Kendra Schwartz, MD, MSPH

Evidence summary

VVC is a common cause of vaginitis; Candida albicans accounts for 85% to 90% of cases. Risk factors include pregnancy, diabetes mellitus, and systemic antibiotics.¹ Incidence increases with onset of sexual activity, but there is no direct evidence it is sexually transmitted.¹ About 75% of women experience 1 VVC episode during their lifetime, 40% to 45% have 2 or more, and 5% to 8% have recurrent VVC (defined as 4 or more annually).^1,2

The candidiasis/antibiotics link

A 2003 systematic review found the evidence supporting the association between antibiotics and VVC limited and contradictory.³ Most were case-control or cohort studies with small sample sizes. No RCTs compare the incidence of culture-confirmed VVC among women receiving antibiotics or placebo.

Nineteen reports of 18 original studies had sufficient data to calculate a relative risk or odds ratio for antibiotic-associated VVC. Thirteen of the 19 reports showed an increase (around twofold; range, 0.43–5) in vaginal Candida prevalence; however, 3 of the 13 reports had no mycological culture data. Six studies did not show significant association between antibiotics and vaginal yeast.³

Antibiotics are thought to increase risk of VVC by killing endogenous vaginal flora (particularly Lactobacillus), allowing microorganisms resistant to the antibiotics, like Candida, to flourish.¹ Yet there is evidence that numbers of genital Lactobacillus are similar for women with and without symptomatic VVC.⁴ Further, decreasing Lactobacillus does not increase the risk of VVC.⁵

Topical and oral antifungals—both do the job

For the treatment of uncomplicated VVC, both topical and oral antifungals are clinically and mycologically effective, with comparable clinical cure rates >80%.⁶ No difference in persistent symptoms between single and multiple doses, or different durations of multiple dose regimens have been found, but samples may have been too small to detect clinically significant effects. An RCT found less nausea, headache, and abdominal pain with intravaginal imidazoles, but more vulvar irritation and vaginal discharge than oral fluconazole.⁶

For treatment of recurrent VVC, RCTs have shown the effectiveness of oral fluconazole and itraconazole maintenance therapy taken for 6 months after an initial regimen.^7,8 Treating male sexual partners did not significantly improve resolution of the woman’s symptoms or reduce the rate of symptomatic relapse.⁹

Yogurt may not live up to its rep

Two poor-quality crossover RCTs provided insufficient evidence regarding effectiveness of a diet containing oral Lactobacillus yogurt to prevent recurrent VVC.⁹ A recent RCT of 278 women on short courses of antibiotics were randomized to oral lactobacilli or placebo and vaginal lactobacilli or placebo.¹⁰ The study was stopped early because there was no effect seen. Overall, 23% developed symptomatic vulvovaginitis.

Recommendations from others

The Infectious Diseases Society of America¹¹ recommends treating uncomplicated VVC with short-course of oral or topical antifungals; treating complicated VVC with antimycotic therapy for 7 days, either daily as topical therapy or as two 150-mg doses of fluconazole 72 hours apart; treating non-albicans species of Candida with topical boric acid (600 mg/day for 14 days) or topical flucytosine; and treating recurrent VVC with induction therapy with 2 weeks of a topical or oral azole followed by a maintenance regimen for 6 months (fluconazole once a week or itraconazole twice a week).¹¹

Evidence summary

VVC is a common cause of vaginitis; Candida albicans accounts for 85% to 90% of cases. Risk factors include pregnancy, diabetes mellitus, and systemic antibiotics.¹ Incidence increases with onset of sexual activity, but there is no direct evidence it is sexually transmitted.¹ About 75% of women experience 1 VVC episode during their lifetime, 40% to 45% have 2 or more, and 5% to 8% have recurrent VVC (defined as 4 or more annually).^1,2

The candidiasis/antibiotics link

A 2003 systematic review found the evidence supporting the association between antibiotics and VVC limited and contradictory.³ Most were case-control or cohort studies with small sample sizes. No RCTs compare the incidence of culture-confirmed VVC among women receiving antibiotics or placebo.

Nineteen reports of 18 original studies had sufficient data to calculate a relative risk or odds ratio for antibiotic-associated VVC. Thirteen of the 19 reports showed an increase (around twofold; range, 0.43–5) in vaginal Candida prevalence; however, 3 of the 13 reports had no mycological culture data. Six studies did not show significant association between antibiotics and vaginal yeast.³

Antibiotics are thought to increase risk of VVC by killing endogenous vaginal flora (particularly Lactobacillus), allowing microorganisms resistant to the antibiotics, like Candida, to flourish.¹ Yet there is evidence that numbers of genital Lactobacillus are similar for women with and without symptomatic VVC.⁴ Further, decreasing Lactobacillus does not increase the risk of VVC.⁵

Topical and oral antifungals—both do the job

For the treatment of uncomplicated VVC, both topical and oral antifungals are clinically and mycologically effective, with comparable clinical cure rates >80%.⁶ No difference in persistent symptoms between single and multiple doses, or different durations of multiple dose regimens have been found, but samples may have been too small to detect clinically significant effects. An RCT found less nausea, headache, and abdominal pain with intravaginal imidazoles, but more vulvar irritation and vaginal discharge than oral fluconazole.⁶

For treatment of recurrent VVC, RCTs have shown the effectiveness of oral fluconazole and itraconazole maintenance therapy taken for 6 months after an initial regimen.^7,8 Treating male sexual partners did not significantly improve resolution of the woman’s symptoms or reduce the rate of symptomatic relapse.⁹

Yogurt may not live up to its rep

Two poor-quality crossover RCTs provided insufficient evidence regarding effectiveness of a diet containing oral Lactobacillus yogurt to prevent recurrent VVC.⁹ A recent RCT of 278 women on short courses of antibiotics were randomized to oral lactobacilli or placebo and vaginal lactobacilli or placebo.¹⁰ The study was stopped early because there was no effect seen. Overall, 23% developed symptomatic vulvovaginitis.

Recommendations from others

The Infectious Diseases Society of America¹¹ recommends treating uncomplicated VVC with short-course of oral or topical antifungals; treating complicated VVC with antimycotic therapy for 7 days, either daily as topical therapy or as two 150-mg doses of fluconazole 72 hours apart; treating non-albicans species of Candida with topical boric acid (600 mg/day for 14 days) or topical flucytosine; and treating recurrent VVC with induction therapy with 2 weeks of a topical or oral azole followed by a maintenance regimen for 6 months (fluconazole once a week or itraconazole twice a week).¹¹

Evidence summary

VVC is a common cause of vaginitis; Candida albicans accounts for 85% to 90% of cases. Risk factors include pregnancy, diabetes mellitus, and systemic antibiotics.¹ Incidence increases with onset of sexual activity, but there is no direct evidence it is sexually transmitted.¹ About 75% of women experience 1 VVC episode during their lifetime, 40% to 45% have 2 or more, and 5% to 8% have recurrent VVC (defined as 4 or more annually).^1,2

The candidiasis/antibiotics link

A 2003 systematic review found the evidence supporting the association between antibiotics and VVC limited and contradictory.³ Most were case-control or cohort studies with small sample sizes. No RCTs compare the incidence of culture-confirmed VVC among women receiving antibiotics or placebo.

Nineteen reports of 18 original studies had sufficient data to calculate a relative risk or odds ratio for antibiotic-associated VVC. Thirteen of the 19 reports showed an increase (around twofold; range, 0.43–5) in vaginal Candida prevalence; however, 3 of the 13 reports had no mycological culture data. Six studies did not show significant association between antibiotics and vaginal yeast.³

Antibiotics are thought to increase risk of VVC by killing endogenous vaginal flora (particularly Lactobacillus), allowing microorganisms resistant to the antibiotics, like Candida, to flourish.¹ Yet there is evidence that numbers of genital Lactobacillus are similar for women with and without symptomatic VVC.⁴ Further, decreasing Lactobacillus does not increase the risk of VVC.⁵

Topical and oral antifungals—both do the job

For the treatment of uncomplicated VVC, both topical and oral antifungals are clinically and mycologically effective, with comparable clinical cure rates >80%.⁶ No difference in persistent symptoms between single and multiple doses, or different durations of multiple dose regimens have been found, but samples may have been too small to detect clinically significant effects. An RCT found less nausea, headache, and abdominal pain with intravaginal imidazoles, but more vulvar irritation and vaginal discharge than oral fluconazole.⁶

For treatment of recurrent VVC, RCTs have shown the effectiveness of oral fluconazole and itraconazole maintenance therapy taken for 6 months after an initial regimen.^7,8 Treating male sexual partners did not significantly improve resolution of the woman’s symptoms or reduce the rate of symptomatic relapse.⁹

Yogurt may not live up to its rep

Two poor-quality crossover RCTs provided insufficient evidence regarding effectiveness of a diet containing oral Lactobacillus yogurt to prevent recurrent VVC.⁹ A recent RCT of 278 women on short courses of antibiotics were randomized to oral lactobacilli or placebo and vaginal lactobacilli or placebo.¹⁰ The study was stopped early because there was no effect seen. Overall, 23% developed symptomatic vulvovaginitis.

Recommendations from others

The Infectious Diseases Society of America¹¹ recommends treating uncomplicated VVC with short-course of oral or topical antifungals; treating complicated VVC with antimycotic therapy for 7 days, either daily as topical therapy or as two 150-mg doses of fluconazole 72 hours apart; treating non-albicans species of Candida with topical boric acid (600 mg/day for 14 days) or topical flucytosine; and treating recurrent VVC with induction therapy with 2 weeks of a topical or oral azole followed by a maintenance regimen for 6 months (fluconazole once a week or itraconazole twice a week).¹¹

Prostate cancer screening in asymptomatic men remains controversial, and it is difficult to present its benefits and risks quickly in a way that is understandable to patients. Yet many expert groups agree that physicians should enter into a mutual decision-making process with patients.^1-3

This article reviews the latest information relevant to the controversy, offers “talking points” for family physicians to use when discussing screening with patients, and lists websites that patients may find helpful when making a decision about prostate cancer screening.

For this review, we searched for recent articles that are generalizable to a primary care population and of the highest evidence level available. We preferentially discuss population-based studies, studies from randomized trials of screening, and meta-analyses, rather than results that are hospital- or clinic-based. For a complete systematic review of this topic (from 2002), readers are referred to one conducted for the Agency for Healthcare Research and Quality.⁴ (See Scope of the problem).

Key components of the controversy

How effective is screening?

A good screening test does 2 things. First, it detects a disease earlier than it would be detected with no screening at all, and it does so with sufficient accuracy to avoid a large number of false-positive and false-negative results.

Second, it leads to treatment of early disease that will likely produce a more favorable health outcome than waiting to treat patients who have signs and symptoms of disease.⁹ Unfortunately it is still unclear whether screening tests for prostate cancer meet these 2 criteria.

Skewed numbers. Yes, estimates of false-positive and false-negative results are available from numerous studies of different populations. However, in most studies, only men with abnormal test results receive a biopsy. Men with normal screening test results are not biopsied. Therefore the number of false negatives (and true negatives) is unknown. Furthermore, these estimates are often based on patients from urology clinics, a group more likely to have disease, thereby increasing the positive predictive value of a given screening test.

Whether current screening methods—in particular, prostate-specific antigen (PSA)—identify prostate cancers destined to become clinically relevant is also unknown. If screening does identify such cancers, a decrease in prostate cancer mortality among men who were screened is expected. If it does not, overdiagnosis and treatment of clinically insignificant cancers will negatively impact the quality of men’s lives without extending their life spans.

Clinical variability of the disease

The natural history of prostate cancer is uncertain. If cancer is left unidentified or untreated, more men will die with prostate cancer than of prostate cancer.

Clinically, prostate cancer ranges from an asymptomatic slow-growing tumor to an aggressive cancer with painful metastases. Treatment may be unnecessary at one end of the spectrum and palliative at the other.

The goal of screening is to identify slow-growing tumors destined to extend beyond the prostate while they are confined to the prostate and amenable to treatment, thereby decreasing the risk of prostate cancer morbidity and death.

Usually low morbidity. For a 50-year-old man, the risk of being diagnosed with prostate cancer by age 80 years is 15%; however, the same man has a 1.4% chance of dying of prostate cancer over that 30-year period.⁷ This 10-fold difference shows that prostate cancer usually is not a fatal illness. Another indication of the often benign nature of the disease is the high percentage of prostate cancers identified at surgery for bladder cancer; up to 40% of specimens contain unsuspected prostate cancer (level of evidence [LOE]: 2c).^10,11

Most clinical diagnoses (80%) and prostate cancer deaths (90%) occur among men older than 65 years;⁶ the median age at diagnosis is 72 years.¹² More than 75% of men older than 85 years will have histological prostate cancer (LOE: 2c).¹³ Many men live with their disease for more than 10 years, but do not die of it (LOE: 2c).¹² Additionally, a review of several decision analyses indicates that men 75 years of age and older are not likely to benefit from screening and aggressive treatment (LOE: 2a).¹⁴

Hence the recommendation: if performing prostate cancer screening, limit to men with greater than 10 years life expectancy.¹⁵

Details of screening tests

Digital rectal examination insufficient

Digital rectal examination (DRE)—palpating the prostate gland to determine size and consistency—is one screening tool for prostate cancer, usually performed in conjunction with PSA testing.¹⁵ It is not a difficult or expensive test, but its reproducibility is only fair, even among experienced urologists (LOE: 2b).¹⁶

The sensitivity and specificity of DRE can only be estimated because men with a normal finding on DRE are not routinely biopsied in any studies (TABLE 1).^17,18 One of two pertinent meta-analyses¹⁸ included studies that followed men with a negative DRE finding for development of prostate cancer (LOE: 2a). Both meta-analyses of DRE as a screening tool found that the included studies were heterogeneous in their study populations and definition of abnormal DRE test result (LOE: 2a). The positive predictive value (PPV) of DRE was 28% in one meta-analysis and 18% in the other.

TABLE 1
Characteristics of screening tests for prostate cancer

Prostate-specific antigen: Improving its clinical usefulness

Prostate-specific antigen (PSA), first detected in serum in 1979, is a protein produced by prostate epithelial cells. It was originally used to follow men treated for prostate cancer for evidence of recurrence. In the late 1980s, it became widely used in the US to screen for prostate cancer.

An elevated PSA level is suggestive but not diagnostic of prostate cancer. Elevated levels also occur with advancing age, increased prostate size, and prostatitis, and following ejaculation. Prostate manipulation such as biopsy and surgery (but not digital examination) also elevates PSA.

Customary cutpoint. An abnormal serum PSA level is commonly regarded as 4 ng/mL or greater. At this cutpoint, most studies report sensitivity for cancer of around 70%, with more variability in specificity (TABLE 1). Again, in most studies that report sensitivity and specificity, men with a PSA level less than 4 ng/mL did not undergo prostate biopsy; therefore, the number of false negatives and true negatives are only estimates.

In nested case-control studies of longitudinal cohorts, eligible cases were defined by the length of time between an abnormal PSA result and prostate cancer diagnosis, while controls were men who were not diagnosed with prostate cancer during the same time period (however, a biopsy was not usually performed). The PPV of an abnormal PSA level is estimated at 25% to 37%. As a comparison, the PPV of a positive mammogram finding for women 50 to 59 years is 4% to 9%, and 10% to 19% for women age 60 to 69.²¹

Suggested strategies to improve PSA accuracy. Approximately 70% of men with an elevated serum PSA level do not have cancer. To decrease the number of unnecessary biopsies, experts have suggested several strategies:

• using DRE with PSA
• calculating a ratio of free PSA (unbound to protein) or complexed PSA (bound to protein) to total PSA
• measuring PSA density, which incorporates the volume of the prostate gland and is subject to observer variability in prostate volume measurement
• recording PSA velocity, which is the annual rate of change in PSA level and requires 3 or more measurements.

Revise the cutpoint? The cutpoint of 4 ng/mL has been deemed too high by some clinicians. A recent report of men enrolled in a large prostate cancer prevention trial found that among the 9459 men receiving placebo, 2950 had a PSA level of 4 ng/mL or less and had normal prostates on DRE (LOE: 1b).²² All 2950 underwent a prostate biopsy; 449 (15.2%) were positive for cancer. Nearly 30% of those had a PSA level of 2 ng/mL or less.

Approximately 25% of tumors in men referred for biopsy because of abnormal PSA or DRE findings are thought to be discovered by chance due to the biopsy procedure and the high prevalence of disease (LOE: 2c).²³ Furthermore, a recent study indicates that annual PSA fluctuation is substantial; 45% of men who initially had a value of 4 ng/mL or greater, subsequently had a normal level (LOE: 2b).²⁴ Isolated abnormal levels should be confirmed before referral for biopsy.

Identifying clinically relevant cancers

The continued controversy around PSA screening relates to the potential over-diagnosis of localized prostate cancers that are not likely to become clinically significant. The ideal screening test would predictably identify cancers likely to progress.

Screening over-detects

Several computer simulations have estimated the amount of over-detection of prostate cancer. The National Cancer Institute’s Surveillance, Epidemiology and End Results (SEER) registry data demonstrated that the proportion of prostate cancer found through PSA testing that otherwise would not have been diagnosed in the patient’s lifetime was 29% for white men and 44% for black men (LOE: 2c).²⁵

An Italian study found that the proportional excess of cancers detected by screening over those that would have been expected in the absence of screening is greater than 50% (LOE: 2c).²⁶

Finally, a model based on results from the European randomized study of prostate cancer screening estimated that a screening program with a 4-year interval from age 55 to 67 has an over-detection rate of 48% (cancers that would not have been diagnosed in the absence of screening) (LOE: 2c).²⁷ Additionally, the authors determined that such screening might advance prostate cancer diagnosis by at least 10 years. Previous estimates of diagnosis lead time have been in the 5- to 7-year range (LOE: 2b).²⁸ These findings indicate that screening intervals of 2 to 4 years are unlikely to cause a loss of sensitivity.

Spread of tumor

Spread of the tumor beyond the prostate capsule is a poor prognostic sign.²⁹ Unfortunately, this occurrence is often not known until surgery, and the result usually is an “up-staging” between clinical diagnosis and pathological diagnosis. Nevertheless, most cancers (86%) diagnosed between 1992 and 1999 were localized to the prostate.⁶ Because these cancers have not spread beyond the prostate at the time of diagnosis, they are more likely to be curable. They are also more likely to represent tumors that may grow so slowly that the host will die of something other than prostate cancer.

Gleason score

Gleason score is another predictor of cancers destined to become clinically relevant (FIGURE 1). A Gleason score of 7 or greater denotes moderate to poor cellular differentiation and indicates a greater potential for progression than lower Gleason values.²⁹ A recent long-term follow-up report on a cohort of men with localized cancer treated conservatively demonstrated that men with low-grade tumors (Gleason score 2–4) have a minimal risk of dying from prostate cancer after 20 years, while men with high-grade tumors (Gleason score of 8–10) have high probability of prostate cancer death within 10 years of diagnosis (LOE: 2b).³⁰ Ecological³¹ and clinical studies³² indicate that a substantial proportion of PSA-detected cancers are moderately differentiated. This is especially true in a first round of screening; as in the European randomized study of screening, where 36% of cancers were Gleason score 7 or higher (LOE: 1b).³³

FIGURE 1
Calculating the Gleason score

The Gleason score is based on the level of differentiation and growth pattern of prostate cancer cells. Cancer cells that closely resemble the normal prostate cells when viewed under low-power magnification are well differentiated. Cancer cells that do not retain the structure of the surrounding normal cells are poorly differentiated. Scores range from 1 to 5.

In examining histologic samples of a patient’s prostate tissue, the pathologist will identify the 2 most commonly occurring patterns (types of differentiation) among the cancer cells and assign a numerical value to each pattern. The 2 numbers are then added to yield the final Gleason score. If a single pattern dominates, the pathologist will simply double the corresponding value.

Total scores range from 2 to 10. Scores in the range of 2–4 are considered well-differentiated, 5–7 are moderately differentiated, and 8–10 are poorly differentiated. In general, the higher the score, the worse the prognosis. Men with well-differentiated tumors that are treated conservatively have minimal risk of dying from prostate cancer.

Is declining mortality a sign of screening success?

Prostate cancer mortality has been declining since the mid-1990s in numerous parts of the world; the US,^6,34 Canada,³⁵ Australia,³⁶ and the United Kingdom³⁷ have all reported a reduction in the rate of prostate cancer deaths. Advocates of PSA screening point to this trend as evidence of the effectiveness of screening. But such ecological data are difficult to interpret. For instance, although much less PSA screening is performed in the UK, mortality trends are similar to those in the US where PSA testing has been used more widely.³⁸

Aggressive screening not necessarily the reason. In the US, 2 geographic areas—Seattle, Washington and Connecticut—provided a natural experiment to compare the effect of aggressive screening on prostate cancer mortality (LOE: 2c).³⁹ Although more aggressive screening and treatment took place in the Seattle area, prostate cancer mortality rates were similar to those in Connecticut over 11 years of follow-up. Similarly, in a study in British Columbia, prostate cancer mortality from 1985 to 1999 was not associated with the intensity of PSA screening (LOE: 2c).⁴⁰

Other possible explanations. If the mortality decrease is not related to PSA screening, what could cause it? One explanation is “attribution bias.” Death certificate misattribution of cause of death from prostate cancer may partially explain the pattern of increasing, then decreasing mortality rates (LOE: 2c).⁴¹ Improvement in prostate cancer treatment, especially for advanced stage, and in particular hormone therapy, is another possible explanation for the decreasing prostate cancer mortality (LOE: 2c).^14,42

Benefits of screening

The benefit of any effective screening test is a decrease in the risk of the screened-disease mortality. The best way to demonstrate decreased risk is through a randomized controlled study of the screening test, and 2 such trials are underway for prostate cancer. In the meantime, a decision model estimates that aggressive treatment of organ-confined disease potentially adds 3 years of life for men in their fifties, 1.5 years for men in their sixties, and 0.4 years for men in their seventies (LOE: 2c).³

Others have concluded that 25 men with clinically detected prostate cancer would need to be treated with surgery to prevent 1 prostate cancer death during a 6-year period, without evidence that quality of life is improved (LOE: 2c).⁴³

Consider quality of life. With uncertainty surrounding improvement in the quantity of life as a result of prostate cancer screening, improved quality of life may be an issue for patients. Focus group research has demonstrated that some patients believe it is better to know if a cancer is present than to wonder if it will be diagnosed when it is too late for cure.⁴⁴

General quality of life has been found to be similar among men treated for prostate cancer and age-matched controls without prostate cancer; however, urinary, sexual, and bowel function vary substantially between treated and untreated men and by treatment type (LOE: 3b) (TABLE 2).^45,46 In general, men treated with radical prostatectomy and brachytherapy often report better general quality of life than men who undergo radiation treatment, despite having more urinary and sexual problems (LOE: 2b).^47,48

TABLE 2
Estimates of risk associated with specific prostate cancer treatments 12 months or more after treatment

Harms of screening

The chances of undergoing a biopsy based on an abnormal screening PSA are estimated at 15% to 40% depending on the patient’s age (FIGURE 2).³ There are adverse effects associated with transrectal biopsy of the prostate. In 2 large population-basedstudies of screening, the most frequent complications were hematuria and hematospermia (LOE: 1b, 2b) (TABLE 3), with more serious consequences such as sepsis and hospitalization occurring in fewer than 1% of patients. A study of 100 screened men with an abnormal PSA who underwent prostate biopsy found that although 69% felt moderate to severe pain with the biopsy, 80% would be willing to undergo a repeat biopsy (LOE: 1b).⁴⁹

Treatment options. If the biopsy result is positive, the most common treatment options for localized cancer—which represents over 80% of all prostate cancers diagnosed⁶—include radical prostatectomy, external beam radiation therapy, brachytherapy (internal radiation therapy) or expectant management (watchful waiting). Population-based studies have reported outcomes for these treatment options (TABLE 2). Outcomes derived from hospital-based series of other prostate cancer treatments, such as cryotherapy and 3-dimensional radiation, are available, but the estimates often reflect the experience of only a few hospitals and are not representative of other facilities. Androgen ablation is the standard treatment for metastatic prostate cancer.

Untoward effects of treatment. Approximately 60% of radical prostatectomy patients report some incontinence 12 months or more after surgery (LOE: 2b),^50,51 and about 30% of patients need to wear pads for urine leakage (LOE: 2b).^50-53 Men undergoing radiation therapy have less urinary incontinence, but about 30% complain of diarrhea and loose stools (LOE: 2b).^51,52 Both therapies are associated with a high percentage of erectile dysfunction: approximately 60% of radiation therapy patients and 75% of surgery patients report their erections are not firm enough for intercourse (LOE: 2b).^51,52

Expectant management (following the cancer with regular PSA and ultrasound testing) is sometimes difficult to “sell” to patients whose fear of cancer dictates that the only logical response is to “cut it out.”⁴⁴ A recent randomized trial indicated that radical prostatectomy lowers prostate cancer mortality, local progression, distant metastasis, and overall survival as compared with watchful waiting over a median of 8.2 years of follow-up (LOE: 1b).⁵⁴ However, these results may have little relevance to prostate cancer screening since only 5% of the cancers were screen-detected and 76% were palpable.

FIGURE 2
Yield of screening 1000 men for prostate cancer

TABLE 3
Percentage of patients with specific complication of transrectal prostate biopsy

Recommendations from expert groups

Different expert groups have conflicting recommendations. Both the American Urological Association and the American Cancer Society recommend annual PSA screening starting at age 50 for most men; younger if risk factors are present. Groups that are evidence based tend to recommend a shared decision making process with patients. The AAFP and American College of Physicians advise physicians to counsel men on the known risks and uncertain benefits of screening for prostate cancer. The US Preventive Services Task Force 2002 update concluded that evidence is insufficient to recommend for or against routine screening for prostate cancer using PSA or DRE. The National Cancer Institute cites a lack of evidence to determine a net benefit for PSA or DRE screening.

When will we know more?

Only 1 randomized controlled trial of prostate cancer screening has been completed⁵⁵: 46,193 men were randomized to either PSA and DRE or no screening from 1989 to 1996. The study had methodological problems; for instance, only 23% of the group randomized to screening was screened. The investigators in the trial have interpreted its results as demonstrating a decrease in prostate cancer deaths in the screened group compared with the unscreened group (15 vs 48.7 per 100,000 man-years).⁵⁵ Others have criticized the statistical analysis and calculated the results using an “intent to screen” analysis, finding no difference in prostate cancer deaths between the 2 groups.^3,56

Two randomized controlled trials of screening are ongoing: the National Cancer Institute’s Prostate, Lung, Colon, Ovarian (PLCO) Screening Trial⁵⁷ and the European Randomized Study of Screening for Prostate Cancer.⁵⁸ Both were started in the mid-1990s and will not have results available for a few more years. Also underway is a randomized trial of intervention (radical prostatectomy) versus expectant management, called the Prostate Cancer Intervention Versus Observation Trial (PIVOT).⁵⁹

Counseling recommendations

However, providing men with information on prostate cancer screening before they discussed it with their family physician, rather than after the visit, resulted in patients having a significantly more active role in making a screening decision, and lower levels of decisional conflict (LOE: 2b).⁶¹ Informational pamphlets are available through the AAFP and CDC websites listed in TABLE 4. Additional websites containing prostate cancer screening information are found in TABLE 4. We also provide a bullet item list of key points for discussion with patients (TABLE 5), which can be used along with the balance sheet provided here (TABLE 2).

Shared decision-making is not an easy or quick process. Yet, the majority of patients will benefit from the discussion, regardless of the final decision. Of course, there are instances when a shared decision-making process is well-documented, and still results in an undesirable outcome;⁶² however, while the evidence for screening remains controversial, patients have the right to know that those controversies exist and why they exist.

TABLE 4
Useful websites for patients to find prostate cancer screening information

TABLE 5
Talking points for patients and physicians

CORRESPONDING AUTHOR
Kendra Schwartz, MD, MSPH, 101 E. Alexandrine, Detroit, MI 48201, E-mail: [email protected]

Prostate cancer screening in asymptomatic men remains controversial, and it is difficult to present its benefits and risks quickly in a way that is understandable to patients. Yet many expert groups agree that physicians should enter into a mutual decision-making process with patients.^1-3

This article reviews the latest information relevant to the controversy, offers “talking points” for family physicians to use when discussing screening with patients, and lists websites that patients may find helpful when making a decision about prostate cancer screening.

For this review, we searched for recent articles that are generalizable to a primary care population and of the highest evidence level available. We preferentially discuss population-based studies, studies from randomized trials of screening, and meta-analyses, rather than results that are hospital- or clinic-based. For a complete systematic review of this topic (from 2002), readers are referred to one conducted for the Agency for Healthcare Research and Quality.⁴ (See Scope of the problem).

Key components of the controversy

How effective is screening?

A good screening test does 2 things. First, it detects a disease earlier than it would be detected with no screening at all, and it does so with sufficient accuracy to avoid a large number of false-positive and false-negative results.

Second, it leads to treatment of early disease that will likely produce a more favorable health outcome than waiting to treat patients who have signs and symptoms of disease.⁹ Unfortunately it is still unclear whether screening tests for prostate cancer meet these 2 criteria.

Skewed numbers. Yes, estimates of false-positive and false-negative results are available from numerous studies of different populations. However, in most studies, only men with abnormal test results receive a biopsy. Men with normal screening test results are not biopsied. Therefore the number of false negatives (and true negatives) is unknown. Furthermore, these estimates are often based on patients from urology clinics, a group more likely to have disease, thereby increasing the positive predictive value of a given screening test.

Whether current screening methods—in particular, prostate-specific antigen (PSA)—identify prostate cancers destined to become clinically relevant is also unknown. If screening does identify such cancers, a decrease in prostate cancer mortality among men who were screened is expected. If it does not, overdiagnosis and treatment of clinically insignificant cancers will negatively impact the quality of men’s lives without extending their life spans.

Clinical variability of the disease

The natural history of prostate cancer is uncertain. If cancer is left unidentified or untreated, more men will die with prostate cancer than of prostate cancer.

Clinically, prostate cancer ranges from an asymptomatic slow-growing tumor to an aggressive cancer with painful metastases. Treatment may be unnecessary at one end of the spectrum and palliative at the other.

The goal of screening is to identify slow-growing tumors destined to extend beyond the prostate while they are confined to the prostate and amenable to treatment, thereby decreasing the risk of prostate cancer morbidity and death.

Usually low morbidity. For a 50-year-old man, the risk of being diagnosed with prostate cancer by age 80 years is 15%; however, the same man has a 1.4% chance of dying of prostate cancer over that 30-year period.⁷ This 10-fold difference shows that prostate cancer usually is not a fatal illness. Another indication of the often benign nature of the disease is the high percentage of prostate cancers identified at surgery for bladder cancer; up to 40% of specimens contain unsuspected prostate cancer (level of evidence [LOE]: 2c).^10,11

Most clinical diagnoses (80%) and prostate cancer deaths (90%) occur among men older than 65 years;⁶ the median age at diagnosis is 72 years.¹² More than 75% of men older than 85 years will have histological prostate cancer (LOE: 2c).¹³ Many men live with their disease for more than 10 years, but do not die of it (LOE: 2c).¹² Additionally, a review of several decision analyses indicates that men 75 years of age and older are not likely to benefit from screening and aggressive treatment (LOE: 2a).¹⁴

Hence the recommendation: if performing prostate cancer screening, limit to men with greater than 10 years life expectancy.¹⁵

Details of screening tests

Digital rectal examination insufficient

Digital rectal examination (DRE)—palpating the prostate gland to determine size and consistency—is one screening tool for prostate cancer, usually performed in conjunction with PSA testing.¹⁵ It is not a difficult or expensive test, but its reproducibility is only fair, even among experienced urologists (LOE: 2b).¹⁶

The sensitivity and specificity of DRE can only be estimated because men with a normal finding on DRE are not routinely biopsied in any studies (TABLE 1).^17,18 One of two pertinent meta-analyses¹⁸ included studies that followed men with a negative DRE finding for development of prostate cancer (LOE: 2a). Both meta-analyses of DRE as a screening tool found that the included studies were heterogeneous in their study populations and definition of abnormal DRE test result (LOE: 2a). The positive predictive value (PPV) of DRE was 28% in one meta-analysis and 18% in the other.

TABLE 1
Characteristics of screening tests for prostate cancer

Prostate-specific antigen: Improving its clinical usefulness

Prostate-specific antigen (PSA), first detected in serum in 1979, is a protein produced by prostate epithelial cells. It was originally used to follow men treated for prostate cancer for evidence of recurrence. In the late 1980s, it became widely used in the US to screen for prostate cancer.

An elevated PSA level is suggestive but not diagnostic of prostate cancer. Elevated levels also occur with advancing age, increased prostate size, and prostatitis, and following ejaculation. Prostate manipulation such as biopsy and surgery (but not digital examination) also elevates PSA.

Customary cutpoint. An abnormal serum PSA level is commonly regarded as 4 ng/mL or greater. At this cutpoint, most studies report sensitivity for cancer of around 70%, with more variability in specificity (TABLE 1). Again, in most studies that report sensitivity and specificity, men with a PSA level less than 4 ng/mL did not undergo prostate biopsy; therefore, the number of false negatives and true negatives are only estimates.

In nested case-control studies of longitudinal cohorts, eligible cases were defined by the length of time between an abnormal PSA result and prostate cancer diagnosis, while controls were men who were not diagnosed with prostate cancer during the same time period (however, a biopsy was not usually performed). The PPV of an abnormal PSA level is estimated at 25% to 37%. As a comparison, the PPV of a positive mammogram finding for women 50 to 59 years is 4% to 9%, and 10% to 19% for women age 60 to 69.²¹

Suggested strategies to improve PSA accuracy. Approximately 70% of men with an elevated serum PSA level do not have cancer. To decrease the number of unnecessary biopsies, experts have suggested several strategies:

• using DRE with PSA
• calculating a ratio of free PSA (unbound to protein) or complexed PSA (bound to protein) to total PSA
• measuring PSA density, which incorporates the volume of the prostate gland and is subject to observer variability in prostate volume measurement
• recording PSA velocity, which is the annual rate of change in PSA level and requires 3 or more measurements.

Revise the cutpoint? The cutpoint of 4 ng/mL has been deemed too high by some clinicians. A recent report of men enrolled in a large prostate cancer prevention trial found that among the 9459 men receiving placebo, 2950 had a PSA level of 4 ng/mL or less and had normal prostates on DRE (LOE: 1b).²² All 2950 underwent a prostate biopsy; 449 (15.2%) were positive for cancer. Nearly 30% of those had a PSA level of 2 ng/mL or less.

Approximately 25% of tumors in men referred for biopsy because of abnormal PSA or DRE findings are thought to be discovered by chance due to the biopsy procedure and the high prevalence of disease (LOE: 2c).²³ Furthermore, a recent study indicates that annual PSA fluctuation is substantial; 45% of men who initially had a value of 4 ng/mL or greater, subsequently had a normal level (LOE: 2b).²⁴ Isolated abnormal levels should be confirmed before referral for biopsy.

Identifying clinically relevant cancers

The continued controversy around PSA screening relates to the potential over-diagnosis of localized prostate cancers that are not likely to become clinically significant. The ideal screening test would predictably identify cancers likely to progress.

Screening over-detects

Several computer simulations have estimated the amount of over-detection of prostate cancer. The National Cancer Institute’s Surveillance, Epidemiology and End Results (SEER) registry data demonstrated that the proportion of prostate cancer found through PSA testing that otherwise would not have been diagnosed in the patient’s lifetime was 29% for white men and 44% for black men (LOE: 2c).²⁵

An Italian study found that the proportional excess of cancers detected by screening over those that would have been expected in the absence of screening is greater than 50% (LOE: 2c).²⁶

Finally, a model based on results from the European randomized study of prostate cancer screening estimated that a screening program with a 4-year interval from age 55 to 67 has an over-detection rate of 48% (cancers that would not have been diagnosed in the absence of screening) (LOE: 2c).²⁷ Additionally, the authors determined that such screening might advance prostate cancer diagnosis by at least 10 years. Previous estimates of diagnosis lead time have been in the 5- to 7-year range (LOE: 2b).²⁸ These findings indicate that screening intervals of 2 to 4 years are unlikely to cause a loss of sensitivity.

Spread of tumor

Spread of the tumor beyond the prostate capsule is a poor prognostic sign.²⁹ Unfortunately, this occurrence is often not known until surgery, and the result usually is an “up-staging” between clinical diagnosis and pathological diagnosis. Nevertheless, most cancers (86%) diagnosed between 1992 and 1999 were localized to the prostate.⁶ Because these cancers have not spread beyond the prostate at the time of diagnosis, they are more likely to be curable. They are also more likely to represent tumors that may grow so slowly that the host will die of something other than prostate cancer.

Gleason score

Gleason score is another predictor of cancers destined to become clinically relevant (FIGURE 1). A Gleason score of 7 or greater denotes moderate to poor cellular differentiation and indicates a greater potential for progression than lower Gleason values.²⁹ A recent long-term follow-up report on a cohort of men with localized cancer treated conservatively demonstrated that men with low-grade tumors (Gleason score 2–4) have a minimal risk of dying from prostate cancer after 20 years, while men with high-grade tumors (Gleason score of 8–10) have high probability of prostate cancer death within 10 years of diagnosis (LOE: 2b).³⁰ Ecological³¹ and clinical studies³² indicate that a substantial proportion of PSA-detected cancers are moderately differentiated. This is especially true in a first round of screening; as in the European randomized study of screening, where 36% of cancers were Gleason score 7 or higher (LOE: 1b).³³

FIGURE 1
Calculating the Gleason score

The Gleason score is based on the level of differentiation and growth pattern of prostate cancer cells. Cancer cells that closely resemble the normal prostate cells when viewed under low-power magnification are well differentiated. Cancer cells that do not retain the structure of the surrounding normal cells are poorly differentiated. Scores range from 1 to 5.

In examining histologic samples of a patient’s prostate tissue, the pathologist will identify the 2 most commonly occurring patterns (types of differentiation) among the cancer cells and assign a numerical value to each pattern. The 2 numbers are then added to yield the final Gleason score. If a single pattern dominates, the pathologist will simply double the corresponding value.

Total scores range from 2 to 10. Scores in the range of 2–4 are considered well-differentiated, 5–7 are moderately differentiated, and 8–10 are poorly differentiated. In general, the higher the score, the worse the prognosis. Men with well-differentiated tumors that are treated conservatively have minimal risk of dying from prostate cancer.

Is declining mortality a sign of screening success?

Prostate cancer mortality has been declining since the mid-1990s in numerous parts of the world; the US,^6,34 Canada,³⁵ Australia,³⁶ and the United Kingdom³⁷ have all reported a reduction in the rate of prostate cancer deaths. Advocates of PSA screening point to this trend as evidence of the effectiveness of screening. But such ecological data are difficult to interpret. For instance, although much less PSA screening is performed in the UK, mortality trends are similar to those in the US where PSA testing has been used more widely.³⁸

Aggressive screening not necessarily the reason. In the US, 2 geographic areas—Seattle, Washington and Connecticut—provided a natural experiment to compare the effect of aggressive screening on prostate cancer mortality (LOE: 2c).³⁹ Although more aggressive screening and treatment took place in the Seattle area, prostate cancer mortality rates were similar to those in Connecticut over 11 years of follow-up. Similarly, in a study in British Columbia, prostate cancer mortality from 1985 to 1999 was not associated with the intensity of PSA screening (LOE: 2c).⁴⁰

Other possible explanations. If the mortality decrease is not related to PSA screening, what could cause it? One explanation is “attribution bias.” Death certificate misattribution of cause of death from prostate cancer may partially explain the pattern of increasing, then decreasing mortality rates (LOE: 2c).⁴¹ Improvement in prostate cancer treatment, especially for advanced stage, and in particular hormone therapy, is another possible explanation for the decreasing prostate cancer mortality (LOE: 2c).^14,42

Benefits of screening

The benefit of any effective screening test is a decrease in the risk of the screened-disease mortality. The best way to demonstrate decreased risk is through a randomized controlled study of the screening test, and 2 such trials are underway for prostate cancer. In the meantime, a decision model estimates that aggressive treatment of organ-confined disease potentially adds 3 years of life for men in their fifties, 1.5 years for men in their sixties, and 0.4 years for men in their seventies (LOE: 2c).³

Others have concluded that 25 men with clinically detected prostate cancer would need to be treated with surgery to prevent 1 prostate cancer death during a 6-year period, without evidence that quality of life is improved (LOE: 2c).⁴³

Consider quality of life. With uncertainty surrounding improvement in the quantity of life as a result of prostate cancer screening, improved quality of life may be an issue for patients. Focus group research has demonstrated that some patients believe it is better to know if a cancer is present than to wonder if it will be diagnosed when it is too late for cure.⁴⁴

General quality of life has been found to be similar among men treated for prostate cancer and age-matched controls without prostate cancer; however, urinary, sexual, and bowel function vary substantially between treated and untreated men and by treatment type (LOE: 3b) (TABLE 2).^45,46 In general, men treated with radical prostatectomy and brachytherapy often report better general quality of life than men who undergo radiation treatment, despite having more urinary and sexual problems (LOE: 2b).^47,48

TABLE 2
Estimates of risk associated with specific prostate cancer treatments 12 months or more after treatment

Harms of screening

The chances of undergoing a biopsy based on an abnormal screening PSA are estimated at 15% to 40% depending on the patient’s age (FIGURE 2).³ There are adverse effects associated with transrectal biopsy of the prostate. In 2 large population-basedstudies of screening, the most frequent complications were hematuria and hematospermia (LOE: 1b, 2b) (TABLE 3), with more serious consequences such as sepsis and hospitalization occurring in fewer than 1% of patients. A study of 100 screened men with an abnormal PSA who underwent prostate biopsy found that although 69% felt moderate to severe pain with the biopsy, 80% would be willing to undergo a repeat biopsy (LOE: 1b).⁴⁹

Treatment options. If the biopsy result is positive, the most common treatment options for localized cancer—which represents over 80% of all prostate cancers diagnosed⁶—include radical prostatectomy, external beam radiation therapy, brachytherapy (internal radiation therapy) or expectant management (watchful waiting). Population-based studies have reported outcomes for these treatment options (TABLE 2). Outcomes derived from hospital-based series of other prostate cancer treatments, such as cryotherapy and 3-dimensional radiation, are available, but the estimates often reflect the experience of only a few hospitals and are not representative of other facilities. Androgen ablation is the standard treatment for metastatic prostate cancer.

Untoward effects of treatment. Approximately 60% of radical prostatectomy patients report some incontinence 12 months or more after surgery (LOE: 2b),^50,51 and about 30% of patients need to wear pads for urine leakage (LOE: 2b).^50-53 Men undergoing radiation therapy have less urinary incontinence, but about 30% complain of diarrhea and loose stools (LOE: 2b).^51,52 Both therapies are associated with a high percentage of erectile dysfunction: approximately 60% of radiation therapy patients and 75% of surgery patients report their erections are not firm enough for intercourse (LOE: 2b).^51,52

Expectant management (following the cancer with regular PSA and ultrasound testing) is sometimes difficult to “sell” to patients whose fear of cancer dictates that the only logical response is to “cut it out.”⁴⁴ A recent randomized trial indicated that radical prostatectomy lowers prostate cancer mortality, local progression, distant metastasis, and overall survival as compared with watchful waiting over a median of 8.2 years of follow-up (LOE: 1b).⁵⁴ However, these results may have little relevance to prostate cancer screening since only 5% of the cancers were screen-detected and 76% were palpable.

FIGURE 2
Yield of screening 1000 men for prostate cancer

TABLE 3
Percentage of patients with specific complication of transrectal prostate biopsy

Recommendations from expert groups

Different expert groups have conflicting recommendations. Both the American Urological Association and the American Cancer Society recommend annual PSA screening starting at age 50 for most men; younger if risk factors are present. Groups that are evidence based tend to recommend a shared decision making process with patients. The AAFP and American College of Physicians advise physicians to counsel men on the known risks and uncertain benefits of screening for prostate cancer. The US Preventive Services Task Force 2002 update concluded that evidence is insufficient to recommend for or against routine screening for prostate cancer using PSA or DRE. The National Cancer Institute cites a lack of evidence to determine a net benefit for PSA or DRE screening.

When will we know more?

Only 1 randomized controlled trial of prostate cancer screening has been completed⁵⁵: 46,193 men were randomized to either PSA and DRE or no screening from 1989 to 1996. The study had methodological problems; for instance, only 23% of the group randomized to screening was screened. The investigators in the trial have interpreted its results as demonstrating a decrease in prostate cancer deaths in the screened group compared with the unscreened group (15 vs 48.7 per 100,000 man-years).⁵⁵ Others have criticized the statistical analysis and calculated the results using an “intent to screen” analysis, finding no difference in prostate cancer deaths between the 2 groups.^3,56

Two randomized controlled trials of screening are ongoing: the National Cancer Institute’s Prostate, Lung, Colon, Ovarian (PLCO) Screening Trial⁵⁷ and the European Randomized Study of Screening for Prostate Cancer.⁵⁸ Both were started in the mid-1990s and will not have results available for a few more years. Also underway is a randomized trial of intervention (radical prostatectomy) versus expectant management, called the Prostate Cancer Intervention Versus Observation Trial (PIVOT).⁵⁹

Counseling recommendations

However, providing men with information on prostate cancer screening before they discussed it with their family physician, rather than after the visit, resulted in patients having a significantly more active role in making a screening decision, and lower levels of decisional conflict (LOE: 2b).⁶¹ Informational pamphlets are available through the AAFP and CDC websites listed in TABLE 4. Additional websites containing prostate cancer screening information are found in TABLE 4. We also provide a bullet item list of key points for discussion with patients (TABLE 5), which can be used along with the balance sheet provided here (TABLE 2).

Shared decision-making is not an easy or quick process. Yet, the majority of patients will benefit from the discussion, regardless of the final decision. Of course, there are instances when a shared decision-making process is well-documented, and still results in an undesirable outcome;⁶² however, while the evidence for screening remains controversial, patients have the right to know that those controversies exist and why they exist.

TABLE 4
Useful websites for patients to find prostate cancer screening information

TABLE 5
Talking points for patients and physicians

CORRESPONDING AUTHOR
Kendra Schwartz, MD, MSPH, 101 E. Alexandrine, Detroit, MI 48201, E-mail: [email protected]

Prostate cancer screening in asymptomatic men remains controversial, and it is difficult to present its benefits and risks quickly in a way that is understandable to patients. Yet many expert groups agree that physicians should enter into a mutual decision-making process with patients.^1-3

This article reviews the latest information relevant to the controversy, offers “talking points” for family physicians to use when discussing screening with patients, and lists websites that patients may find helpful when making a decision about prostate cancer screening.

For this review, we searched for recent articles that are generalizable to a primary care population and of the highest evidence level available. We preferentially discuss population-based studies, studies from randomized trials of screening, and meta-analyses, rather than results that are hospital- or clinic-based. For a complete systematic review of this topic (from 2002), readers are referred to one conducted for the Agency for Healthcare Research and Quality.⁴ (See Scope of the problem).

Key components of the controversy

How effective is screening?

A good screening test does 2 things. First, it detects a disease earlier than it would be detected with no screening at all, and it does so with sufficient accuracy to avoid a large number of false-positive and false-negative results.

Second, it leads to treatment of early disease that will likely produce a more favorable health outcome than waiting to treat patients who have signs and symptoms of disease.⁹ Unfortunately it is still unclear whether screening tests for prostate cancer meet these 2 criteria.

Skewed numbers. Yes, estimates of false-positive and false-negative results are available from numerous studies of different populations. However, in most studies, only men with abnormal test results receive a biopsy. Men with normal screening test results are not biopsied. Therefore the number of false negatives (and true negatives) is unknown. Furthermore, these estimates are often based on patients from urology clinics, a group more likely to have disease, thereby increasing the positive predictive value of a given screening test.

Whether current screening methods—in particular, prostate-specific antigen (PSA)—identify prostate cancers destined to become clinically relevant is also unknown. If screening does identify such cancers, a decrease in prostate cancer mortality among men who were screened is expected. If it does not, overdiagnosis and treatment of clinically insignificant cancers will negatively impact the quality of men’s lives without extending their life spans.

Clinical variability of the disease

The natural history of prostate cancer is uncertain. If cancer is left unidentified or untreated, more men will die with prostate cancer than of prostate cancer.

Clinically, prostate cancer ranges from an asymptomatic slow-growing tumor to an aggressive cancer with painful metastases. Treatment may be unnecessary at one end of the spectrum and palliative at the other.

The goal of screening is to identify slow-growing tumors destined to extend beyond the prostate while they are confined to the prostate and amenable to treatment, thereby decreasing the risk of prostate cancer morbidity and death.

Usually low morbidity. For a 50-year-old man, the risk of being diagnosed with prostate cancer by age 80 years is 15%; however, the same man has a 1.4% chance of dying of prostate cancer over that 30-year period.⁷ This 10-fold difference shows that prostate cancer usually is not a fatal illness. Another indication of the often benign nature of the disease is the high percentage of prostate cancers identified at surgery for bladder cancer; up to 40% of specimens contain unsuspected prostate cancer (level of evidence [LOE]: 2c).^10,11

Most clinical diagnoses (80%) and prostate cancer deaths (90%) occur among men older than 65 years;⁶ the median age at diagnosis is 72 years.¹² More than 75% of men older than 85 years will have histological prostate cancer (LOE: 2c).¹³ Many men live with their disease for more than 10 years, but do not die of it (LOE: 2c).¹² Additionally, a review of several decision analyses indicates that men 75 years of age and older are not likely to benefit from screening and aggressive treatment (LOE: 2a).¹⁴

Hence the recommendation: if performing prostate cancer screening, limit to men with greater than 10 years life expectancy.¹⁵

Details of screening tests

Digital rectal examination insufficient

Digital rectal examination (DRE)—palpating the prostate gland to determine size and consistency—is one screening tool for prostate cancer, usually performed in conjunction with PSA testing.¹⁵ It is not a difficult or expensive test, but its reproducibility is only fair, even among experienced urologists (LOE: 2b).¹⁶

The sensitivity and specificity of DRE can only be estimated because men with a normal finding on DRE are not routinely biopsied in any studies (TABLE 1).^17,18 One of two pertinent meta-analyses¹⁸ included studies that followed men with a negative DRE finding for development of prostate cancer (LOE: 2a). Both meta-analyses of DRE as a screening tool found that the included studies were heterogeneous in their study populations and definition of abnormal DRE test result (LOE: 2a). The positive predictive value (PPV) of DRE was 28% in one meta-analysis and 18% in the other.

TABLE 1
Characteristics of screening tests for prostate cancer

Prostate-specific antigen: Improving its clinical usefulness

Prostate-specific antigen (PSA), first detected in serum in 1979, is a protein produced by prostate epithelial cells. It was originally used to follow men treated for prostate cancer for evidence of recurrence. In the late 1980s, it became widely used in the US to screen for prostate cancer.

An elevated PSA level is suggestive but not diagnostic of prostate cancer. Elevated levels also occur with advancing age, increased prostate size, and prostatitis, and following ejaculation. Prostate manipulation such as biopsy and surgery (but not digital examination) also elevates PSA.

Customary cutpoint. An abnormal serum PSA level is commonly regarded as 4 ng/mL or greater. At this cutpoint, most studies report sensitivity for cancer of around 70%, with more variability in specificity (TABLE 1). Again, in most studies that report sensitivity and specificity, men with a PSA level less than 4 ng/mL did not undergo prostate biopsy; therefore, the number of false negatives and true negatives are only estimates.

In nested case-control studies of longitudinal cohorts, eligible cases were defined by the length of time between an abnormal PSA result and prostate cancer diagnosis, while controls were men who were not diagnosed with prostate cancer during the same time period (however, a biopsy was not usually performed). The PPV of an abnormal PSA level is estimated at 25% to 37%. As a comparison, the PPV of a positive mammogram finding for women 50 to 59 years is 4% to 9%, and 10% to 19% for women age 60 to 69.²¹

Suggested strategies to improve PSA accuracy. Approximately 70% of men with an elevated serum PSA level do not have cancer. To decrease the number of unnecessary biopsies, experts have suggested several strategies:

• using DRE with PSA
• calculating a ratio of free PSA (unbound to protein) or complexed PSA (bound to protein) to total PSA
• measuring PSA density, which incorporates the volume of the prostate gland and is subject to observer variability in prostate volume measurement
• recording PSA velocity, which is the annual rate of change in PSA level and requires 3 or more measurements.

Revise the cutpoint? The cutpoint of 4 ng/mL has been deemed too high by some clinicians. A recent report of men enrolled in a large prostate cancer prevention trial found that among the 9459 men receiving placebo, 2950 had a PSA level of 4 ng/mL or less and had normal prostates on DRE (LOE: 1b).²² All 2950 underwent a prostate biopsy; 449 (15.2%) were positive for cancer. Nearly 30% of those had a PSA level of 2 ng/mL or less.

Approximately 25% of tumors in men referred for biopsy because of abnormal PSA or DRE findings are thought to be discovered by chance due to the biopsy procedure and the high prevalence of disease (LOE: 2c).²³ Furthermore, a recent study indicates that annual PSA fluctuation is substantial; 45% of men who initially had a value of 4 ng/mL or greater, subsequently had a normal level (LOE: 2b).²⁴ Isolated abnormal levels should be confirmed before referral for biopsy.

Identifying clinically relevant cancers

The continued controversy around PSA screening relates to the potential over-diagnosis of localized prostate cancers that are not likely to become clinically significant. The ideal screening test would predictably identify cancers likely to progress.

Screening over-detects

Several computer simulations have estimated the amount of over-detection of prostate cancer. The National Cancer Institute’s Surveillance, Epidemiology and End Results (SEER) registry data demonstrated that the proportion of prostate cancer found through PSA testing that otherwise would not have been diagnosed in the patient’s lifetime was 29% for white men and 44% for black men (LOE: 2c).²⁵

An Italian study found that the proportional excess of cancers detected by screening over those that would have been expected in the absence of screening is greater than 50% (LOE: 2c).²⁶

Finally, a model based on results from the European randomized study of prostate cancer screening estimated that a screening program with a 4-year interval from age 55 to 67 has an over-detection rate of 48% (cancers that would not have been diagnosed in the absence of screening) (LOE: 2c).²⁷ Additionally, the authors determined that such screening might advance prostate cancer diagnosis by at least 10 years. Previous estimates of diagnosis lead time have been in the 5- to 7-year range (LOE: 2b).²⁸ These findings indicate that screening intervals of 2 to 4 years are unlikely to cause a loss of sensitivity.

Spread of tumor

Spread of the tumor beyond the prostate capsule is a poor prognostic sign.²⁹ Unfortunately, this occurrence is often not known until surgery, and the result usually is an “up-staging” between clinical diagnosis and pathological diagnosis. Nevertheless, most cancers (86%) diagnosed between 1992 and 1999 were localized to the prostate.⁶ Because these cancers have not spread beyond the prostate at the time of diagnosis, they are more likely to be curable. They are also more likely to represent tumors that may grow so slowly that the host will die of something other than prostate cancer.

Gleason score

Gleason score is another predictor of cancers destined to become clinically relevant (FIGURE 1). A Gleason score of 7 or greater denotes moderate to poor cellular differentiation and indicates a greater potential for progression than lower Gleason values.²⁹ A recent long-term follow-up report on a cohort of men with localized cancer treated conservatively demonstrated that men with low-grade tumors (Gleason score 2–4) have a minimal risk of dying from prostate cancer after 20 years, while men with high-grade tumors (Gleason score of 8–10) have high probability of prostate cancer death within 10 years of diagnosis (LOE: 2b).³⁰ Ecological³¹ and clinical studies³² indicate that a substantial proportion of PSA-detected cancers are moderately differentiated. This is especially true in a first round of screening; as in the European randomized study of screening, where 36% of cancers were Gleason score 7 or higher (LOE: 1b).³³

FIGURE 1
Calculating the Gleason score

The Gleason score is based on the level of differentiation and growth pattern of prostate cancer cells. Cancer cells that closely resemble the normal prostate cells when viewed under low-power magnification are well differentiated. Cancer cells that do not retain the structure of the surrounding normal cells are poorly differentiated. Scores range from 1 to 5.

In examining histologic samples of a patient’s prostate tissue, the pathologist will identify the 2 most commonly occurring patterns (types of differentiation) among the cancer cells and assign a numerical value to each pattern. The 2 numbers are then added to yield the final Gleason score. If a single pattern dominates, the pathologist will simply double the corresponding value.

Total scores range from 2 to 10. Scores in the range of 2–4 are considered well-differentiated, 5–7 are moderately differentiated, and 8–10 are poorly differentiated. In general, the higher the score, the worse the prognosis. Men with well-differentiated tumors that are treated conservatively have minimal risk of dying from prostate cancer.

Is declining mortality a sign of screening success?

Prostate cancer mortality has been declining since the mid-1990s in numerous parts of the world; the US,^6,34 Canada,³⁵ Australia,³⁶ and the United Kingdom³⁷ have all reported a reduction in the rate of prostate cancer deaths. Advocates of PSA screening point to this trend as evidence of the effectiveness of screening. But such ecological data are difficult to interpret. For instance, although much less PSA screening is performed in the UK, mortality trends are similar to those in the US where PSA testing has been used more widely.³⁸

Aggressive screening not necessarily the reason. In the US, 2 geographic areas—Seattle, Washington and Connecticut—provided a natural experiment to compare the effect of aggressive screening on prostate cancer mortality (LOE: 2c).³⁹ Although more aggressive screening and treatment took place in the Seattle area, prostate cancer mortality rates were similar to those in Connecticut over 11 years of follow-up. Similarly, in a study in British Columbia, prostate cancer mortality from 1985 to 1999 was not associated with the intensity of PSA screening (LOE: 2c).⁴⁰

Other possible explanations. If the mortality decrease is not related to PSA screening, what could cause it? One explanation is “attribution bias.” Death certificate misattribution of cause of death from prostate cancer may partially explain the pattern of increasing, then decreasing mortality rates (LOE: 2c).⁴¹ Improvement in prostate cancer treatment, especially for advanced stage, and in particular hormone therapy, is another possible explanation for the decreasing prostate cancer mortality (LOE: 2c).^14,42

Benefits of screening

The benefit of any effective screening test is a decrease in the risk of the screened-disease mortality. The best way to demonstrate decreased risk is through a randomized controlled study of the screening test, and 2 such trials are underway for prostate cancer. In the meantime, a decision model estimates that aggressive treatment of organ-confined disease potentially adds 3 years of life for men in their fifties, 1.5 years for men in their sixties, and 0.4 years for men in their seventies (LOE: 2c).³

Others have concluded that 25 men with clinically detected prostate cancer would need to be treated with surgery to prevent 1 prostate cancer death during a 6-year period, without evidence that quality of life is improved (LOE: 2c).⁴³

Consider quality of life. With uncertainty surrounding improvement in the quantity of life as a result of prostate cancer screening, improved quality of life may be an issue for patients. Focus group research has demonstrated that some patients believe it is better to know if a cancer is present than to wonder if it will be diagnosed when it is too late for cure.⁴⁴

General quality of life has been found to be similar among men treated for prostate cancer and age-matched controls without prostate cancer; however, urinary, sexual, and bowel function vary substantially between treated and untreated men and by treatment type (LOE: 3b) (TABLE 2).^45,46 In general, men treated with radical prostatectomy and brachytherapy often report better general quality of life than men who undergo radiation treatment, despite having more urinary and sexual problems (LOE: 2b).^47,48

TABLE 2
Estimates of risk associated with specific prostate cancer treatments 12 months or more after treatment

Harms of screening

The chances of undergoing a biopsy based on an abnormal screening PSA are estimated at 15% to 40% depending on the patient’s age (FIGURE 2).³ There are adverse effects associated with transrectal biopsy of the prostate. In 2 large population-basedstudies of screening, the most frequent complications were hematuria and hematospermia (LOE: 1b, 2b) (TABLE 3), with more serious consequences such as sepsis and hospitalization occurring in fewer than 1% of patients. A study of 100 screened men with an abnormal PSA who underwent prostate biopsy found that although 69% felt moderate to severe pain with the biopsy, 80% would be willing to undergo a repeat biopsy (LOE: 1b).⁴⁹

Treatment options. If the biopsy result is positive, the most common treatment options for localized cancer—which represents over 80% of all prostate cancers diagnosed⁶—include radical prostatectomy, external beam radiation therapy, brachytherapy (internal radiation therapy) or expectant management (watchful waiting). Population-based studies have reported outcomes for these treatment options (TABLE 2). Outcomes derived from hospital-based series of other prostate cancer treatments, such as cryotherapy and 3-dimensional radiation, are available, but the estimates often reflect the experience of only a few hospitals and are not representative of other facilities. Androgen ablation is the standard treatment for metastatic prostate cancer.

Untoward effects of treatment. Approximately 60% of radical prostatectomy patients report some incontinence 12 months or more after surgery (LOE: 2b),^50,51 and about 30% of patients need to wear pads for urine leakage (LOE: 2b).^50-53 Men undergoing radiation therapy have less urinary incontinence, but about 30% complain of diarrhea and loose stools (LOE: 2b).^51,52 Both therapies are associated with a high percentage of erectile dysfunction: approximately 60% of radiation therapy patients and 75% of surgery patients report their erections are not firm enough for intercourse (LOE: 2b).^51,52

Expectant management (following the cancer with regular PSA and ultrasound testing) is sometimes difficult to “sell” to patients whose fear of cancer dictates that the only logical response is to “cut it out.”⁴⁴ A recent randomized trial indicated that radical prostatectomy lowers prostate cancer mortality, local progression, distant metastasis, and overall survival as compared with watchful waiting over a median of 8.2 years of follow-up (LOE: 1b).⁵⁴ However, these results may have little relevance to prostate cancer screening since only 5% of the cancers were screen-detected and 76% were palpable.

FIGURE 2
Yield of screening 1000 men for prostate cancer

TABLE 3
Percentage of patients with specific complication of transrectal prostate biopsy

Recommendations from expert groups

Different expert groups have conflicting recommendations. Both the American Urological Association and the American Cancer Society recommend annual PSA screening starting at age 50 for most men; younger if risk factors are present. Groups that are evidence based tend to recommend a shared decision making process with patients. The AAFP and American College of Physicians advise physicians to counsel men on the known risks and uncertain benefits of screening for prostate cancer. The US Preventive Services Task Force 2002 update concluded that evidence is insufficient to recommend for or against routine screening for prostate cancer using PSA or DRE. The National Cancer Institute cites a lack of evidence to determine a net benefit for PSA or DRE screening.

When will we know more?

Only 1 randomized controlled trial of prostate cancer screening has been completed⁵⁵: 46,193 men were randomized to either PSA and DRE or no screening from 1989 to 1996. The study had methodological problems; for instance, only 23% of the group randomized to screening was screened. The investigators in the trial have interpreted its results as demonstrating a decrease in prostate cancer deaths in the screened group compared with the unscreened group (15 vs 48.7 per 100,000 man-years).⁵⁵ Others have criticized the statistical analysis and calculated the results using an “intent to screen” analysis, finding no difference in prostate cancer deaths between the 2 groups.^3,56

Two randomized controlled trials of screening are ongoing: the National Cancer Institute’s Prostate, Lung, Colon, Ovarian (PLCO) Screening Trial⁵⁷ and the European Randomized Study of Screening for Prostate Cancer.⁵⁸ Both were started in the mid-1990s and will not have results available for a few more years. Also underway is a randomized trial of intervention (radical prostatectomy) versus expectant management, called the Prostate Cancer Intervention Versus Observation Trial (PIVOT).⁵⁹

Counseling recommendations

However, providing men with information on prostate cancer screening before they discussed it with their family physician, rather than after the visit, resulted in patients having a significantly more active role in making a screening decision, and lower levels of decisional conflict (LOE: 2b).⁶¹ Informational pamphlets are available through the AAFP and CDC websites listed in TABLE 4. Additional websites containing prostate cancer screening information are found in TABLE 4. We also provide a bullet item list of key points for discussion with patients (TABLE 5), which can be used along with the balance sheet provided here (TABLE 2).

Shared decision-making is not an easy or quick process. Yet, the majority of patients will benefit from the discussion, regardless of the final decision. Of course, there are instances when a shared decision-making process is well-documented, and still results in an undesirable outcome;⁶² however, while the evidence for screening remains controversial, patients have the right to know that those controversies exist and why they exist.

TABLE 4
Useful websites for patients to find prostate cancer screening information

TABLE 5
Talking points for patients and physicians

CORRESPONDING AUTHOR
Kendra Schwartz, MD, MSPH, 101 E. Alexandrine, Detroit, MI 48201, E-mail: [email protected]

Family physicians are strongly encouraged to support and promote breastfeeding, the optimal form of infant nutrition.¹ Despite its known benefits (fewer infant infections^2-6 and decreased maternal risks of premenopausal breast cancer⁷ and post-menopausal hip fractures⁸), only 64% of mothers initiated breastfeeding in 1998⁹ and only 29% of mothers fed their 6-month-old infant by breast, well below the Healthy People 2010 goal of 50% breastfeeding at 6 months.¹⁰ Clearly, determining the factors that influence breastfeeding beyond the early postpartum period would be beneficial.

Returning to work is a consistent risk factor for weaning.^11-14 The impact of early bottle-feeding on the duration of breastfeeding has been studied with less consistent results.^15,20 Insufficient milk supply is a common subjective reason given for termination.^15,19,21,22 Older women and those with a higher level of education are at less risk of early breastfeeding termination.^{9,11,15,16,21,23,24}

Few investigators have described how breastfeeding patterns may affect breastfeeding duration. Little is known about the effects of timing, frequency, and duration of individual breastfeedings, or the roles of breast pain and infection, sleep, and manual expression on early weaning. We studied women who indicated their intent to breastfeed prenatally to identify demographic factors and breastfeeding patterns associated with weaning in the first 12 weeks postpartum.

Methods

Population

We interviewed breastfeeding women by telephone at 3, 6, 9, and 12 weeks postpartum to investigate lactation mastitis risk factors and predictors of weaning. Pregnant women intending to breastfeed were recruited from 2 geographic sites between June 1994 and January 1998. In suburban Detroit, Michigan, women attending orientation at a freestanding birthing center were asked to participate. In Omaha, Nebraska, women at a single large company were recruited when applying for maternity leave.

Data collection

During the computer-assisted interview, subjects were asked to recall each of the previous 3 weeks. The initial interview, which collected demographic information, typically lasted 15 to 20 minutes; subsequent interviews were shorter. The survey addressed breastfeeding practices and recent health events. Exclusive breastfeeders were women who fed their infants only by breast. We did not collect information on pacifiers; therefore, exclusively breastfed infants may have also received pacifiers. Women who manually expressed or used a device to assist in expression were classified as “pumping” their breasts. Respondents were asked if they had bottle-fed the infant; they were not asked about bottle contents or volume.

Subjects were queried on potential difficulties including breast or nipple pain while nursing, nipple cracks, and mastitis (diagnosed by a health care provider), as well as other health problems and behaviors. Subjects who had stopped breastfeeding in the previous 3 weeks were asked when and why, given a list of possible explanations and an open-ended opportunity. Respondents could provide multiple reasons for termination.

Data analysis

Kaplan-Meier estimates describe the distribution of weaning times for the 2 sites. A log-rank test was used to assess group differences. Relationships between demographic factors and time of weaning were assessed by Cox regression analysis. Discrete survival analysis was used to determine whether variables measured on a weekly basis were related to breastfeeding cessation. Hazard ratios describe the association of the exposures between women who stopped breastfeeding at a given time and those who continued. Because breastfeeding cessation was a rare event in later weeks of the study, as were certain clinical or behavioral breastfeeding factors, weeks 4-12 were collapsed into a single interval. Two variables, number of daily feedings and duration of each feeding, were examined only in the first 3 weeks because the information was often missing beyond 3 weeks. All analyses were performed using the Statistical Package for the Social Sciences.²⁵

Results

Description of subjects

A total of 1057 women agreed to be contacted. Of those, 946 (89.5%) participated in at least 1 interview. Of the 111 women who did not participate, 11 refused and 100 could not be located. Six hundred fifty-eight (69.6%) women completed all 4 interviews. The 56 women who entered the study at week 6 because they could not be reached for the first interview were similar in all factors to women who entered earlier. Of the 946, 711 (75.2%) were from Michigan and 235 (24.8%) were from Nebraska.

Subjects from Michigan were significantly more likely than those from Nebraska to be older than 30 years (52.0% vs 38.3%), have at least a bachelor’s degree (62.9% vs 48.5%), have 3 or more children (38.5% vs 19.6%), and have had a vaginal delivery (99.6% vs 77.0%) (Table W1).* The groups were similar in race, household income, and marital status.

Demographic factors

A total of 673 women (71.1%) continued breastfeeding until 12 weeks; 28% were exclusive breastfeeders. Michigan women were more likely to breastfeed at weeks 2 through 12 than their Nebraskan counterparts (P < .0001, Figure). A college degree was associated with 40% less weaning (Table 1). Age and annual household income were directly related to continued breastfeeding at both sites. Number of children in the household was not associated with termination. Previous breastfeeding experience showed a nonsignificant but consistent trend toward lower weaning risk.

TABLE 1
Relationships of demographics and other characteristics with time to weaning, by site

FIGURE
Probability of breastfeeding, by site, by postpartum week

Clinical and behavioral factors

Because time to weaning differed significantly by site, the survival analyses of clinical and behavioral factors were performed separately for Michigan and Nebraska and controlled for education, age, and previous breastfeeding experience.

During the first 3 weeks, Michigan women with mastitis were nearly 6 times more likely than Michigan women without mastitis to stop breast-feeding in the week of diagnosis (Table 2). Women from Nebraska showed nonsignificant results in the same direction in weeks 4 to 12. (No women from Nebraska with mastitis terminated during weeks 1 through 3.) Although nipple sores and cracks were not associated with weaning, breast pain was associated with weaning. For each day of pain in the first 3 weeks, there was a 10% increase in risk of cessation among Michigan women and a 26% increase among Nebraska women. The association between pain and weaning in weeks 4 through 12 is less clear. In these later weeks, women who reported pain were unexpectedly 75% to 80% more likely to continue breastfeeding than women who did not report pain, yet for Nebraska women the number of days with pain remained significantly associated with breastfeeding cessation.

Subjective depression and breastfeeding cessation were not related. The association between daily sleep and weaning varied by site. During weeks 4 through 12, Michigan women with more daily sleep were less likely to terminate. An opposite, but marginally significant trend, was observed for Nebraska women. Weaning was not associated with outside household help. Nonvaginal birth was not associated with weaning for Nebraska women. (There were only 2 cesarean sections in the Michigan group.)

Michigan women who expressed breast milk during the first 3 weeks were twice as likely to stop breastfeeding as those who did not pump. During the same period, Michigan women who used a bottle for some feedings were 9 times more likely to wean than nonbottle users. Respondents in Nebraska showed similar nonsignificant trends in the first 3 weeks. By contrast, during weeks 4 through 12, both Nebraska and Michigan women who pumped were about 75% less likely to wean, while women who used a bottle for some feedings were 98% less likely to stop breastfeeding.

Breast milk expression increased gradually over time, from 30% of women pumping an average of 3 times per day in the first 3 weeks to 45% of women pumping 5 times per day in the last 3 weeks. To determine if pumping and bottle-feeding had an effect independent of pain or mastitis on weaning in the first 3 weeks, we performed additional analyses controlling for pain, cracks and sores, and mastitis in the same week. The results were similar to those presented in Table 2. Michigan women who pumped were 3 times more likely to wean than those who did not pump (hazard ratio [HR] = 3.0, 95% confidence interval [CI], 1.3 - 6.7), while for Nebraska women there was no association between pumping and weaning (HR = 0.6, 95% CI, 0.3 - 1.5). Bottle-feeding was again significantly associated with weaning in weeks 1 through 3 for Michigan women (HR = 10.9, 95% CI, 4.5 - 26.7) and not associated in Nebraskans (HR = 0.8, 95% CI, 0.4 - 2.0).

Duration and frequency of feedings were investigated as weaning risk factors. There appeared to be a threshold for both variables during the first 3 weeks in Michigan women. Michigan women who breastfed less than 10 minutes per feeding were nearly 5 times more likely to stop breastfeeding than women who breastfed longer. Michigan women who breastfed 6 or fewer times per day were 8 times more likely to stop than those who breastfed more often. Results for Nebraska women fell in the same direction but were not statistically significant.

TABLE 2
Relationships of clinical and behavioral factors to breastfeeding cessation in the same week, adjusted for mother’s age, education, and previous breastfeeding experience

Subjective factors

At each interview, women who had stopped breastfeeding in the previous 3 weeks were asked why they had made that decision. Most women (75%) provided only one reason. At the first interview, insufficient milk supply (37.3%) and breast pain or mastitis (32.9%) were the most common reasons for termination (Table 3). Insufficient milk supply was the reason most often given (35.0%) during weeks 4 through 6. At both weeks 9 and 12, return to work was the reason given most often (53.1% and 58.3%, respectively).

TABLE 3
Percentage of women citing given reason for termination of breastfeeding

Discussion

Mastitis, pain, and days with pain in the first 3 weeks were important clinical factors associated with breastfeeding cessation in this cohort of women who prenatally self-identified as intending to breastfeed. Women who intend to breastfeed should be counseled regarding these possible complications, their temporary nature, prevention, and treatment. Mastitis is not an indication for breastfeeding termination; in fact, increased feedings and milk expression are considered treatment.^26,27 Women who reported pain the first 3 weeks were more likely to stop breastfeeding than women who reported pain after the first 3 weeks. It is difficult to explain this finding; perhaps there are women who have pain during their entire breastfeeding career and yet continue to breastfeed because they are more pain-tolerant, have less severe or frequent pain than those who wean, or are more committed to breastfeeding.

Other clinical factors investigated were depression and daily sleep hours. Weaning was not associated with subjective depression. However, subjects did not undergo formal psychological testing as in the study that reported an association.²⁴ The relationship between daily sleep hours and termination was not consistent, and likely not clinically significant.

The demographic risk factors related to breast-feeding termination in our study are similar to those previously reported,^{14,15,20,21,23,24} namely, younger maternal age and lower educational level. Investigations of parity have been inconsistent.^16,28 We found no association of weaning with parity. Prior breastfeeding experience has been reported as improving breastfeeding rates^15,28; our results are consistent with those findings, but not significantly so. All subjects had access to prenatal breastfeeding education and postnatal breastfeeding support, which may have diminished the differences between women with breastfeeding experience and those without experience.²⁰

Michigan and Nebraska women who pumped or bottle-fed during weeks 4 through 12 were significantly less likely to terminate breastfeeding. In contrast, Michigan women who pumped or bottle-fed during the first 3 weeks postpartum were more likely to terminate even after controlling for pain and mastitis. A commitment to exclusive breastfeeding may be necessary in the early postpartum period for long-term success.^15,19 To our knowledge, the seemingly protective effect associated with pumping and bottle-feeding after the first 3 weeks has not been previously reported.

Breastfeeding 6 or fewer times per day and feedings of 10 minutes or less were associated with termination during the first 3 weeks. Other studies also indicate that the ratio of breast to bottle feedings is important for long-term success. Feinstein and colleagues¹⁵ found that more than one daily bottle of formula supplementation was associated with shorter breastfeeding duration, which was minimized if there were 7 or more breastfeedings per day. Another study found no weaning difference between women who offered their infant only one bottle daily during weeks 2 through 6 and a bottle-avoiding group.¹⁷

The most frequent reasons given for termination were similar to those reported by others, namely, insufficient milk supply and return to work.^11-15,21,22 Insufficient milk supply was a more common reason in the first few weeks after birth; return to work became an increasingly common reason after week 6.

We were unable to examine the role of pacifiers or smoking in breastfeedng termination because pacifier information was not collected and there were too few smokers for meaningful analysis. Smoking has been consistently reported as associated with early cessation.^15,20,29,30 Although pacifier use does not appear to be directly related,^31,32 it has been proposed as a marker for breastfeeding problems. The homogeneity of the sample limits our ability to make generalizations regarding other populations, such as women of color. However, the large sample size and the similarity of termination risk factors between 2 different populations of women lend confidence to our conclusions. As we did not assess mothers’ intentions, some of the variables found associated with termination might be intentional activities of weaning rather than risk factors for termination. The significant difference in termination risk between the sites also may be related to mothers’ intentions or level of commitment. The Michigan women may have intended to breastfeed longer from the outset. The Michigan recruitment site was an alternative birthing center. Women being delivered there may be more persistent in their breast-feeding efforts. Both sites provided access to breast-feeding support personnel, but the Michigan women, as a group, may have been more motivated to continue.

Our results provide clinically useful information. Additional support may be needed for younger and less educated women. Special efforts should be made for early diagnosis and treatment of mastitis and breast pain, particularly during the first 3 weeks. Exclusive breastfeeding without bottle supplementation should be recommended for the first 3 weeks, with at least 7 feedings per day. Each feeding should preferably last more than 10 minutes.

These results should also reassure breastfeeding women and their providers regarding the use of bottles. Bottle-feeding after 3 weeks does not appear to jeopardize breastfeeding success up to 12 weeks and may even improve it.

* Table W1 appears on the JFP Web site at www.jfponline.com.

Acknowledgments

This study was supported by National Institutes of Health grant #30866.

Family physicians are strongly encouraged to support and promote breastfeeding, the optimal form of infant nutrition.¹ Despite its known benefits (fewer infant infections^2-6 and decreased maternal risks of premenopausal breast cancer⁷ and post-menopausal hip fractures⁸), only 64% of mothers initiated breastfeeding in 1998⁹ and only 29% of mothers fed their 6-month-old infant by breast, well below the Healthy People 2010 goal of 50% breastfeeding at 6 months.¹⁰ Clearly, determining the factors that influence breastfeeding beyond the early postpartum period would be beneficial.

Returning to work is a consistent risk factor for weaning.^11-14 The impact of early bottle-feeding on the duration of breastfeeding has been studied with less consistent results.^15,20 Insufficient milk supply is a common subjective reason given for termination.^15,19,21,22 Older women and those with a higher level of education are at less risk of early breastfeeding termination.^{9,11,15,16,21,23,24}

Few investigators have described how breastfeeding patterns may affect breastfeeding duration. Little is known about the effects of timing, frequency, and duration of individual breastfeedings, or the roles of breast pain and infection, sleep, and manual expression on early weaning. We studied women who indicated their intent to breastfeed prenatally to identify demographic factors and breastfeeding patterns associated with weaning in the first 12 weeks postpartum.

Methods

Population

We interviewed breastfeeding women by telephone at 3, 6, 9, and 12 weeks postpartum to investigate lactation mastitis risk factors and predictors of weaning. Pregnant women intending to breastfeed were recruited from 2 geographic sites between June 1994 and January 1998. In suburban Detroit, Michigan, women attending orientation at a freestanding birthing center were asked to participate. In Omaha, Nebraska, women at a single large company were recruited when applying for maternity leave.

Data collection

During the computer-assisted interview, subjects were asked to recall each of the previous 3 weeks. The initial interview, which collected demographic information, typically lasted 15 to 20 minutes; subsequent interviews were shorter. The survey addressed breastfeeding practices and recent health events. Exclusive breastfeeders were women who fed their infants only by breast. We did not collect information on pacifiers; therefore, exclusively breastfed infants may have also received pacifiers. Women who manually expressed or used a device to assist in expression were classified as “pumping” their breasts. Respondents were asked if they had bottle-fed the infant; they were not asked about bottle contents or volume.

Subjects were queried on potential difficulties including breast or nipple pain while nursing, nipple cracks, and mastitis (diagnosed by a health care provider), as well as other health problems and behaviors. Subjects who had stopped breastfeeding in the previous 3 weeks were asked when and why, given a list of possible explanations and an open-ended opportunity. Respondents could provide multiple reasons for termination.

Data analysis

Kaplan-Meier estimates describe the distribution of weaning times for the 2 sites. A log-rank test was used to assess group differences. Relationships between demographic factors and time of weaning were assessed by Cox regression analysis. Discrete survival analysis was used to determine whether variables measured on a weekly basis were related to breastfeeding cessation. Hazard ratios describe the association of the exposures between women who stopped breastfeeding at a given time and those who continued. Because breastfeeding cessation was a rare event in later weeks of the study, as were certain clinical or behavioral breastfeeding factors, weeks 4-12 were collapsed into a single interval. Two variables, number of daily feedings and duration of each feeding, were examined only in the first 3 weeks because the information was often missing beyond 3 weeks. All analyses were performed using the Statistical Package for the Social Sciences.²⁵

Results

Description of subjects

A total of 1057 women agreed to be contacted. Of those, 946 (89.5%) participated in at least 1 interview. Of the 111 women who did not participate, 11 refused and 100 could not be located. Six hundred fifty-eight (69.6%) women completed all 4 interviews. The 56 women who entered the study at week 6 because they could not be reached for the first interview were similar in all factors to women who entered earlier. Of the 946, 711 (75.2%) were from Michigan and 235 (24.8%) were from Nebraska.

Subjects from Michigan were significantly more likely than those from Nebraska to be older than 30 years (52.0% vs 38.3%), have at least a bachelor’s degree (62.9% vs 48.5%), have 3 or more children (38.5% vs 19.6%), and have had a vaginal delivery (99.6% vs 77.0%) (Table W1).* The groups were similar in race, household income, and marital status.

Demographic factors

A total of 673 women (71.1%) continued breastfeeding until 12 weeks; 28% were exclusive breastfeeders. Michigan women were more likely to breastfeed at weeks 2 through 12 than their Nebraskan counterparts (P < .0001, Figure). A college degree was associated with 40% less weaning (Table 1). Age and annual household income were directly related to continued breastfeeding at both sites. Number of children in the household was not associated with termination. Previous breastfeeding experience showed a nonsignificant but consistent trend toward lower weaning risk.

TABLE 1
Relationships of demographics and other characteristics with time to weaning, by site

FIGURE
Probability of breastfeeding, by site, by postpartum week

Clinical and behavioral factors

Because time to weaning differed significantly by site, the survival analyses of clinical and behavioral factors were performed separately for Michigan and Nebraska and controlled for education, age, and previous breastfeeding experience.

During the first 3 weeks, Michigan women with mastitis were nearly 6 times more likely than Michigan women without mastitis to stop breast-feeding in the week of diagnosis (Table 2). Women from Nebraska showed nonsignificant results in the same direction in weeks 4 to 12. (No women from Nebraska with mastitis terminated during weeks 1 through 3.) Although nipple sores and cracks were not associated with weaning, breast pain was associated with weaning. For each day of pain in the first 3 weeks, there was a 10% increase in risk of cessation among Michigan women and a 26% increase among Nebraska women. The association between pain and weaning in weeks 4 through 12 is less clear. In these later weeks, women who reported pain were unexpectedly 75% to 80% more likely to continue breastfeeding than women who did not report pain, yet for Nebraska women the number of days with pain remained significantly associated with breastfeeding cessation.

Subjective depression and breastfeeding cessation were not related. The association between daily sleep and weaning varied by site. During weeks 4 through 12, Michigan women with more daily sleep were less likely to terminate. An opposite, but marginally significant trend, was observed for Nebraska women. Weaning was not associated with outside household help. Nonvaginal birth was not associated with weaning for Nebraska women. (There were only 2 cesarean sections in the Michigan group.)

Michigan women who expressed breast milk during the first 3 weeks were twice as likely to stop breastfeeding as those who did not pump. During the same period, Michigan women who used a bottle for some feedings were 9 times more likely to wean than nonbottle users. Respondents in Nebraska showed similar nonsignificant trends in the first 3 weeks. By contrast, during weeks 4 through 12, both Nebraska and Michigan women who pumped were about 75% less likely to wean, while women who used a bottle for some feedings were 98% less likely to stop breastfeeding.

Breast milk expression increased gradually over time, from 30% of women pumping an average of 3 times per day in the first 3 weeks to 45% of women pumping 5 times per day in the last 3 weeks. To determine if pumping and bottle-feeding had an effect independent of pain or mastitis on weaning in the first 3 weeks, we performed additional analyses controlling for pain, cracks and sores, and mastitis in the same week. The results were similar to those presented in Table 2. Michigan women who pumped were 3 times more likely to wean than those who did not pump (hazard ratio [HR] = 3.0, 95% confidence interval [CI], 1.3 - 6.7), while for Nebraska women there was no association between pumping and weaning (HR = 0.6, 95% CI, 0.3 - 1.5). Bottle-feeding was again significantly associated with weaning in weeks 1 through 3 for Michigan women (HR = 10.9, 95% CI, 4.5 - 26.7) and not associated in Nebraskans (HR = 0.8, 95% CI, 0.4 - 2.0).

Duration and frequency of feedings were investigated as weaning risk factors. There appeared to be a threshold for both variables during the first 3 weeks in Michigan women. Michigan women who breastfed less than 10 minutes per feeding were nearly 5 times more likely to stop breastfeeding than women who breastfed longer. Michigan women who breastfed 6 or fewer times per day were 8 times more likely to stop than those who breastfed more often. Results for Nebraska women fell in the same direction but were not statistically significant.

TABLE 2
Relationships of clinical and behavioral factors to breastfeeding cessation in the same week, adjusted for mother’s age, education, and previous breastfeeding experience

Subjective factors

At each interview, women who had stopped breastfeeding in the previous 3 weeks were asked why they had made that decision. Most women (75%) provided only one reason. At the first interview, insufficient milk supply (37.3%) and breast pain or mastitis (32.9%) were the most common reasons for termination (Table 3). Insufficient milk supply was the reason most often given (35.0%) during weeks 4 through 6. At both weeks 9 and 12, return to work was the reason given most often (53.1% and 58.3%, respectively).

TABLE 3
Percentage of women citing given reason for termination of breastfeeding

Discussion

Mastitis, pain, and days with pain in the first 3 weeks were important clinical factors associated with breastfeeding cessation in this cohort of women who prenatally self-identified as intending to breastfeed. Women who intend to breastfeed should be counseled regarding these possible complications, their temporary nature, prevention, and treatment. Mastitis is not an indication for breastfeeding termination; in fact, increased feedings and milk expression are considered treatment.^26,27 Women who reported pain the first 3 weeks were more likely to stop breastfeeding than women who reported pain after the first 3 weeks. It is difficult to explain this finding; perhaps there are women who have pain during their entire breastfeeding career and yet continue to breastfeed because they are more pain-tolerant, have less severe or frequent pain than those who wean, or are more committed to breastfeeding.

Other clinical factors investigated were depression and daily sleep hours. Weaning was not associated with subjective depression. However, subjects did not undergo formal psychological testing as in the study that reported an association.²⁴ The relationship between daily sleep hours and termination was not consistent, and likely not clinically significant.

The demographic risk factors related to breast-feeding termination in our study are similar to those previously reported,^{14,15,20,21,23,24} namely, younger maternal age and lower educational level. Investigations of parity have been inconsistent.^16,28 We found no association of weaning with parity. Prior breastfeeding experience has been reported as improving breastfeeding rates^15,28; our results are consistent with those findings, but not significantly so. All subjects had access to prenatal breastfeeding education and postnatal breastfeeding support, which may have diminished the differences between women with breastfeeding experience and those without experience.²⁰

Michigan and Nebraska women who pumped or bottle-fed during weeks 4 through 12 were significantly less likely to terminate breastfeeding. In contrast, Michigan women who pumped or bottle-fed during the first 3 weeks postpartum were more likely to terminate even after controlling for pain and mastitis. A commitment to exclusive breastfeeding may be necessary in the early postpartum period for long-term success.^15,19 To our knowledge, the seemingly protective effect associated with pumping and bottle-feeding after the first 3 weeks has not been previously reported.

Breastfeeding 6 or fewer times per day and feedings of 10 minutes or less were associated with termination during the first 3 weeks. Other studies also indicate that the ratio of breast to bottle feedings is important for long-term success. Feinstein and colleagues¹⁵ found that more than one daily bottle of formula supplementation was associated with shorter breastfeeding duration, which was minimized if there were 7 or more breastfeedings per day. Another study found no weaning difference between women who offered their infant only one bottle daily during weeks 2 through 6 and a bottle-avoiding group.¹⁷

The most frequent reasons given for termination were similar to those reported by others, namely, insufficient milk supply and return to work.^11-15,21,22 Insufficient milk supply was a more common reason in the first few weeks after birth; return to work became an increasingly common reason after week 6.

We were unable to examine the role of pacifiers or smoking in breastfeedng termination because pacifier information was not collected and there were too few smokers for meaningful analysis. Smoking has been consistently reported as associated with early cessation.^15,20,29,30 Although pacifier use does not appear to be directly related,^31,32 it has been proposed as a marker for breastfeeding problems. The homogeneity of the sample limits our ability to make generalizations regarding other populations, such as women of color. However, the large sample size and the similarity of termination risk factors between 2 different populations of women lend confidence to our conclusions. As we did not assess mothers’ intentions, some of the variables found associated with termination might be intentional activities of weaning rather than risk factors for termination. The significant difference in termination risk between the sites also may be related to mothers’ intentions or level of commitment. The Michigan women may have intended to breastfeed longer from the outset. The Michigan recruitment site was an alternative birthing center. Women being delivered there may be more persistent in their breast-feeding efforts. Both sites provided access to breast-feeding support personnel, but the Michigan women, as a group, may have been more motivated to continue.

Our results provide clinically useful information. Additional support may be needed for younger and less educated women. Special efforts should be made for early diagnosis and treatment of mastitis and breast pain, particularly during the first 3 weeks. Exclusive breastfeeding without bottle supplementation should be recommended for the first 3 weeks, with at least 7 feedings per day. Each feeding should preferably last more than 10 minutes.

These results should also reassure breastfeeding women and their providers regarding the use of bottles. Bottle-feeding after 3 weeks does not appear to jeopardize breastfeeding success up to 12 weeks and may even improve it.

* Table W1 appears on the JFP Web site at www.jfponline.com.

Acknowledgments

This study was supported by National Institutes of Health grant #30866.

Family physicians are strongly encouraged to support and promote breastfeeding, the optimal form of infant nutrition.¹ Despite its known benefits (fewer infant infections^2-6 and decreased maternal risks of premenopausal breast cancer⁷ and post-menopausal hip fractures⁸), only 64% of mothers initiated breastfeeding in 1998⁹ and only 29% of mothers fed their 6-month-old infant by breast, well below the Healthy People 2010 goal of 50% breastfeeding at 6 months.¹⁰ Clearly, determining the factors that influence breastfeeding beyond the early postpartum period would be beneficial.

Returning to work is a consistent risk factor for weaning.^11-14 The impact of early bottle-feeding on the duration of breastfeeding has been studied with less consistent results.^15,20 Insufficient milk supply is a common subjective reason given for termination.^15,19,21,22 Older women and those with a higher level of education are at less risk of early breastfeeding termination.^{9,11,15,16,21,23,24}

Few investigators have described how breastfeeding patterns may affect breastfeeding duration. Little is known about the effects of timing, frequency, and duration of individual breastfeedings, or the roles of breast pain and infection, sleep, and manual expression on early weaning. We studied women who indicated their intent to breastfeed prenatally to identify demographic factors and breastfeeding patterns associated with weaning in the first 12 weeks postpartum.

Methods

Population

We interviewed breastfeeding women by telephone at 3, 6, 9, and 12 weeks postpartum to investigate lactation mastitis risk factors and predictors of weaning. Pregnant women intending to breastfeed were recruited from 2 geographic sites between June 1994 and January 1998. In suburban Detroit, Michigan, women attending orientation at a freestanding birthing center were asked to participate. In Omaha, Nebraska, women at a single large company were recruited when applying for maternity leave.

Data collection

During the computer-assisted interview, subjects were asked to recall each of the previous 3 weeks. The initial interview, which collected demographic information, typically lasted 15 to 20 minutes; subsequent interviews were shorter. The survey addressed breastfeeding practices and recent health events. Exclusive breastfeeders were women who fed their infants only by breast. We did not collect information on pacifiers; therefore, exclusively breastfed infants may have also received pacifiers. Women who manually expressed or used a device to assist in expression were classified as “pumping” their breasts. Respondents were asked if they had bottle-fed the infant; they were not asked about bottle contents or volume.

Subjects were queried on potential difficulties including breast or nipple pain while nursing, nipple cracks, and mastitis (diagnosed by a health care provider), as well as other health problems and behaviors. Subjects who had stopped breastfeeding in the previous 3 weeks were asked when and why, given a list of possible explanations and an open-ended opportunity. Respondents could provide multiple reasons for termination.

Data analysis

Kaplan-Meier estimates describe the distribution of weaning times for the 2 sites. A log-rank test was used to assess group differences. Relationships between demographic factors and time of weaning were assessed by Cox regression analysis. Discrete survival analysis was used to determine whether variables measured on a weekly basis were related to breastfeeding cessation. Hazard ratios describe the association of the exposures between women who stopped breastfeeding at a given time and those who continued. Because breastfeeding cessation was a rare event in later weeks of the study, as were certain clinical or behavioral breastfeeding factors, weeks 4-12 were collapsed into a single interval. Two variables, number of daily feedings and duration of each feeding, were examined only in the first 3 weeks because the information was often missing beyond 3 weeks. All analyses were performed using the Statistical Package for the Social Sciences.²⁵

Results

Description of subjects

A total of 1057 women agreed to be contacted. Of those, 946 (89.5%) participated in at least 1 interview. Of the 111 women who did not participate, 11 refused and 100 could not be located. Six hundred fifty-eight (69.6%) women completed all 4 interviews. The 56 women who entered the study at week 6 because they could not be reached for the first interview were similar in all factors to women who entered earlier. Of the 946, 711 (75.2%) were from Michigan and 235 (24.8%) were from Nebraska.

Subjects from Michigan were significantly more likely than those from Nebraska to be older than 30 years (52.0% vs 38.3%), have at least a bachelor’s degree (62.9% vs 48.5%), have 3 or more children (38.5% vs 19.6%), and have had a vaginal delivery (99.6% vs 77.0%) (Table W1).* The groups were similar in race, household income, and marital status.

Demographic factors

A total of 673 women (71.1%) continued breastfeeding until 12 weeks; 28% were exclusive breastfeeders. Michigan women were more likely to breastfeed at weeks 2 through 12 than their Nebraskan counterparts (P < .0001, Figure). A college degree was associated with 40% less weaning (Table 1). Age and annual household income were directly related to continued breastfeeding at both sites. Number of children in the household was not associated with termination. Previous breastfeeding experience showed a nonsignificant but consistent trend toward lower weaning risk.

TABLE 1
Relationships of demographics and other characteristics with time to weaning, by site

FIGURE
Probability of breastfeeding, by site, by postpartum week

Clinical and behavioral factors

Because time to weaning differed significantly by site, the survival analyses of clinical and behavioral factors were performed separately for Michigan and Nebraska and controlled for education, age, and previous breastfeeding experience.

During the first 3 weeks, Michigan women with mastitis were nearly 6 times more likely than Michigan women without mastitis to stop breast-feeding in the week of diagnosis (Table 2). Women from Nebraska showed nonsignificant results in the same direction in weeks 4 to 12. (No women from Nebraska with mastitis terminated during weeks 1 through 3.) Although nipple sores and cracks were not associated with weaning, breast pain was associated with weaning. For each day of pain in the first 3 weeks, there was a 10% increase in risk of cessation among Michigan women and a 26% increase among Nebraska women. The association between pain and weaning in weeks 4 through 12 is less clear. In these later weeks, women who reported pain were unexpectedly 75% to 80% more likely to continue breastfeeding than women who did not report pain, yet for Nebraska women the number of days with pain remained significantly associated with breastfeeding cessation.

Subjective depression and breastfeeding cessation were not related. The association between daily sleep and weaning varied by site. During weeks 4 through 12, Michigan women with more daily sleep were less likely to terminate. An opposite, but marginally significant trend, was observed for Nebraska women. Weaning was not associated with outside household help. Nonvaginal birth was not associated with weaning for Nebraska women. (There were only 2 cesarean sections in the Michigan group.)

Michigan women who expressed breast milk during the first 3 weeks were twice as likely to stop breastfeeding as those who did not pump. During the same period, Michigan women who used a bottle for some feedings were 9 times more likely to wean than nonbottle users. Respondents in Nebraska showed similar nonsignificant trends in the first 3 weeks. By contrast, during weeks 4 through 12, both Nebraska and Michigan women who pumped were about 75% less likely to wean, while women who used a bottle for some feedings were 98% less likely to stop breastfeeding.

Breast milk expression increased gradually over time, from 30% of women pumping an average of 3 times per day in the first 3 weeks to 45% of women pumping 5 times per day in the last 3 weeks. To determine if pumping and bottle-feeding had an effect independent of pain or mastitis on weaning in the first 3 weeks, we performed additional analyses controlling for pain, cracks and sores, and mastitis in the same week. The results were similar to those presented in Table 2. Michigan women who pumped were 3 times more likely to wean than those who did not pump (hazard ratio [HR] = 3.0, 95% confidence interval [CI], 1.3 - 6.7), while for Nebraska women there was no association between pumping and weaning (HR = 0.6, 95% CI, 0.3 - 1.5). Bottle-feeding was again significantly associated with weaning in weeks 1 through 3 for Michigan women (HR = 10.9, 95% CI, 4.5 - 26.7) and not associated in Nebraskans (HR = 0.8, 95% CI, 0.4 - 2.0).

Duration and frequency of feedings were investigated as weaning risk factors. There appeared to be a threshold for both variables during the first 3 weeks in Michigan women. Michigan women who breastfed less than 10 minutes per feeding were nearly 5 times more likely to stop breastfeeding than women who breastfed longer. Michigan women who breastfed 6 or fewer times per day were 8 times more likely to stop than those who breastfed more often. Results for Nebraska women fell in the same direction but were not statistically significant.

TABLE 2
Relationships of clinical and behavioral factors to breastfeeding cessation in the same week, adjusted for mother’s age, education, and previous breastfeeding experience

Subjective factors

At each interview, women who had stopped breastfeeding in the previous 3 weeks were asked why they had made that decision. Most women (75%) provided only one reason. At the first interview, insufficient milk supply (37.3%) and breast pain or mastitis (32.9%) were the most common reasons for termination (Table 3). Insufficient milk supply was the reason most often given (35.0%) during weeks 4 through 6. At both weeks 9 and 12, return to work was the reason given most often (53.1% and 58.3%, respectively).

TABLE 3
Percentage of women citing given reason for termination of breastfeeding

Discussion

Mastitis, pain, and days with pain in the first 3 weeks were important clinical factors associated with breastfeeding cessation in this cohort of women who prenatally self-identified as intending to breastfeed. Women who intend to breastfeed should be counseled regarding these possible complications, their temporary nature, prevention, and treatment. Mastitis is not an indication for breastfeeding termination; in fact, increased feedings and milk expression are considered treatment.^26,27 Women who reported pain the first 3 weeks were more likely to stop breastfeeding than women who reported pain after the first 3 weeks. It is difficult to explain this finding; perhaps there are women who have pain during their entire breastfeeding career and yet continue to breastfeed because they are more pain-tolerant, have less severe or frequent pain than those who wean, or are more committed to breastfeeding.

Other clinical factors investigated were depression and daily sleep hours. Weaning was not associated with subjective depression. However, subjects did not undergo formal psychological testing as in the study that reported an association.²⁴ The relationship between daily sleep hours and termination was not consistent, and likely not clinically significant.

The demographic risk factors related to breast-feeding termination in our study are similar to those previously reported,^{14,15,20,21,23,24} namely, younger maternal age and lower educational level. Investigations of parity have been inconsistent.^16,28 We found no association of weaning with parity. Prior breastfeeding experience has been reported as improving breastfeeding rates^15,28; our results are consistent with those findings, but not significantly so. All subjects had access to prenatal breastfeeding education and postnatal breastfeeding support, which may have diminished the differences between women with breastfeeding experience and those without experience.²⁰

Michigan and Nebraska women who pumped or bottle-fed during weeks 4 through 12 were significantly less likely to terminate breastfeeding. In contrast, Michigan women who pumped or bottle-fed during the first 3 weeks postpartum were more likely to terminate even after controlling for pain and mastitis. A commitment to exclusive breastfeeding may be necessary in the early postpartum period for long-term success.^15,19 To our knowledge, the seemingly protective effect associated with pumping and bottle-feeding after the first 3 weeks has not been previously reported.

Breastfeeding 6 or fewer times per day and feedings of 10 minutes or less were associated with termination during the first 3 weeks. Other studies also indicate that the ratio of breast to bottle feedings is important for long-term success. Feinstein and colleagues¹⁵ found that more than one daily bottle of formula supplementation was associated with shorter breastfeeding duration, which was minimized if there were 7 or more breastfeedings per day. Another study found no weaning difference between women who offered their infant only one bottle daily during weeks 2 through 6 and a bottle-avoiding group.¹⁷

The most frequent reasons given for termination were similar to those reported by others, namely, insufficient milk supply and return to work.^11-15,21,22 Insufficient milk supply was a more common reason in the first few weeks after birth; return to work became an increasingly common reason after week 6.

We were unable to examine the role of pacifiers or smoking in breastfeedng termination because pacifier information was not collected and there were too few smokers for meaningful analysis. Smoking has been consistently reported as associated with early cessation.^15,20,29,30 Although pacifier use does not appear to be directly related,^31,32 it has been proposed as a marker for breastfeeding problems. The homogeneity of the sample limits our ability to make generalizations regarding other populations, such as women of color. However, the large sample size and the similarity of termination risk factors between 2 different populations of women lend confidence to our conclusions. As we did not assess mothers’ intentions, some of the variables found associated with termination might be intentional activities of weaning rather than risk factors for termination. The significant difference in termination risk between the sites also may be related to mothers’ intentions or level of commitment. The Michigan women may have intended to breastfeed longer from the outset. The Michigan recruitment site was an alternative birthing center. Women being delivered there may be more persistent in their breast-feeding efforts. Both sites provided access to breast-feeding support personnel, but the Michigan women, as a group, may have been more motivated to continue.

Our results provide clinically useful information. Additional support may be needed for younger and less educated women. Special efforts should be made for early diagnosis and treatment of mastitis and breast pain, particularly during the first 3 weeks. Exclusive breastfeeding without bottle supplementation should be recommended for the first 3 weeks, with at least 7 feedings per day. Each feeding should preferably last more than 10 minutes.

These results should also reassure breastfeeding women and their providers regarding the use of bottles. Bottle-feeding after 3 weeks does not appear to jeopardize breastfeeding success up to 12 weeks and may even improve it.

* Table W1 appears on the JFP Web site at www.jfponline.com.

Acknowledgments

This study was supported by National Institutes of Health grant #30866.

ABSTRACT

BACKGROUND: Based on randomized clinical trials (RCTs), the most effective first-line eradication therapy for Helicobacter pylori is a combination of proton pump inhibitor (PPI) and 2 antimicrobial agents.¹ Yet there remains a significant treatment failure rate of 5% to 25%. Antibiotic resistance is the major impediment of cure.² Using a pooled analysis approach, the authors determined the second-line treatment strategy resulting in the greatest percentage of H pylorieradication.

POPULATION STUDIED: Studies of H pylori re-eradication in adults were retrieved from MEDLINE database, reference lists of retrieved research papers, and major congress abstract lists. All studies were performed between 1994 and 1999, were conducted prospectively (or without information on study design), contained detailed information on eradication agents, and included subjects with only one treatment failure. Eighteen articles and 47 abstracts were identified; 16 articles and 24 abstracts met the inclusion criteria.

STUDY DESIGN AND VALIDITY: The authors performed a pooled efficacy analysis of re-treatment regimens for H pylori eradication in adults. They included all prospective studies—randomized and nonrandomized—that reported eradication rates in patients previously treated with antibiotic therapy. The inclusion criteria were appropriate, and the search for relevant articles was complete in that the authors included abstracts from international gastroenterology meetings. An analysis strategy of simple pooling was used (total number of patients successfully treated divided by all those enrolled in given treatment category), which is appropriate since the primary outcome was the eradication rate.

OUTCOMES MEASURED: The authors calculated a pooled eradication rate with 95% confidence intervals for each of the 6 general treatment categories. They also determined the eradication rate of each second treatment, accounting for differences in initial treatment therapy regimen. Finally, they ascertained if the effectiveness of second-line therapy regimens improved when distinct antimicrobials not used in the first attempt at treatment were used.

RESULTS: The most effective second-line therapies for eradication of H pylori were quadruple therapy, either ranitidine-bismuth-based triple therapy (ranitidine-bismuth product plus 2 antimicrobials) at 80.2% (95% confidence interval [CI], 75%-85%) or H2-blocker or PPI, bismuth compound, and 2 antimicrobials) at 75.8% (95% CI, 73%-79%). The second-line treatment eradication rate was lower when the initial therapy was a PPI with 2 antimicrobials versus a PPI with one antimicrobial. Re-treatment was more difficult with an increased number of antimicrobials used in initial therapy. The re-treatment eradication rate was greater when 2 new antimicrobials were included in the regimen than when a single new antimicrobial was added (P=.0064).

ABSTRACT

BACKGROUND: Based on randomized clinical trials (RCTs), the most effective first-line eradication therapy for Helicobacter pylori is a combination of proton pump inhibitor (PPI) and 2 antimicrobial agents.¹ Yet there remains a significant treatment failure rate of 5% to 25%. Antibiotic resistance is the major impediment of cure.² Using a pooled analysis approach, the authors determined the second-line treatment strategy resulting in the greatest percentage of H pylorieradication.

POPULATION STUDIED: Studies of H pylori re-eradication in adults were retrieved from MEDLINE database, reference lists of retrieved research papers, and major congress abstract lists. All studies were performed between 1994 and 1999, were conducted prospectively (or without information on study design), contained detailed information on eradication agents, and included subjects with only one treatment failure. Eighteen articles and 47 abstracts were identified; 16 articles and 24 abstracts met the inclusion criteria.

STUDY DESIGN AND VALIDITY: The authors performed a pooled efficacy analysis of re-treatment regimens for H pylori eradication in adults. They included all prospective studies—randomized and nonrandomized—that reported eradication rates in patients previously treated with antibiotic therapy. The inclusion criteria were appropriate, and the search for relevant articles was complete in that the authors included abstracts from international gastroenterology meetings. An analysis strategy of simple pooling was used (total number of patients successfully treated divided by all those enrolled in given treatment category), which is appropriate since the primary outcome was the eradication rate.

OUTCOMES MEASURED: The authors calculated a pooled eradication rate with 95% confidence intervals for each of the 6 general treatment categories. They also determined the eradication rate of each second treatment, accounting for differences in initial treatment therapy regimen. Finally, they ascertained if the effectiveness of second-line therapy regimens improved when distinct antimicrobials not used in the first attempt at treatment were used.

RESULTS: The most effective second-line therapies for eradication of H pylori were quadruple therapy, either ranitidine-bismuth-based triple therapy (ranitidine-bismuth product plus 2 antimicrobials) at 80.2% (95% confidence interval [CI], 75%-85%) or H2-blocker or PPI, bismuth compound, and 2 antimicrobials) at 75.8% (95% CI, 73%-79%). The second-line treatment eradication rate was lower when the initial therapy was a PPI with 2 antimicrobials versus a PPI with one antimicrobial. Re-treatment was more difficult with an increased number of antimicrobials used in initial therapy. The re-treatment eradication rate was greater when 2 new antimicrobials were included in the regimen than when a single new antimicrobial was added (P=.0064).

ABSTRACT

BACKGROUND: Based on randomized clinical trials (RCTs), the most effective first-line eradication therapy for Helicobacter pylori is a combination of proton pump inhibitor (PPI) and 2 antimicrobial agents.¹ Yet there remains a significant treatment failure rate of 5% to 25%. Antibiotic resistance is the major impediment of cure.² Using a pooled analysis approach, the authors determined the second-line treatment strategy resulting in the greatest percentage of H pylorieradication.

POPULATION STUDIED: Studies of H pylori re-eradication in adults were retrieved from MEDLINE database, reference lists of retrieved research papers, and major congress abstract lists. All studies were performed between 1994 and 1999, were conducted prospectively (or without information on study design), contained detailed information on eradication agents, and included subjects with only one treatment failure. Eighteen articles and 47 abstracts were identified; 16 articles and 24 abstracts met the inclusion criteria.

STUDY DESIGN AND VALIDITY: The authors performed a pooled efficacy analysis of re-treatment regimens for H pylori eradication in adults. They included all prospective studies—randomized and nonrandomized—that reported eradication rates in patients previously treated with antibiotic therapy. The inclusion criteria were appropriate, and the search for relevant articles was complete in that the authors included abstracts from international gastroenterology meetings. An analysis strategy of simple pooling was used (total number of patients successfully treated divided by all those enrolled in given treatment category), which is appropriate since the primary outcome was the eradication rate.

OUTCOMES MEASURED: The authors calculated a pooled eradication rate with 95% confidence intervals for each of the 6 general treatment categories. They also determined the eradication rate of each second treatment, accounting for differences in initial treatment therapy regimen. Finally, they ascertained if the effectiveness of second-line therapy regimens improved when distinct antimicrobials not used in the first attempt at treatment were used.

RESULTS: The most effective second-line therapies for eradication of H pylori were quadruple therapy, either ranitidine-bismuth-based triple therapy (ranitidine-bismuth product plus 2 antimicrobials) at 80.2% (95% confidence interval [CI], 75%-85%) or H2-blocker or PPI, bismuth compound, and 2 antimicrobials) at 75.8% (95% CI, 73%-79%). The second-line treatment eradication rate was lower when the initial therapy was a PPI with 2 antimicrobials versus a PPI with one antimicrobial. Re-treatment was more difficult with an increased number of antimicrobials used in initial therapy. The re-treatment eradication rate was greater when 2 new antimicrobials were included in the regimen than when a single new antimicrobial was added (P=.0064).

Evidence Summary

Many trials report bone mineral density (BMD) as their primary outcome, which may or may not be associated with fracture rates. Relatively few studies have reported patient-oriented outcomes, such as vertebral or nonvertebral fracture.

Alendronate and risedronate have both reduced vertebral and nonvertebral fractures and increased BMD in clinical trials. The pooled risk estimates for alendronate’s antifracture effect are 48% to 53% vertebral and hip fracture reduction and 30% reduction of all clinical fractures (relative risk [RR] = 0.70; 95% confidence interval [CI], 0.59-0.82).1 Risedronate reduced hip fractures in osteoporotic women aged 70 to 79 years (RR = 0.6; 95% CI, 0.4-0.9), but women 80 years and older with nonskeletal risk factors did not benefit.² One alendronate trial in 241 osteoporotic men demonstrated increased spine and hip BMD and a decreased incidence of vertebral fractures.³

HRT has been associated with increased BMD in RCTs, fewer fractures in observational studies, and fewer fractures in a recent meta-analysis of RCTs. HRT had the greatest benefit among women younger than 60 years (RR = 0.67; 95% CI, 0.56-0.94).4 For women older than 60 years, a nonsignificant trend toward benefit was found (RR = 0.88, 95% CI, 0.71-1.08). However, in a large RCT of postmenopausal women, 85% without osteoporotic BMD, 4 years of HRT did not reduce fracture incidence.⁵ Testosterone replacement in men is less studied; some small studies demonstrate increased BMD, but long-term benefits are unknown.⁶

Raloxifene and calcitonin have some supporting evidence as well. In a large 4-year placebo-controlled trial, raloxifene decreased vertebral fracture risk (RR = 0.7; 95% CI, 0.5-0.8) but had no effect on hip fracture and a threefold increase in thromboembolic risk (RR = 3.1; 95% CI, 1.5-6.2).7 Regarding calcitonin, one RCT compared 3 doses of intranasal calcitonin and placebo in osteoporotic women and showed an increase in lumbar spine BMD in treated groups.8 Over the 5-year study (59% of participants were lost to follow-up), there was a 39% reduction in risk of radiologic deformities (P=.03) only for women taking 200 IU.

Many subjects also used vitamin D and calcium, but effective doses of both agents are still in question. Combined calcium and vitamin D supplements have decreased fracture risk using 700 to 800 IU of vitamin D and 500 to 1200 mg of calcium daily,9 but a systematic review of vitamin D without calcium found no clear benefit.10 A meta-analysis showed that exercise training programs may play a preventive role, potentially preventing or reversing 1% of bone loss annually among women.11 Smoking cessation may also increase bone density.¹²

Recommendations From Others

The Royal College of Physician Guidelines on Osteoporosis Prevention and Treatment gives HRT an A grade of recommendation for preventing spinal fractures and a B grade for nonvertebral fracture prevention, while alendronate and risedronate received an A grade for both vertebral and nonvertebral fracture prevention. The American Association of Clinical Endocrinologists (AACE) states that HRT is the standard of care for preventing and treating postmenopausal bone loss but supports the use of bisphosphonates as an alternative. The US Preventive Services Task Force emphasizes smoking cessation, regular exercise, and adequate calcium intake, and recommends discussion with patients of the risks and benefits of HRT.

Evidence Summary

Many trials report bone mineral density (BMD) as their primary outcome, which may or may not be associated with fracture rates. Relatively few studies have reported patient-oriented outcomes, such as vertebral or nonvertebral fracture.

Alendronate and risedronate have both reduced vertebral and nonvertebral fractures and increased BMD in clinical trials. The pooled risk estimates for alendronate’s antifracture effect are 48% to 53% vertebral and hip fracture reduction and 30% reduction of all clinical fractures (relative risk [RR] = 0.70; 95% confidence interval [CI], 0.59-0.82).1 Risedronate reduced hip fractures in osteoporotic women aged 70 to 79 years (RR = 0.6; 95% CI, 0.4-0.9), but women 80 years and older with nonskeletal risk factors did not benefit.² One alendronate trial in 241 osteoporotic men demonstrated increased spine and hip BMD and a decreased incidence of vertebral fractures.³

HRT has been associated with increased BMD in RCTs, fewer fractures in observational studies, and fewer fractures in a recent meta-analysis of RCTs. HRT had the greatest benefit among women younger than 60 years (RR = 0.67; 95% CI, 0.56-0.94).4 For women older than 60 years, a nonsignificant trend toward benefit was found (RR = 0.88, 95% CI, 0.71-1.08). However, in a large RCT of postmenopausal women, 85% without osteoporotic BMD, 4 years of HRT did not reduce fracture incidence.⁵ Testosterone replacement in men is less studied; some small studies demonstrate increased BMD, but long-term benefits are unknown.⁶

Raloxifene and calcitonin have some supporting evidence as well. In a large 4-year placebo-controlled trial, raloxifene decreased vertebral fracture risk (RR = 0.7; 95% CI, 0.5-0.8) but had no effect on hip fracture and a threefold increase in thromboembolic risk (RR = 3.1; 95% CI, 1.5-6.2).7 Regarding calcitonin, one RCT compared 3 doses of intranasal calcitonin and placebo in osteoporotic women and showed an increase in lumbar spine BMD in treated groups.8 Over the 5-year study (59% of participants were lost to follow-up), there was a 39% reduction in risk of radiologic deformities (P=.03) only for women taking 200 IU.

Many subjects also used vitamin D and calcium, but effective doses of both agents are still in question. Combined calcium and vitamin D supplements have decreased fracture risk using 700 to 800 IU of vitamin D and 500 to 1200 mg of calcium daily,9 but a systematic review of vitamin D without calcium found no clear benefit.10 A meta-analysis showed that exercise training programs may play a preventive role, potentially preventing or reversing 1% of bone loss annually among women.11 Smoking cessation may also increase bone density.¹²

Recommendations From Others

The Royal College of Physician Guidelines on Osteoporosis Prevention and Treatment gives HRT an A grade of recommendation for preventing spinal fractures and a B grade for nonvertebral fracture prevention, while alendronate and risedronate received an A grade for both vertebral and nonvertebral fracture prevention. The American Association of Clinical Endocrinologists (AACE) states that HRT is the standard of care for preventing and treating postmenopausal bone loss but supports the use of bisphosphonates as an alternative. The US Preventive Services Task Force emphasizes smoking cessation, regular exercise, and adequate calcium intake, and recommends discussion with patients of the risks and benefits of HRT.

Evidence Summary

Many trials report bone mineral density (BMD) as their primary outcome, which may or may not be associated with fracture rates. Relatively few studies have reported patient-oriented outcomes, such as vertebral or nonvertebral fracture.

Alendronate and risedronate have both reduced vertebral and nonvertebral fractures and increased BMD in clinical trials. The pooled risk estimates for alendronate’s antifracture effect are 48% to 53% vertebral and hip fracture reduction and 30% reduction of all clinical fractures (relative risk [RR] = 0.70; 95% confidence interval [CI], 0.59-0.82).1 Risedronate reduced hip fractures in osteoporotic women aged 70 to 79 years (RR = 0.6; 95% CI, 0.4-0.9), but women 80 years and older with nonskeletal risk factors did not benefit.² One alendronate trial in 241 osteoporotic men demonstrated increased spine and hip BMD and a decreased incidence of vertebral fractures.³

HRT has been associated with increased BMD in RCTs, fewer fractures in observational studies, and fewer fractures in a recent meta-analysis of RCTs. HRT had the greatest benefit among women younger than 60 years (RR = 0.67; 95% CI, 0.56-0.94).4 For women older than 60 years, a nonsignificant trend toward benefit was found (RR = 0.88, 95% CI, 0.71-1.08). However, in a large RCT of postmenopausal women, 85% without osteoporotic BMD, 4 years of HRT did not reduce fracture incidence.⁵ Testosterone replacement in men is less studied; some small studies demonstrate increased BMD, but long-term benefits are unknown.⁶

Raloxifene and calcitonin have some supporting evidence as well. In a large 4-year placebo-controlled trial, raloxifene decreased vertebral fracture risk (RR = 0.7; 95% CI, 0.5-0.8) but had no effect on hip fracture and a threefold increase in thromboembolic risk (RR = 3.1; 95% CI, 1.5-6.2).7 Regarding calcitonin, one RCT compared 3 doses of intranasal calcitonin and placebo in osteoporotic women and showed an increase in lumbar spine BMD in treated groups.8 Over the 5-year study (59% of participants were lost to follow-up), there was a 39% reduction in risk of radiologic deformities (P=.03) only for women taking 200 IU.

Many subjects also used vitamin D and calcium, but effective doses of both agents are still in question. Combined calcium and vitamin D supplements have decreased fracture risk using 700 to 800 IU of vitamin D and 500 to 1200 mg of calcium daily,9 but a systematic review of vitamin D without calcium found no clear benefit.10 A meta-analysis showed that exercise training programs may play a preventive role, potentially preventing or reversing 1% of bone loss annually among women.11 Smoking cessation may also increase bone density.¹²

Recommendations From Others

The Royal College of Physician Guidelines on Osteoporosis Prevention and Treatment gives HRT an A grade of recommendation for preventing spinal fractures and a B grade for nonvertebral fracture prevention, while alendronate and risedronate received an A grade for both vertebral and nonvertebral fracture prevention. The American Association of Clinical Endocrinologists (AACE) states that HRT is the standard of care for preventing and treating postmenopausal bone loss but supports the use of bisphosphonates as an alternative. The US Preventive Services Task Force emphasizes smoking cessation, regular exercise, and adequate calcium intake, and recommends discussion with patients of the risks and benefits of HRT.