Survival of patients who receive a diagnosis of early-stage disease is substantially better than that of patients who receive a diagnosis of late-stage disease [4, 6]. If digital rectal examinations and measurements of prostate-specific antigen (PSA) are used to screen men who are asymptomatic or those who have symptoms that are consistent with benign prostatic hyperplasia, the disease is diagnosed earlier and the proportion of cases that are discovered while the cancer is still confined to the prostate is increased. This so-called stage shift has generated considerable advocacy for the methods that result in early detection [7-9].

Some experts consider stage shifts to be inadmissible evidence for screening because of the effects of lead-time and length biases [10]. Lead-time bias occurs if the time that appears to be added to survival after diagnosis is actually a result of earlier detection. Length blas occurs if selective screening identifies more indolent tumors with better prognoses. Both of these biases can produce an apparent survival benefit even when screening is ineffective [11, 12]. Lead-time and length biases are most likely to mislead when many patients with the target condition have undiagnosed but detectable disease for a long time and when the disease's biological behavior is heterogeneous. This appears to be the case with prostate cancer [7, 12, 13].

Only a minority of prostate cancer that are detected as a result of PSA measurement and are treated surgically have been considered clinically insignificant [14, 15]. However, even histologically significant prostate cancer may not result in early death or reduced quality of life, particularly when it occurs in older men who face other risks for death. The ratio of cumulative incidence, which is increasing because of early detection efforts, to mortality rate, which is relatively stable, suggests that most cases of prostate cancer that are being discovered with present methods are not fatal.

Moreover, aggressive treatment of prostate cancer confers substantial risk for illness and a small but finite risk for death, which must be borne immediately in return for a putative benefit that may be realized only in the distant future [16-18]. An overview of observational studies indicates that expectant management provides a 10-year disease-specific survival rate greater than 80% for many men with prostate cancer of low-to-moderate grade [19]. No published controlled trials have proven that radical prostatectomy or radiotherapy reduces rates of death from clinically localized prostate cancer. Decision analyses have raised doubts about the benefits of early detection and aggressive treatment [20, 21]. Debate also exists about which prostate cancers merit aggressive treatment [22-24].

Some unanswered questions about screening for prostate cancer are being addressed in randomized trials in the United States and Europe [25-28]. However, the results of these trials may not be available for more than a decade. Meanwhile, clinicians and patients receive conflicting advice. Those who are willing to assume that the benefits outweigh the risks until direct evidence is available favor screening [29-32]. The American Cancer Society has recommended that patients begin having annual digital rectal examination and PSA measurement at 50 years of age (40 years for men at increased risk) [33]. Conversely, the U.S. Preventive Services Task Force and the Canadian Task Force on the Periodic Health Examination have found the evidence insufficient to recommend routine screening with digital rectal examination and PSA measurement [34, 35].

These conflicting recommendations reflect differences in the level of evidence required to make a positive recommendation rather than different interpretations of the results of existing studies [36-40]. We provide a thorough review and balanced synthesis of evidence that relates to the decision to screen for prostate cancer and thereby support clinical decision making. Our findings are presented in two parts. This first part considers the epidemiology of prostate cancer and its biological behavior as it bears on determining the appropriate target of a screening strategy. It also presents an analysis of the effectiveness of available tests for identifying this type of cancer. Part II, to be published in the 15 March issue of Annals, adds reviews of the natural history and the outcomes of alternate treatments and provides estimates of the benefits, risks, and costs of prostate cancer screening for men of various ages. The American College of Physicians' guidelines for prostate cancer screening will also be presented in the 15 March issue.

Methods

Top Methods Author & Article Info References

Literature Review

Our review of the literature began with a systematic general search of the MEDLINE database for articles published from 1966 to 1995. The methods and detailed results of this structured review have been reported elsewhere [41].

Estimates of the Age-Specific Prevalence of Target Prostate Cancer

Assessment of the value of early prostate cancer detection, through derivation of test likelihood ratios and a cost-effectiveness analysis, requires knowledge of the age-specific prevalence of the proper target for screening: unrecognized cancer that is destined to cause illness or death if not treated but that is still curable. We estimated the prevalence of undiagnosed prostate cancer by first analyzing data from eight autopsy series that had a specified age distribution; were prospective; had consecutive, unselected participants; excluded patients with prostate cancer that was suspected before death; and did serial step sectioning of the entire gland [42-49]. We then incorporated the unique morphometric and histologic analysis of McNeal and colleagues [50] to estimate the age-specific prevalence for three categories of tumors: intracapsular and smaller than 0.5 mL in volume, intracapsular and larger than 0.5 mL, and extracapsular and larger than 0.5 mL. Although any definition of clinical significance or curability of prostate cancer is arbitrary given the current limitations in primary data, intracapsular tumors larger than 0.5 mL seem to be logical primary targets for early detection [2, 14, 23].

Effectiveness of Tests for Early Detection of Prostate Cancer

We also searched for English-language articles about the early detection of prostate cancer (using the Medical Subject Heading terms prostatic neoplasms, mass screening, diagnosis, sensitivity/specificity, predictive value, and evaluation studies) involving digital rectal examination (adding the key words palpation, physical examination, and rectum) and measurement of PSA (adding the key words prostate-specific antigen; tumor markers, biological; and blood). Similar searches were done for transrectal ultrasonography and transrectal needle biopsy. Articles that provided data for fewer than 30 men were excluded. Because the number of prospective, population-based studies was limited, studies with invited volunteers were also included.

Reasons for the subsequent exclusion of articles (n = 154) are listed in Table 1; excluded studies had various selection biases [10, 51]. No relevant randomized, controlled trials of screening were identified. Several case–control studies [7, 52, 53] were identified and included.

None of the identified prospective population-based or invitational screening studies that used digital rectal examination or PSA measurement applied an adequate criterion standard (gold standard) to all patients for the diagnosis of cancer. To assess the operating characteristics of digital rectal examination or PSA measurement, we included studies of referral populations that uniformly applied a criterion standard, such as transrectal needle biopsy [54-60]. Furthermore, we included studies of consecutive men who had newly diagnosed, clinically localized cancer that was not initially diagnosed by PSA measurement and who subsequently underwent pathological staging [61-69]. Although they are potentially biased, such studies provide the best available direct information about test sensitivity [11, 51].

Calculating Likelihood Ratios for Tests for Early Detection

Given data limitations that prevent calculation of true sensitivity and specificity, we estimated likelihood ratios (sensitivity/[1 –specificity]) for digital rectal examination and PSA measurement. For these calculations, we selected only studies of tests used for primary screening that provided positive predictive values for a study population with a specified age distribution [8, 9, 70-72]. Predictive values were converted into post-test odds of disease (odds = probability/[1 –probability]), assuming perfect compliance with biopsy [10]. Estimates of age-specific prevalence from the autopsy studies were used to derive the pretest odds of cancer in the screened population, and post-test odds were divided by the pretest odds to estimate likelihood ratios.

Data Synthesis

Epidemiology of Prostate Cancer

The incidence of diagnosed prostate cancer has increased sharply in the United States since 1985 [1, 73]. This trend, as well as a trend toward detection at earlier clinical stages [6, 74], reflects several factors, including the growing use of PSA measurement as a detection method [31, 32]. The favorable stage shift has been accompanied by a small increase in age-adjusted rates of death from prostate cancer [75].

Risk Factors

Risk factors for prostate cancer have recently been reviewed [76, 77]. Age remains the most powerful risk factor. Depending on the number of first-degree relatives with prostate cancer, family history may increase a man's risk twofold to fivefold [78, 79]. The incidence in black men is as much as 1.6 times the incidence in white men [75, 80]. Whether a previous vasectomy increases a man's future risk is controversial [81-84]. A high-fat diet may also increase risk [76, 85].

Age-Specific Prevalence

Table 2 presents estimates of the prevalence of prostate tumors of various volumes, stratified by patient age and capsular status. These estimates serve to derive likelihood ratios for digital rectal examination and PSA measurement and to estimate the age-specific cost-effectiveness of early detection.

Biology Relevant to Early Detection

The disparity between the approximate 30% prevalence of histologic prostate cancer in men older than 50 years of age and the 3% lifetime risk for death from this disease shows the difficulty in distinguishing cancer that is destined to cause illness and death from cancer that is not [5, 11, 77]. This uncertainty is central to the debate about whether early detection efforts are appropriate.

The most commonly used histologic grading system for prostate cancer is the Gleason score, which assigns a grade of 1 to 5 for the two predominant areas of the pathologic specimen, yielding a Gleason sum of 2 to 10 [88]. Tumors that are assigned Gleason sums of 2 to 4 are considered well differentiated; those with sums of 5 to 6 or 7, moderately differentiated; and those with sums of 7 or 8 to 10, poorly differentiated. The two predominant clinical staging systems presently used are the Whitmore-Jewett and the TNM (tumor, node, metastasis) systems [89].

Histologic grade is the strongest prognostic factor [2, 90]. However, the grade determined from biopsy specimens may differ from that found in surgical specimens in as many as 30% of cases [91-93]. Although a tumor volume that exceeds 0.5 mL (a characteristic of approximately one fifth of cases of prostate cancer discovered at autopsy) has been considered by many experts to confer clinical significance, it remains uncertain whether tumor volume, adjusted for grade, is an independent predictor of disease-specific survival [87, 94]. With occasional exceptions [15], capsular penetration is associated with tumors whose volume exceeds 0.5 mL [50, 90, 95]. Recent studies suggest that patient age is not an independent predictor of prognosis [19, 96], but some older data indicate that it may be [97, 98]. Whether nondiploid tumors have an independently worse prognosis is unclear, although they are commonly associated with other adverse characteristics [99-101]. Given an average PSA doubling time of about 4 years, most tumors that are smaller than 0.5 mL when detected pose little risk in an average lifetime [90].

Effectiveness of Tests for Early Detection of Prostate Cancer

Digital Rectal Examination

Despite its time-honored place as part of a comprehensive physical examination, routine digital rectal examination has not been shown to reduce a patient's chance of dying of prostate cancer or to improve future quality of life. Among men 50 years of age and older, approximately 2% to 3% who receive one screening examination are found to have prostatic induration, marked asymmetry, or nodularity. Such findings increase the odds of harboring a clinically significant (>0.5 mL) intracapsular tumor by twofold. However, these findings increase the odds of having extracapsular prostate tumors by threefold to ninefold.

We reached these conclusions by using the following method. We examined 25 articles that met our selection criteria for studies of the effectiveness of primary digital rectal examination in nonreferral populations [41]. Our review found no controlled studies that prove that one-time or repeated digital rectal examination reduces the rates of morbidity or mortality attributable to prostate cancer. If liberal criteria for the definition of an abnormal test result and an aggressive strategy of follow-up biopsies are used, the digital rectal examination may have a positive predictive value of 15% and provide an overall cancer detection rate of 2%; as many as 70% of detected cases of cancer may be pathologically confined to the prostate [9]. In the absence of similar data from a population-based study, it remains uncertain whether similar results might be achieved in the primary care setting, where the detection rate of organ-confined prostate cancer might be much lower. Moreover, it seems likely that fewer examinations would be considered sufficiently suspicious to recommend systematic transrectal needle biopsy.

Studies of interrater reliability for digital rectal examination are few, but the results of those that have been done show poor reproducibility [102, 103]. Digital rectal examination is more sensitive for detecting cancer in the peripheral zone (the area of the gland that is closest to the examining finger) than in the deeper transition and central zones. However, in a recent cohort of cases of cancer detected by PSA measurement among patients who had had normal results of digital rectal examination, half of the detected cases were still found adjacent to the peripheral zone capsule [15]. Because larger tumors are more easily palpable, it is also often assumed that digital rectal examination has a low probability of detecting insignificant, low-volume tumors. However, the probability increases if a suspicious result of a digital rectal examination prompts systematic biopsies in addition to those directed to the palpable abnormality [104].

In a rare study that reported long-term outcome, Gerber and colleagues [13] found that men who had cancer that was discovered on a serial digital rectal examination seemed to have a more favorable stage shift than men who had cancer that was discovered on the initial examination. However, the former group did not have an improved survival rate [72, 105], which suggests length bias [10].

Obvious selection biases were common among the 25 studies that we reviewed. However, two population-based Scandinavian studies [106, 107], which appear least likely to be affected by such biases, provide estimates of the positive predictive value of digital rectal examination of 22% and 29%. In contrast to a recent large study of U.S. volunteers [9], fewer than one third of detected cases of cancer in one Scandinavian study [106] remained localized in the gland after pathologic staging, although more than 90% seemed localized before surgery. The proportion of clinically localized cancer that was discovered in this cohort actually decreased to 71% in a second round of screening [108].

Only two studies [9, 72], both of which involved volunteers, provided sufficient age-stratified data to estimate operating characteristics for digital rectal examination in the detection of pathologically confined and unconfined prostate cancer. Digital rectal examination in these studies had a positive predictive value of 15% [9] and 26% [72]. In the study by Richie and colleagues [9], the positive predictive value varied from 11% for men 50 to 59 years of age to 17% for men 70 to 79 years of age. Detection rates in the two studies were 2.2% [9] and 1.5% [72]; the proportion of surgically staged cancer confined to the prostate was 70% [9] and 50% [72]. On the basis of these two reports, if a patient has suspicious results on a digital rectal examination, we estimate that the odds of an intracapsular tumor larger than 0.5 mL are increased by 1.5- to 2-fold (Table 3). The odds of extracapsular tumors are increased 2.7- to 8.6-fold.

One case–control study of digital rectal examination in a large health maintenance organization reported that men with metastatic prostate cancer were no less likely to have previously had a digital rectal examination than were controls (relative risk, 0.9 [95% CI, 0.5 to 1.7]) [52]. Using estimates of available indirect evidence, three decision models have produced varying estimates of survival benefit [21, 110, 111].

Prostate-Specific Antigen

Routine measurement of PSA has not been proven to reduce the overall or disease-specific mortality rate or to improve health-related quality of life. Measurement of PSA is the most sensitive noninvasive test available for early detection of prostate cancer and may allow for earlier diagnosis of aggressive cancer [7]. If a cutoff of 4 ng/mL is used to define an abnormal result, a PSA level above this point approximately triples the odds that a man 50 years of age or older has a larger (>0.5 mL) but still localized prostate tumor. Such PSA levels also increase the odds of extracapsular tumors by three-fold to fivefold.

Measurement of PSA plays a potentially valuable role in prostate cancer detection because cancer tissue generates more PSA than does normal or hyperplastic tissue, and cancer tissue may disrupt the prostate-blood barrier [112-114]. An autopsy study showed that PSA levels are consistently elevated only when tumor volume exceeds 1.0 mL [115]. More poorly differentiated tumors appear to produce less PSA per unit of volume [61, 62]. Levels of PSA can be elevated for several weeks after acute prostatitis, acute urinary retention, transrectal needle biopsy, or prostate surgery [116, 117]. Several studies did not show clinically important elevation in PSA levels after routine digital rectal examination [118, 119].

Several PSA assays are available for clinical use [120-124]. Theoretical concerns about calibration of these assays, relating to differential detection of free compared with complexed PSA, appear to have little clinical significance [125-128]. In most PSA assays, levels of PSA greater than 4.0 ng/mL are considered abnormal.

However, PSA levels normally increase with age. More than 90% of this change reflects an age-dependent increase in prostate volume [129-131]. As a result, age-specific reference ranges for PSA levels have been proposed [129, 131, 132]. This strategy, which trades lower specificity for higher sensitivity in younger men and lower sensitivity for higher specificity in older men, has not been universally embraced [133, 134].

We examined 15 studies that met the criterion of prospectively evaluating PSA measurement as a primary early detection test among men from a general population or that invited patients who were screened [41]. We also analyzed three recent studies of serial PSA measurement [8, 133, 135-137]. No prospective controlled studies of PSA measurement have determined whether routine screening reduces morbidity and mortality rates. Although comparisons between the effectiveness of digital rectal examination and that of PSA measurement against a reference standard are lacking, available indirect evidence suggests that PSA measurement has a greater sensitivity but a lower specificity than digital rectal examination [9].

Only four studies of volunteers provide positive predictive values for PSA measurement that are not confounded by the results of digital rectal examination or transrectal ultrasonography [9, 71, 138-140]. Unfortunately, in the only two population-based studies, abnormal results on digital rectal examination or transrectal ultrasonography, not PSA measurement, were the principal criteria for biopsy [107, 141]. Similarly, criteria for biopsy in the American Cancer Society's National Prostate Cancer Detection Project (ACS-NPCDP) were determined almost exclusively on the basis of results of digital rectal examination and transrectal ultrasonography [113, 142, 143]. The positive predictive value of PSA measurement in the four studies of volunteers varied from 17% to 28%. The positive predictive value was 21% for PSA levels of 4 to 10 ng/mL and increased to 42% to 64% for PSA levels greater than 10 ng/mL. The positive predictive value appears to be independent of age, suggesting that increased cancer prevalence is balanced by decreased test specificity in older men [9]. The weighted mean detection rate for PSA measurement in these studies was 3.2%. Pathologic surgical staging was reported in two of these studies; 56% to 66% of cases of cancer were found to be confined to the prostate [9, 71]. None of the 15 studies in our primary analysis provides long-term outcome data.

Gann and colleagues [7] provided estimates of the sensitivity and specificity of PSA measurement in a relatively unselected population and applied an acceptable reference standard (long-term follow-up) to all participants. The sensitivity of a baseline PSA measurement for all cases of prostate cancer diagnosed within 4 years was 73%; an 87% sensitivity for aggressive cancer was discovered within that interval. The specificity of PSA measurement was approximately 91% (in a population with a mean age of 63 years). The sensitivity of baseline PSA measurement was 46% for all prostate cancer and 56% for aggressive cancer detected during a 10-year period. Although these results suggest that discovery of and intervention for an elevated PSA level might advance detection by an average of 5.4 years for aggressive cancer, such data do not prove that realtime PSA measurement among these men would have reduced the disease-specific mortality rate. One interpretation of the data of Gann and colleagues [7] is that PSA measurement is much more useful for detecting advanced cancer than for detecting localized cancer, which is a more "suitable target" for screening. For this subgroup of nonaggressive cancer, PSA measurement was only 37% sensitive for cancer diagnosed during the full 10-year period and 53% sensitive for cancer detected within 4 years of enrollment.

Guinan and colleagues [57] uniformly applied transrectal needle biopsy of the prostate as a criterion standard to 280 men who were hospitalized on a urology service. The detection rate for the test was 28%; however, the PSA level was greater than 4 ng/mL in only 75% of detected cases with a similar specificity of 75% [57]. Similar sensitivities may be found in studies of PSA measurements obtained before radical prostatectomy. Such designs are expected to inflate test sensitivity because results of PSA measurement may prompt diagnosis (inclusion bias) [51]. However, in seven surgical studies of consecutive men with clinically localized disease, PSA level was less than 4 ng/mL in 30% to 55% of men with pathologically organ-confined tumors [61-6365-6769, 144].

Data from the four studies of volunteers that provided age distributions allow estimation of likelihood ratios for several ranges of PSA level, as shown in Table 4. Elevations of PSA level of 4.1 to 10 ng/mL generally increase the odds that a man has an intracapsular tumor larger than 0.5 mL by 1.5- to 3-fold. Levels of PSA greater than 10 ng/mL substantially increase the odds that a man has an extracapsular tumor.

Less information is available for serial PSA measurement. The detection rates and positive predictive value decline significantly by the second year of consecutive testing [8, 135, 136]. Two studies suggest that repeated testing increases the likelihood that detected tumors will be pathologically organ-confined [8, 135]. However, one study found that the magnitude of further stage shift with serial compared with initial screening was small (74% compared with 69%) [137].

Modifications in Measurement of Prostate-Specific Antigen Levels

In addition to age-adjusted reference ranges, other variations in PSA measurement have been devised to improve the specificity of the test among men with benign prostatic hyperplasia; specificity among such men may be as low as 50% to 79% [61, 66, 109, 112, 144]. Improvements in test specificity with maintenance of comparable test sensitivity could substantially reduce the number of unnecessary prostate biopsies that are done because of false-positive test results in men with benign prostatic hyperplasia. However, in the absence of strong evidence that current therapy for clinically localized prostate cancer is effective, improvements in the accuracy of PSA measurement cannot be assumed to translate into improved life expectancy for persons who are screened. To date, none of these modifications in PSA measurement has been proven on the basis of patient outcome to be superior to using raw PSA levels in clinical practice.

Prostate-specific antigen density adjusts the raw PSA levels by gland volume derived from transrectal ultrasonography [145, 146]. The studies that suggest that PSA densities greater than 0.15 ng/mL better separate men with and without cancer than does measurement of raw PSA levels (especially those between 4.1 and 10 ng/mL) were done in select populations [145-150]. Recent retrospective studies have not shown significant improvements in specificity obtained by using PSA density and have found that adjustment from the raw PSA level reduces sensitivity [131, 151, 152]. These discrepancies may be partly explained by the poor reliability of measurements of volume. Substantial cost is added to early detection by using PSA density instead of the raw PSA level, and measuring PSA density is not practical in primary care settings [153, 154].

The PSA velocity is the rate of change of the PSA level over time. In a longitudinal study of aging, a PSA velocity greater than 0.75 ng/mL per year better discriminated cancer from benign prostatic hyperplasia than did a single PSA value [155, 156]. Specificity among men with benign prostatic hyperplasia improved from 60% to 90%, with a similar sensitivity for cancer (72% compared with 78%). However, intrapatient variation of PSA levels greatly limits the usefulness of PSA velocity until a patient has had at least three PSA levels measured for a period of several years [156-158]. Moreover, in practice, an initially abnormal PSA level in such men would probably prompt one or more prostate biopsies earlier than would theoretically be warranted. Thus, although the belief that using PSA velocity would reduce the proportion of men receiving unnecessary biopsies is attractive, that result has not been shown in the primary care setting. A recent study determined that for men whose initial PSA level was less than 4.0 ng/mL, a PSA velocity of 0.75 ng/mL per year optimized the post-test odds ratio for prostate cancer (odds ratio, 7.2 [CI, 4.2 to 11.5]) [137].

Measurement of free and complexed PSA may ultimately allow better discrimination between men with prostate cancer and men with benign prostatic hyperplasia [159]. Men with prostate cancer appear to have less circulating free PSA relative to PSA circulating in complexes with such macromolecules as ₁-antichymotrypsin [160]. Available data are limited [159]. Further prospective studies are needed to define optimal cutoff points.

Transrectal Ultrasonography

No direct or convincing indirect evidence shows that use of transrectal ultrasonography as a screening test improves disease-specific survival rates. The limited sensitivity and specificity of transrectal ultrasonography as a primary screening tool are well documented [41].

Transrectal ultrasonography uses biplanar probes with 7- or 7.5-MHz transducers to identify areas in the prostate that are suspicious for cancer. Although the criteria for abnormal test results vary, focal hypoechogenicity has been investigated most extensively [113]. Transrectal ultrasonography appears to be more sensitive for more peripheral and larger tumors [161-163]. As it does with PSA measurement, benign prostatic hyperplasia reduces the specificity of transrectal ultrasonography [59, 164, 165]. Even for referral-based studies among patients with a high prevalence of cancer, the positive predictive value of prostate ultrasonography is 6% or less, provided the digital rectal examination result and PSA level are normal [59, 166, 167]. Transrectal ultrasonography is no longer widely promoted as a primary screening test, but it is used to investigate abnormalities on digital rectal examination and PSA measurement and to guide biopsies [113]. Moreover, given the limited sensitivity of transrectal ultrasonography, many authorities now recommend systematic biopsies when either digital rectal examination results or the PSA level is suspicious, regardless of ultrasonographic findings [23].

Transrectal Needle Biopsy

We found no evidence that routine transrectal needle biopsy reduces a man's likelihood of dying of prostate cancer. Although such biopsies may be more accurate than available noninvasive tests, their viability as a screening test is greatly limited by cost, potential illness, and patient reluctance.

Transrectal needle biopsy with ultrasonographic guidance using a biopsy "gun" has generally replaced fine-needle aspiration and core biopsy techniques to obtain prostate tissue for histologic diagnosis. Biopsies can be directed at palpable or ultrasonographic abnormalities or to obtain systematic sampling of the entire gland [23].

The true sensitivity of transrectal needle biopsy is unknown because no evaluations have used an independent reference standard, such as long-term follow-up. Although this test is often considered the reference standard for diagnosing prostate cancer, it clearly has an imperfect sensitivity. For example, in one study of men who had needle biopsies if the PSA level was mildly elevated (4.4 to 9 ng/mL), approximately 25% who initially had negative results were found to have cancer on subsequent biopsies [168].

Specificity is also problematic because transrectal needle biopsy can detect low-volume tumors that may not threaten a man's future health. The risk for finding such low-volume tumors at biopsy is unknown for men in the general population. Terris and colleagues [169], using a set of six systematic biopsies, estimated that the likelihood of finding such incidental tumors was 4% in a referral population. Although the definition of low-volume tumors varies across studies, the reported proportion of such tumors detected through PSA measurement followed by systematic transrectal needle biopsy ranges from 11% to 26% [15, 109, 137]. The limitations in the sensitivity and specificity of transrectal needle biopsy have implications for all published positive predictive values for digital rectal examination, PSA measurement, and transrectal ultrasonography as early detection tests.

Transrectal needle biopsy is mildly uncomfortable and causes minor, self-limited complications (principally infections or bleeding) in as many as 40% of patients [54, 59, 167, 170-173]. Hospitalization for any biopsy-related complication has been reported in less than 1% of patients [173].

Combination of Digital Rectal Examination and Prostate-Specific Antigen Screening Followed by Transrectal Ultrasonography and Transrectal Needle Biopsy

Evidence from uncontrolled studies suggests that a combination of digital rectal examination and PSA measurement improves the overall rate of prostate cancer detection when compared with the detection rate of either test alone [9, 138-140]. This combination strategy might provide a 4% yield, possibly doubling the proportion of pathologically localized cancer compared with the fraction of such tumors discovered without screening [174]. Despite this apparent stage shift, however, no proof currently exists to show that this strategy reduces disease-specific morbidity and mortality rates.

Office-based early detection with digital rectal examination and PSA measurement, using transrectal ultrasonography to direct needle biopsy if either test result is abnormal, is the strategy that is currently favored by most proponents of screening. We found only three studies of early detection among volunteers that defined the detection rate and positive predictive value of this strategy [9, 138-140]. In these studies, the proportion of patients with abnormal results on either digital rectal examination or PSA measurement was 18% to 26%; overall cancer detection rates were 3.5% to 4%. If either test result was abnormal, the positive predictive value varied from 15% to 21%. If both test results were abnormal, positive predictive values varied from 38% to 50%. Digital rectal examination and PSA measurement each detected cancer that was not identified by the other test.

In one of these studies [140], digital rectal examination detected only two thirds as many cases of organ-confined cancer as were detected by PSA measurement, even though 68% of cases of cancer that were found with each test were pathologically localized [140]. The incremental gain in cancer detected by adding PSA measurement to digital examination came at a cost of recommending biopsy in 26% of the cohort instead of 15%.

If this combined strategy were routinely adopted, approximately one quarter of screened men older than 50 years of age would be subject to the cost and risk associated with transrectal needle biopsy (15% of men 50 to 59 years of age, 28% of men 60 to 69 years of age, and 40% of men 70 to 79 years of age). It is important to consider that the use of volunteers in screening studies may increase the prevalence of prostate cancer compared with the prevalence among unselected men in the general population. In fact, Oesterling and colleagues [175], using routine transrectal ultrasonography, digital rectal examination, PSA measurement, and systematic biopsies for men with abnormal results on any test, found prostate cancer in only 1% of men randomly selected from Olmsted County, Minnesota.

Effect of Benign Prostatic Hyperplasia on Test Effectiveness

Benign prostatic hyperplasia appears to have an important effect on the operating characteristics of tests for prostate cancer. More than 50% of men older than 50 years of age have histologic benign prostatic hyperplasia, and about one third have lower urinary tract symptoms [176, 177]. In men with clinical and histologic evidence of benign prostatic hyperplasia, the specificity of PSA measurement has been found to decrease to 50% to 79% [116, 144]. The effect of prostatic hyperplasia on the sensitivity and specificity of transrectal ultrasonography and digital rectal examination is less clear but may also be an issue [59, 60, 164, 178]. Reduced test specificity, especially for PSA measurement, in men with benign prostatic hyperplasia leads to high rates of prostatic biopsy and its attendant illness and costs. However, because the prevalence of benign prostatic hyperplasia is strongly related to age, another likely effect of lower test specificity is an increase in the discovery of incidental tumors (<0.5 mL in volume) through the follow-up of positive test results with systematic prostatic biopsies [15, 109, 163].

Men with symptoms of prostatism do not appear to be at higher risk for having unrecognized prostate cancer [142]. One of the large studies of volunteers [140], found that after controlling for age, the presence of symptoms (including bone pain and hematuria) independently reduced the chance of detecting cancer by digital rectal examination and PSA measurement. If the more ominous symptoms had been excluded, lower urinary tract symptoms alone might have led to an even lower cancer detection rate. A recent consecutive series of PSA-detected cancer suggested that, after controlling for other covariates, asymptomatic men whose condition was diagnosed had more extensive disease [15].

Given that men with lower urinary tract symptoms do not seem to be at greater risk for prostate cancer, little evidence supports the traditional distinction between diagnosis and screening when dealing with men with such symptoms [34, 35, 179, 180].

Conclusions

Undiagnosed prostate cancer is highly prevalent, especially among older men [42-49]. Although many of these cancers may be considered incidental, evidence suggests that consideration of screening is warranted because earlier diagnosis of clinically significant cancers often has the potential to improve outcome [10, 12]. In fact, prostate cancer is so common among older men that selecting subpopulations for screening on the basis of risk factors other than age, such as race or family history, would not be necessary if screening and treatment strategies that favorably affect outcome were available.

However, evidence also indicates that available tests for early detection have limited specificity, particularly among older men who have benign prostatic hyperplasia. Therefore, the combined use of digital rectal examination and the currently available assays for PSA results in as many as 15% of men in their 50s and 40% of men in their 70s requiring further invasive evaluation with biopsy. Positive predictive values are 15% to 21%, depending on age.

On the basis of available evidence (Table 3), suspicious results on a digital rectal examination modestly increase the odds of intracapsular tumors (>0.5 mL) and considerably increase the odds of extracapsular tumors. On the other hand, if the results of digital rectal examination are not suspicious, the patient's pretest odds are not affected and little reassurance is provided about the absence of intracapsular or extracapsular tumors. Results of PSA measurement have a greater influence on the probability of cancer, particularly extracapsular tumors (Table 4). However, given that none of the studies that we reviewed did biopsies on all patients with normal results on digital rectal examination and PSA levels, these estimated likelihood ratios remain uncertain; they are accurate only if the data in Table 2 fairly represent the pretest probability of cancer in these studies.

Such refinements in PSA measurement as measurement of free and complexed PSA levels hold promise of enhanced specificity with similar sensitivity, thereby reducing the proportion of men who would have unnecessary biopsies [159]. However, such improvements in test accuracy would not necessarily lead to improved long-term disease-specific health outcome in the absence of definitive evidence that aggressive therapy is effective. Such proof is not yet available [25-28].

At the risk of prompting biopsies on a large proportion of men who are screened, digital rectal examination and PSA measurement can be used to further reduce the post-test odds of cancer among men with normal results; however, the degree of such reduction is uncertain. An unknown but probably substantial proportion of the cases of cancer that are discovered by using this dual test strategy will be discovered through serendipity, that is, as a result of doing biopsies on a large proportion of screened patients. This phenomenon has been noted to be an important factor in fecal occult blood screening for colorectal cancer [181].

The second part of this paper will focus on whether widespread early detection should be recommended on the basis of the ability of current screening methods to increase or decrease patients' odds of prostate cancer.

Drs. Barry and Mulley: Medical Practices Evaluation Center, Massachusetts General Hospital, 50 Staniford Street, Boston, MA 02114.

Dr. Fleming: Health Outcomes Associates, 602 Northeast 134th Street, Vancouver, WA 98685.