission,[23, 25, 26, 28] all were asthma focused; 3 demonstrated a decrease in readmissions,[23, 25, 26] and 1 an increase in readmissions.[28] The 3 effective interventions included 1‐on‐1 inpatient education delivered by an asthma nurse, in addition to postdischarge follow‐up support, either by telephone or clinic visit. Two of these interventions provided rescue oral steroids to some patients on discharge.[25, 26] In contrast, a study from New Zealand evaluated a series of postdischarge visits using an existing public health nurse infrastructure and demonstrated an increase in readmission between 6 to 18 months after admission in European children.[28] An additional study focused on outpatient support after discharge from the NICU, and demonstrated a lower frequency of readmission to the intensive care unit without overall reduction of hospital readmission (Tables 1 and 2).[30]

Findings Across Patient Populations Regarding Subsequent ED Visits

Of all the discharge interventions, 6 demonstrated differences in return to the ED after discharge. Five studies described a decrease in ED visits after hospitalization,[23, 25, 30, 31, 32] and 1 showed an increase.[21] Three studies in the NICU population demonstrated decreased ED utilization through a combination of augmented family engagement during hospitalization and/or enhanced support after discharge. Two inpatient asthma education interventions with structured postdischarge follow‐up decreased return visitation to the ED.[23, 26] The intervention that worsened subsequent ED utilization (ie, increased ED visit hazard compared to matched controls) provided enhanced inpatient education to a nonrandom group of children hospitalized with asthma and provided a follow‐up phone call 3 weeks after discharge (Tables 1 and 2).[21]

Current Conceptual Framework

There are a number of existing discharge transitional care frameworks from prior studies[35, 36] and professional societies.[37] The Stepping Up to the Plate (SUTTP) alliance, a collaborative of 9 professional organizations, including the American Academy of Pediatrics, introduced 1 such framework in 2007. SUTTP sought to enhance care transitions by outlining principles of discharge transitional care including: (1) enhanced accountability, (2) creation of a central coordination hub charged with communicating expectations for care, (3) clear and direct communication of treatment plans and follow‐up, (4) timely feedback/feed‐forward of relevant information, and (5) involvement of family member at every stage.[38] In the context of the SUTTP framework, we present 3 hypotheses based on our findings to guide future work.

Persistent Literature Gaps

Follow‐up with a primary care provider after discharge is another intervention that might decrease postdischarge utilization. We did not identify any studies that specifically examined primary care follow‐up. However, 2 studies[43, 44] that did not meet our inclusion criteria (because they included adults and did not stratify by age group in the analysis) examined any outpatient follow‐up after discharge using state‐specific Medicaid claims. One study found that outpatient follow‐up after sickle cell hospitalization was associated with lower rates of readmission.[43] The other found no difference in readmission across multiple conditions.[44] One recent review of outpatient follow‐up from the ED for asthma found that even when increases in follow‐up were achieved, no reduction in the subsequent utilization was observed.[45]

Additional important questions remain underexplored. First, are condition‐specific interventions superior to those that span conditions? All of the interventions that demonstrated reductions in readmission were condition‐specific, yet no generic interventions met our inclusion criteria. Importantly, only 1 study[29] in our review examined discharge processes from 1 of the pediatric conditions with the most variation[8] in readmission. Further, no studies focused on children with complex medical conditions, who are known to be at increased risk of readmission,[46] indicating a sizable knowledge gap persists in understanding how to prevent readmissions in the most vulnerable pediatric populations.

Lastly, who are the most appropriate effector individuals for discharge‐focused transitional care interventions? Demographically matched effector individuals have shown promise in improving care using community health workers.[47, 48] The degree to which the identity of the intervener mediates subsequent ED and hospital utilization warrants further investigation.

Limitations of This Systematic Review

The studies included in this review assessed different outcomes at different intervals, precluding meta‐analysis. With greater consistency in the collection of data on the quality of discharge processes and their subsequent outcomes, future studies may offer further clarity as to which discharge‐oriented practices are more effective than others. Because we only identified literature in 3 pediatric conditions, generalizability beyond these conditions may be limited. The settings of the interventions also occurred in multiple countries; we excluded countries from low or low‐middle incomes to facilitate generalizability. As many of the discharge processes contained multiple interventions, it is not possible to ascertain which, if any, singular action may decrease posthospitalization utilization. Additionally, some of the included interventions are older, and it is plausible that discharge processes have evolved with the expansion of the hospitalist model.

Methods of data collection influence the quality of results in the included studies. Most of the studies included in this review used either medical record review or parental self‐report of utilization. Parental report may be sufficient for hospitalizations and ED utilization; however, it is subject to recall bias. Chart review likely underestimates the number of postdischarge events, depending on the individual institution's proportion of the market and the tendency of individuals to seek care at multiple institutions. Claims data may offer the most accurate assessments of ED and hospital utilization and cost, but can be more difficult to obtain and do not provide the same potential for granularity as parent report or medical records review.

Finally, subsequent ED visits, readmissions, and cost may not be the best measures of the quality of discharge transitional care. A number of tools have been developed to more specifically evaluate the quality of transitional care in adults,[49, 50] including a validated instrument that consists of only 3 items,[50] which primarily assesses the extent to which patients are prepared for self‐care upon discharge. For pediatric populations, validated tools assessing caregiver experience with discharge[51] and discharge readiness[52] are also available. These instruments may assist those interested in assessing process‐related outcomes that specifically assess discharge transitional care elements and may mediate subsequent ED visits or hospitalizations.

Hospital Quality and Safety at Risk?

The effect of the recession on the quality of hospital care is less clear. Until the 1990s, hospital quality was essentially assumed and virtually unmeasured. Even now, measuring hospital quality is difficult and rarely timely. Medicare data often take 1 to 2 years to become publicly available for analysis. Reports by trade organizations like the American Hospital Association are up‐to‐date but have conflicts of interest and are less rigorous. The most timely measures of hospitals' distressflawed as they may bewill come from the hospitals themselves, just like reports of economic woe from other businesses and government agencies during challenging economic times.

However, since the publication of the 1999 report To Err is Human,12 major improvements in hospital quality and safety have transformed the delivery of inpatient care. These improvements have taken the form of simple interventions like nationally consistent medical abbreviations, management initiatives like Six Sigma, and technological advances including computerized health records.

Nonetheless, during this recession and recovery, slashed hospital budgets may slow or even stop the momentum towards further improvements in quality and safety. Frontline care delivery could be at risk. Understaffed and under financed hospitals are rarely safe. Dissatisfaction and layoffs hurt the interactions between employees and patients. Robust nurse‐to‐patient ratios which have proven vital to patients' hospital outcomes could be at risk.13 Admittedly, recession‐induced threats to quality and safety are conjectures on our part: unfortunately, no recession measures of hospitals' specific spending on staffing, technology, or process improvements exist.

However, there are many small, evidence‐based changes that could improve hospital safety dramatically in the near future. Michigan's Keystone ICU Initiative showed that systematic interventions in routine care delivery could reduce the risk of catheter‐related bloodstream infections, which currently are implicated in the death of 28,000 Americans per year, to nearly zero.14 The Institute for Healthcare Improvement's 100,000 Lives Campaign also illustrated that dramatic improvements in hospital‐related mortality can occur with fairly focused interventions. In the month after discharge, more than one‐quarter of all hospitalized patients go to an emergency room or need to be rehospitalized. This rate can be cut by 30% by inserting a nurse discharge advocate into the discharge process.15 Instituting a simple safety checklist before surgery decreased surgery‐related mortality and complications by over one‐third.16

Such interventions are effective, reasonable, and widely accessible. Over the long‐term, many may even be cost‐saving. But, importantly, they all require an institutional investment in start‐up money and an organizational will to change how things have been done. In a period of recession with severe cost‐cutting, and a recovery period of cautious spending, this may not be possible.

Hospital Quality and Safety at Risk?

The effect of the recession on the quality of hospital care is less clear. Until the 1990s, hospital quality was essentially assumed and virtually unmeasured. Even now, measuring hospital quality is difficult and rarely timely. Medicare data often take 1 to 2 years to become publicly available for analysis. Reports by trade organizations like the American Hospital Association are up‐to‐date but have conflicts of interest and are less rigorous. The most timely measures of hospitals' distressflawed as they may bewill come from the hospitals themselves, just like reports of economic woe from other businesses and government agencies during challenging economic times.

However, since the publication of the 1999 report To Err is Human,12 major improvements in hospital quality and safety have transformed the delivery of inpatient care. These improvements have taken the form of simple interventions like nationally consistent medical abbreviations, management initiatives like Six Sigma, and technological advances including computerized health records.

Nonetheless, during this recession and recovery, slashed hospital budgets may slow or even stop the momentum towards further improvements in quality and safety. Frontline care delivery could be at risk. Understaffed and under financed hospitals are rarely safe. Dissatisfaction and layoffs hurt the interactions between employees and patients. Robust nurse‐to‐patient ratios which have proven vital to patients' hospital outcomes could be at risk.13 Admittedly, recession‐induced threats to quality and safety are conjectures on our part: unfortunately, no recession measures of hospitals' specific spending on staffing, technology, or process improvements exist.

However, there are many small, evidence‐based changes that could improve hospital safety dramatically in the near future. Michigan's Keystone ICU Initiative showed that systematic interventions in routine care delivery could reduce the risk of catheter‐related bloodstream infections, which currently are implicated in the death of 28,000 Americans per year, to nearly zero.14 The Institute for Healthcare Improvement's 100,000 Lives Campaign also illustrated that dramatic improvements in hospital‐related mortality can occur with fairly focused interventions. In the month after discharge, more than one‐quarter of all hospitalized patients go to an emergency room or need to be rehospitalized. This rate can be cut by 30% by inserting a nurse discharge advocate into the discharge process.15 Instituting a simple safety checklist before surgery decreased surgery‐related mortality and complications by over one‐third.16

Such interventions are effective, reasonable, and widely accessible. Over the long‐term, many may even be cost‐saving. But, importantly, they all require an institutional investment in start‐up money and an organizational will to change how things have been done. In a period of recession with severe cost‐cutting, and a recovery period of cautious spending, this may not be possible.

Hospital Quality and Safety at Risk?

The effect of the recession on the quality of hospital care is less clear. Until the 1990s, hospital quality was essentially assumed and virtually unmeasured. Even now, measuring hospital quality is difficult and rarely timely. Medicare data often take 1 to 2 years to become publicly available for analysis. Reports by trade organizations like the American Hospital Association are up‐to‐date but have conflicts of interest and are less rigorous. The most timely measures of hospitals' distressflawed as they may bewill come from the hospitals themselves, just like reports of economic woe from other businesses and government agencies during challenging economic times.

However, since the publication of the 1999 report To Err is Human,12 major improvements in hospital quality and safety have transformed the delivery of inpatient care. These improvements have taken the form of simple interventions like nationally consistent medical abbreviations, management initiatives like Six Sigma, and technological advances including computerized health records.

Nonetheless, during this recession and recovery, slashed hospital budgets may slow or even stop the momentum towards further improvements in quality and safety. Frontline care delivery could be at risk. Understaffed and under financed hospitals are rarely safe. Dissatisfaction and layoffs hurt the interactions between employees and patients. Robust nurse‐to‐patient ratios which have proven vital to patients' hospital outcomes could be at risk.13 Admittedly, recession‐induced threats to quality and safety are conjectures on our part: unfortunately, no recession measures of hospitals' specific spending on staffing, technology, or process improvements exist.

However, there are many small, evidence‐based changes that could improve hospital safety dramatically in the near future. Michigan's Keystone ICU Initiative showed that systematic interventions in routine care delivery could reduce the risk of catheter‐related bloodstream infections, which currently are implicated in the death of 28,000 Americans per year, to nearly zero.14 The Institute for Healthcare Improvement's 100,000 Lives Campaign also illustrated that dramatic improvements in hospital‐related mortality can occur with fairly focused interventions. In the month after discharge, more than one‐quarter of all hospitalized patients go to an emergency room or need to be rehospitalized. This rate can be cut by 30% by inserting a nurse discharge advocate into the discharge process.15 Instituting a simple safety checklist before surgery decreased surgery‐related mortality and complications by over one‐third.16

Such interventions are effective, reasonable, and widely accessible. Over the long‐term, many may even be cost‐saving. But, importantly, they all require an institutional investment in start‐up money and an organizational will to change how things have been done. In a period of recession with severe cost‐cutting, and a recovery period of cautious spending, this may not be possible.

Literature Search

With the assistance of a health services librarian, a search of the following electronic databases from January 1, 1950 through February 5, 2009 was conducted: Medline, Web of Science, Cumulative Index to Nursing and Allied Health Literature (CINAHL), Health Care Advisory Board (HCAB), Lexis‐Nexis, National Guideline Clearinghouse, and Cochrane Reviews. Key words used for the Boolean search are included in Appendix A. In addition, we conducted a manual search of reference lists from reviews, guidelines, and articles meeting inclusion criteria.

We included English language peer‐reviewed publications that reported on pediatric OU care in the United States. Studies were included if they reported outcomes including lengths of stay, admission from observation rates, return visit rates, costs or charges. Descriptive publications of pediatric OU structure and function were also included. Studies were excluded if they were conducted outside the United States, evaluated psychiatric or intensive care, reported on observation status in an ED without an OU or observation status on a traditional inpatient ward. Two reviewers (M.M. and C.K.) identified articles for inclusion. Any disagreements between the reviewers were resolved by discussion and consensus agreement. Interrater reliability was assessed using the kappa statistic.

Quality Assessment

The quality of each study was rated using the Oxford Centre for Evidence‐based Medicine levels of evidence.25 With this system, levels of evidence range from 1a (homogeneous systematic review of randomized, controlled trials) to 5 (expert opinion without explicit critical appraisal).

Data Synthesis

Data on study design, OU characteristics, patient populations, and outcomes were extracted using a standardized form. Heterogeneity of study design, interventions, and outcomes precluded the ability to conduct meta‐analyses.

Observation Unit Characteristics

The majority of research on OUs has been conducted at large academic pediatric centers. One publication was from a community hospital.26 These studies present data on more than 22,000 children cared for in OUs of 11 hospitals over a 32‐year time span. Most studies were level 2 evidence: 2b, retrospective cohort studies and low‐quality randomized, controlled trials; or 2c, outcomes research. Three were descriptive and not assigned a formal evidence level.2729

Table 1 highlights general features of U.S. pediatric OUs. Five institutions renovated or expanded clinical space in order to open the OU.27, 2932 Units ranged in size from 3 to 23 beds. The OU was located in or near the ED in all but 2 hospitals, which had ward‐based units. The ED was the primary entry point into the OU with only 2 open model units accepting patients from other settings.5, 32 The annual number of observation cases ranged from 1000 to 3000 in children's hospitals. Approximately 500 ward‐based observation cases per year were cared for in the single community hospital studied. Three reports included time trends showing increased OU utilization over study years.5, 30, 31

Staffing and Workflow

Staffing models varied and have undergone transitions over time. Prior to 1997, general pediatricians primarily provided physician services. In more recent years, OUs have utilized pediatric emergency medicine (PEM) providers. Three of the 11 units allowed for direct patient care by subspecialists.5, 6, 32 One OU was staffed by nurse practitioners.29 OU nursing backgrounds included pediatrics, emergency medicine, or PEM.

Five institutions assembled multidisciplinary teams to define the unit's role and establish policies and procedures.7, 27, 2931 Workflow in the OU focused on optimizing efficiency through standardized orders, condition‐specific treatment protocols, and bedside charting.7, 26, 33 Several units emphasized the importance of ongoing evaluations by attending physicians who could immediately respond to patient needs. Rounds occurred as often as every 4 hours.5, 7 Two centers utilized combined physician‐nursing rounds to enhance provider communication.7, 34 No publications reported on patient transitions between sites of care or at shift changes.

Criteria for Observation

All 11 hospitals have developed protocols to guide OU admissions (Table 2). Nine publications from 4 OUs commented on treatments delivered prior to observation.33, 3542 The most commonly cited criteria for admission was approval by the unit's supervising physician. Utilization review was not mentioned as an element in the OU admission decision. Common OU exclusions were the need for intensive care or monitoring while awaiting an inpatient bed; however, these were not universal. Eight centers placed bounds around the duration of OU stays, with minimum stays of 2 hours and maximum stays of 8 to 24 hours.

Ages of Children Under Observation

Seven of 11 hospitals reported the age range of patients accepted in their OU (Table 2). All but 1 unit accepted children from infants to young adults, 18 to 21 years of age.43 In the 6 units that reported the age distribution of their OU population, roughly 20% were <1 year, more than 50% were <5 years, and fewer than 30% fell into an adolescent age range.5, 6, 26, 32, 34, 43

Conditions Under Observation

Many conditions under observation were common across time and location (Table 3). The list of conditions cared for in OUs has expanded in recent years. Medical conditions predominated over surgical. While the majority of observation cases required acute care, nearly one‐half of the units accepted children with scheduled care needs (eg, routine postoperative care, procedures requiring sedation, infusions, and extended evaluations such as electroencephalograms or pH probes). These scheduled cases, cared for within the OU structure, provided more steady demand for OU services.

Reimbursement

One publication highlighted the special billing rules that must be considered for observation care.27 In 3 studies, payers recognized cost‐savings associated with the OU's ability to provide outpatient management for cases that would traditionally require inpatient care.31, 35, 38

Observation Unit Outcomes

Outcomes reported for pediatric OU stays fall into 4 major categories: length of stay (LOS), admission rates, return visit rates, and costs. Despite these seemingly straightforward groupings, there was significant heterogeneity in reporting these outcomes.

Observation Unit Operations

The OU care model expands outpatient management of acute conditions to include children who are neither ready for discharge nor clear candidates for inpatient admission. OUs have demonstrated the ability to care for patients across the pediatric age spectrum. Over the decades spanning these publications, advances in medical therapy such as antiemetics for gastroenteritis and early administration of systemic steroids for asthma may have resulted in lower admission rates or shorter time to recovery.44, 45 Despite these advances, there are marked consistencies in the conditions cared for within OUs over time. The data summarized here may help guide institutions as they consider specific pediatric conditions amenable to observation care.

The hospitals included in this review either added physical space or revised services within existing structures to establish their OU. Hospitals facing physical constraints may look to underutilized areas, such as recovery rooms, to provide observation care, as observation does not require the use of licensed inpatient beds. Several units have responded to daily fluctuations in unscheduled observation cases by also serving patients who require outpatient procedures, brief therapeutic interventions, and diagnostic testing. By caring for patients with these scheduled care needs during the day, there is a more steady flow of patients into the OU. While hospitals traditionally have used postanesthesia care units and treatment rooms for scheduled cases, OUs appear to benefit from the consistent resource allocation associated with a constant demand for services.

To date, the vast majority of pediatric OUs in the published literature have emerged as an extension of ED services. Now, with the expansion of pediatric hospitalist services and movement toward 24/7 inpatient physician coverage, there may be increased development of ward‐based OUs and the designation of inpatient observation status. While ward‐based OUs managed by pediatric hospitalists may be well established, we were not able to identify published reports on this structure of care. A national survey of health systems should be undertaken to gather information regarding the current state of pediatric observation services.

When creating policies and procedures for OUs, input should be sought from stakeholders including hospitalists, PEM providers, primary care providers, subspecialists, mid‐level providers, nurses, and ancillary staff. As patients requiring observation level of care do not neatly fit an outpatient or inpatient designation, they present an opportunity for hospitalist and PEM physician groups to collaborate.4648 Calling on the clinical experiences of inpatient and ED providers could offer unique perspectives leading to the development of innovative observation care models.

This review focused on institutions with dedicated observation services, which in all but 1 study26 consisted of a defined geographic unit. It is possible that the practices implemented in an OU could have hospital‐wide impact. For example, 1 study reported reduction in LOS for all asthma cases after opening a ward‐based unit.7 Further, pediatric hospitalist services have been associated with shorter LOS49 and increased use of observation status beds compared with traditional ward services.50 As pediatric hospitalists expand their scope of practice to include both observation and inpatient care, clinical practice may be enhanced across these care areas. It follows that the impact of observation protocols on care in the ward setting should be independently evaluated.

The costs associated with the establishment and daily operations of an OU were not addressed in the reviewed publications. Assertions that observation provides a cost‐effective alternative to inpatient care4, 7, 23, 42 should be balanced by the possibility that OUs extend care for patients who could otherwise be discharged directly home. Studies have not evaluated the cost of OU care compared with ED care alone. Research is also needed to assess variations in testing and treatment intensity in OUs compared with the ED and inpatient alternatives. Reimbursement for observation is dependent in part upon institutional contracts with payers. A full discussion of reimbursement issues around observation services is beyond the scope of this review.

Observation Unit Outcomes

Limitations

The most important limitations to this review are the heterogeneity in interventions and reporting of outcomes, which precluded our ability to combine data or conduct meta‐analyses. We attempted to organize the outcomes data into clear and consistent groupings. However, we could not compare the performance of 1 center with another due to differences in OU structure, function, and design.

In order to focus this systematic review, we chose to include only peer reviewed publications that describe pediatric OUs within the United States. This excludes expert guidelines, which may be of value to institutions developing observation services.

Our search found only a small number of centers that utilize OUs and have published their experience. Thus, our review is likely subject to publication bias. Along this line, we identified 9 additional publications where children were cared for alongside adults within a general OU.5563 This suggests an unmeasured group of children under observation in general EDs, where more than 90% of US children receive acute care.64 These articles were excluded because we were unable to distinguish pediatric specific outcomes from the larger study population.

Finally, retrospective study design is subject to information bias. Without a comparable control group, it is difficult to understand the effects of OUs. Patients directly admitted or discharged from the ED and patients who require admission after observation all differ from patients discharged from observation in ways that should be controlled for with a randomized study design.