Randomized, Controlled Trials of PSA Screening for Prostate Cancer
Unfortunately, it is impossible to tease out from observational evidence the relative effects on prostate cancer mortality of new treatments, early detection with PSA screening, chance, and misclassification bias. Thus, randomized controlled trials are critical to evaluation of the efficacy of PSA screening. Since the introduction of the PSA assay as a screening test in the 1980s, five unique randomized controlled trials of screening for prostate cancer have been reported in the literature.
Three early trials
One of the earliest of these trials, by Labrie et al, identified men aged 45 to 80 years from the electoral rolls of the Quebec City metropolitan area and randomly assigned 31,133 to PSA and digital rectal examination screening and 15,353 to observation. Of the men in the intervention arm, 7,348 (24%) actually underwent screening; 23,785 (76%) did not. Of the men in the observation arm, 1,122 (7%) also underwent screening during the study. Because of the high cross-over rate, the authors decided to analyze the data according to whether the participant actually received screening or not, deviating from the standard “as-randomized” analysis. When viewed this way, the relative risk for death from prostate cancer in those screened compared with the risk in those not screened was 0.39 (95% CI, 0.21-0.71) (11 prostate cancer deaths among screened men vs 217 deaths among unscreened men). A key limitation to this approach is that it breaks randomization; thus, these results essentially represent an observational study. Importantly, the authors did not report information on the baseline demographic characteristics of screened vs unscreened participants, making it impossible to gauge the comparability of these populations—nor did they adjust for potential confounders with their analysis. If the data are analyzed using an intention-to-screen approach, no difference in prostate cancer mortality is observed between the two arms (relative risk [RR], 1.09; 95% CI, 0.82 to 1.43) (153 deaths among men randomly assigned to screening vs 75 deaths among those randomly assigned to observation).
In 1987, investigators in Sweden used the national population register to assign every sixth man living in Norrköping aged 50 to 69 years (n = 1,494) to screening with digital rectal exam—and later, with PSA in addition to DRE—every three years; the remaining 7,532 men, who were not contacted about their participation in the trial, were treated as controls. This quasi-randomized pilot trial was not powered to detect a statistically significant difference in prostate cancer mortality. There were 43 screening-detected cases of prostate cancer in the intervention arm (3%), along with 42 interval cancers, and there were 292 prostate cancer diagnoses (4%) in the control group. The authors found that more tumors in the screened arm were of a lower grade at the time of diagnosis compared with those in the control arm; however—unsurprisingly, given the limitations of the study design—no differences in prostate cancer−specific or overall mortality were seen.
A study of one-time prostate cancer screening using a combination of PSA, digital rectal exam, and transrectal ultrasonography was initiated in Stockholm, Sweden in 1988. Investigators invited for screening 2,400 randomly chosen men who were 55 to 70 years of age and living in the catchment area of Stockholm South Hospital; 1,796 (74%) participated, and 24,804 men from the remaining source population, who were not contacted about their participation in the trial, served as controls. Follow-up was 15 years (median, 12.9 years). The authors found no statistically significant difference in the risk of prostate cancer mortality between the screen-invited arm and control arm (incidence rate ratio, 1.10; 95% CI, 0.83–1.46; 53 prostate cancer deaths in the screen-invited group [26% of diagnosed] vs 506 deaths in the control group [28% of diagnosed]). There are several important limitations to this trial. The PSA threshold for biopsy was >10 ng/ml, and the screening approach was a simultaneous employment of PSA, digital rectal exam, and ultrasound, making applicability to current practice challenging. The authors also note that the treatments employed in this trial likely do not represent current standard of practice. Finally, although causes of death were assigned by a review committee and verified against a national Cause of Death Register, it is unclear whether the reviewers were blinded to study arm allocation, raising the possibility of misattribution bias.
Although each of the above trials provide useful context regarding prostate cancer screening methods and programs, all have sufficient methodological limitations that the ultimate efficacy of PSA testing for reducing prostate cancer mortality cannot be definitely demonstrated or refuted by them.
The PLCO trial
In 2009, interim results from two very large randomized controlled trials were reported.
The Prostate, Lung, Colorectal, and Ovarian (PLCO) cancer screening trial randomly assigned 76,693 men aged 55 to 74 years at 10 US study centers to annual screening with PSA testing for 6 years and with digital rectal examination for 4 years or to usual care (which could potentially include screening); the period covered by the study was 1993 to 2001. Men were excluded from participating if they had received more than one PSA test in the three years prior to randomization. A PSA level > 4 ng/ml was considered positive. Subjects’ personal health care providers received results and determined diagnostic follow-up and treatment for positive findings. Overall rates of compliance with screening in the intervention group were 85% for PSA testing and 86% for digital rectal exam. In the control arm, PSA testing rates increased from a baseline of 40% to 52% at year six; digital rectal exam rates were 41% to 46%.
After seven years of follow-up (98% of men with known vital status), there was a higher incidence of prostate cancer in the screened arm than in the control arm (incidence rate ratio, 1.22, 95% CI, 1.16-1.29). However, there was no statistically significant difference in the prostate cancer mortality rates between the screened and control arms, with a trend towards an increased number of deaths in the screened arm (50 deaths in the screening group vs 44 in the control group; mortality rate ratio, 1.13; 95% CI, 0.75-1.70). At 10 years of follow-up (complete for 67% of men), the findings remained essentially the same: the excess in prostate cancer cases persisted in the screened arm (incidence rate ratio, 1.17; 95% CI, 1.11-1.22), but mortality was roughly equivalent (92 deaths in the screened arm vs 82 in the control arm; rate ratio, 1.11; 95% CI, 0.83-1.50). The number of subjects with advanced tumors (stage III/IV) was also similar in the two groups (122 in the screened arm vs 135 in the control arm). Subgroup analyses stratified by history of PSA testing prior to study entry did not reveal differential effects on prostate cancer mortality rates. Overall mortality rates were also the same in the two study arms.
The ERSPC trial
The European Randomized Study of Screening for Prostate Cancer (ERSPC) is a multi-national trial of eight European countries (previously nine—Portugal dropped out in 2000 without contributing data) begun in the 1990s; 182,160 men between the ages of 50 and 74 years were randomly assigned to either a group offered PSA testing or a control group not offered screening. The study included a predefined “core” group of 162,387 men aged 55 to 69 years. Randomization occurred prior to consent in three countries and after consent in four. Of note, data from France was not included in this interim analysis because that country first began participating in 2001, limiting length of follow-up. Most countries utilized PSA testing alone, although cut-off values varied by country: most considered ≥3 ng/ml as positive, but Finland used a cutoff of ≥4 ng/ml with ancillary testing (digital rectal exam or free:total PSA ratio, depending on year) for values between 3.0 and 3.9 ng/ml, Italy had men with a PSA value between 2.5 and 3.9 ng/ml undergo digital rectal examination and transrectal ultrasonography, and until 1995, Belgium used a PSA cutoff of 10 ng/ml. Belgian and Dutch centers used a combination of digital rectal exam, ultrasound, and PSA testing as the primary screening approach until 1997. There were also variations in the number of core biopsies performed for a positive screening test, as well as differences in the frequency of screening: most of the participating countries tested every 4 years, but Sweden screened every 2 years and Belgium had one 7-year interval. Compliance rates varied across countries, but overall, 82.2% of men in the screening group received at least one test. Although the study was designed to have sufficient power to account for a 20% contamination rate, no specific information was provided about actual rates of screening in the control arm for all but one study center.
After an average of 8.8 years of follow up (median, 9 years), no statistically significant reduction in prostate cancer mortality was observed in the overall study population (rate ratio, 0.85; 95% CI, 0.73-1.00). However, analysis of the prespecified “core” group (men aged 55 to 69 years) revealed a statistically significant 20% relative reduction in prostate cancer mortality (214 deaths in the screened group vs 326 deaths in the controls; rate ratio, 0.80; 95% CI, 0.65-0.98)—or an absolute reduction of 0.71 prostate cancer deaths per 1,000 men, with differences between the screened and control groups first beginning to emerge after about eight years. Additionally, an exploratory analysis of mortality according to age group revealed a statistically significant reduction in men aged 65 to 69 years (rate ratio, 0.74; 95% CI, 0.56-0.99) but a possible trend towards increased prostate cancer–specific deaths for those aged 50 to 54 years and 70 to 74 years (rate ratios, 1.47 [95% CI, 0.41-5.19] and 1.26 [95% CI, 0.80-1.99], respectively). An increased incidence of prostate cancer was observed in the screened arm compared with controls: 5,990 cases vs 4,307 cases, or a net increase in cumulative incidence of 34 cases per 1,000 men. These findings translate into 1,410 men aged 55 to 69 years needing to be screened and 48 additional prostate cancers treated to prevent or delay one prostate cancer death.
In 2010, Hugosson et al separately reported findings from Göteborg in Sweden, one of the participating countries in the ERSPC trial (data from participants born between 1930 and 1939 were included in the pooled ERSPC data). A total of 20,000 men aged 50 to 64 years were randomized to PSA screening or a control group not offered screening; screening was every 2 years, and the PSA biopsy threshold was 3.0 ng/ml between 1995 and 1998 and 2.5 ng/dL thereafter. Median follow-up was 14 years (complete for 78% of men). As with the larger ERSPC trial, more prostate cancers were diagnosed in the screening arm than in the control arm (11.4% of those screened vs 7.2% of controls). The authors report a statistically significant relative reduction in prostate cancer deaths (rate ratio, 0.56; 95% CI, 0.39-0.82). Overall, there were 44 prostate cancer deaths in the screened group (0.44%) vs 78 deaths in the control group (0.78%)—an absolute risk reduction of 0.34%. Thus, to avert one prostate cancer death, the corresponding number that would need to be invited to be screened would be 293, and the number who would need to be diagnosed and potentially treated (some men chose active surveillance) would be 12. No difference in overall mortality rates between the screened and control arms was found.
Making sense of the differences between the ERSPC and PLCO results
Of the available randomized trials, the ERSPC and PLCO studies represent the current best evidence regarding the efficacy of PSA screening for prostate cancer—although both have limitations. The divergent findings of the two trials have, however, contributed further to the long-standing debate surrounding the widespread population use of PSA screening (a “controversy that refuses to die,” as the editorial accompanying the publication of these two landmark studies termed it). There are several potential explanations for the differences in the findings that are important to consider.
First, the observed disparity between the results of the two trials may eventually resolve. The confidence intervals for the mortality estimates (0.75-1.70 in the PLCO trial and 0.73-1.00 in the ERSPC trial) make the results of both studies potentially consistent with either a modest mortality benefit or no effect. The Göteborg subtrial reported a lower relative risk, and therefore a lower confidence interval (0.39-0.82); the bulk of the results are embedded in the overall ERSPC study. A recent meta-analysis of all five randomized controlled trials reveals a confidence interval of the magnitude of the effect of PSA screening on prostate cancer mortality rates that is similar to those of the two largest trials (risk ratio, 0.88; 95% CI, 0.71-1.09).
Length of follow-up is an important factor. The PLCO trial had essentially complete follow-up for 7 years and data for nearly three-quarters of the men at 10 years, and it demonstrated no statistically significant effect on mortality to date. However, given that the lead time for PSA screening and prostate cancer has been estimated as being as long as 15 years, it is possible that a small benefit might emerge with continued follow-up. The level of screening in the control arm of the PLCO trial is also a consideration. However, the usual statistical methods that are used to adjust for this contamination in the control arm would increase the estimate of harm, since the relative risk was >1.0, and statistical adjustments for the dilution of effect size attributable to contamination move the effect size estimate away from 1.0. Similarly, pre-enrollment PSA screening may also have had an impact on the observable mortality rates. Approximately 44% of men in the PLCO trial had a PSA test before entering the study. This likely reduced the number of prevalent tumors remaining to be detected, lowering the trial’s power to detect a modest mortality difference.
Conversely, the ERSPC trial did show a mortality reduction for its “core” subgroup emerging at about seven to eight years. However, it is important to consider that its mean follow-up period is only slightly longer than the point at which the mortality benefit begins to emerge, and that several of the participating centers have not yet provided data for the period at which the major effect seems to occur. As such, the mortality estimate may change with further analyses, either qualitatively or quantitatively. Contamination rates in the control arm of the ERSPC trial were only captured for the Rotterdam site, but were extrapolated to be about 20% for the overall trial. The trial investigators performed a separate statistical analysis in an attempt to adjust for noncompliance and contamination, and they reported an adjusted relative prostate cancer mortality reduction of about 30%.
The trend towards increased mortality with screening that was seen in the PLCO trial remains an important difference between the two studies. Two factors that might have played a role in this finding include a higher frequency of testing and more interventional management of diagnosed disease in the US trial than in the European trial. In the ERSPC trial, men were most often screened at 4-year intervals, whereas in the PLCO trial, men were screened annually. More frequent screening may increase the probability of overdiagnosis. For example, ERSPC investigators compared the 10-year cumulative incidence of all diagnosed prostate cancers vs interval cancers of two countries that employed different frequencies of screening (the Dutch Rotterdam center, with screening every 4 years, and the Swedish Göteborg center, with screening every 2 years). If more frequent screening successfully increases the proportion of late-stage cancers “pulled out of the future” and prevented from progressing, an associated drop in the number of interval cancers (as symptomatic, later-stage lesions) should be observed. In comparing the data from the two ERPSC screening centers, investigators found that although a greater total number of prostate cancers were diagnosed in the more frequently screened population (1,118 cancers at the Swedish Göteborg center [8.41%] vs 552 cancers at the Dutch Rotterdam center [13.14%]; P < .001), the incidence of aggressive interval cancers was essentially equivalent in the two groups (15 cancers [0.11%] vs 5 cancers [0.12%]; P = .72). In the screening arms, a lower percentage of men employed watchful waiting as their treatment strategy in the PLCO trial than in the ERSPC trial (11% vs 18.6%), whereas a higher percentage of men in the PLCO trial underwent radical prostatectomy. Taken together, these differences may have resulted in a relative increase in the number of indolent prostate cancers diagnosed in the PLCO trial, along with a greater mortality risk from more invasive treatments, resulting in a net harm.
There is one additional difference in design that could have led to the observed difference in the outcomes of the two studies. In several ERSPC centers, control subjects were unaware that they were participating in a trial, and when diagnosed, they received treatment at their usual place of care. However, the participants in the screening arm, who were diagnosed at the major screening centers, tended to receive treatment from high-volume tertiary referral centers. This led to differences in treatment approaches, delivered by urologists of potentially different expertise.[31,32] By contrast, in the PLCO trial, all men knew they were participating in a trial, and analyses confirmed the same stage-specific therapy for diagnosed prostate cancer in each study arm.