Screening for prostate cancer is controversial. Although screening with prostate-specific antigen (PSA) may identify cancer at an early stage, questions remain regarding the impact of screening on mortality and other important clinical outcomes.

Djulbegovic and colleagues performed a systematic review and meta-analysis of randomised controlled trials comparing PSA-based screening with or without digital rectal examination versus no screening including nearly 400 000 asymptomatic men. Six trials met the eligibility criteria and were included in analyses on overall and disease-specific mortality and, if possible, in an analysis on clinical stage at the time of diagnosis.

The authors found that screening was associated with an increased probability of receiving a diagnosis of prostate cancer (RR 1.46, 95% CI 1.21 to 1.77, p<0.001) and stage I prostate cancer (RR 1.95, 95% CI 1.22 to 3.13, p=0.005). There was no significant effect of screening on death from prostate cancer (RR 0.88, 0.71 to 1.09, p=0.25) or overall mortality (RR 0.99, 0.97 to 1.01, p=0.44), and all trials included in this meta-analysis had one or more substantial methodological limitations.

There is no doubt that screening for prostate cancer is associated with a considerable risk of overdiagnosis and overtreatment and that the harms versus the benefits of screening still remains to be elucidated.1

This systematic review included the Quebec study2 and the Norrköping study3 even though they both have substantial methodological weaknesses in design, conduct and analysis.

When interpreting the results of a meta-analysis, it is important to consider the degree to which the studies answer the same clinical question and that their results agree.

With respect to the clinical question, the Norrköping study was designed as a feasibility trial to study the acceptability, organisation and consequences of a screening program.3 It is inevitable that effects vary among studies since inclusion, tests applied and their algorithm are never completely identical. Whether the variability in the study results go beyond chance is called heterogeneity and can be statistically tested. This has been done in the current meta-analysis and resulted in a significant test results when comparing the effect of screening on prostate cancer diagnosis and in a considerable heterogeneity with respect to prostate cancer mortality. The results of testing for heterogeneity on prostate cancer mortality were a p value of 0.06 and an I² of 55%. The I² statistic reflects the inconsistency of the study estimates and ranges from 0% to 100%. I² values > 50% indicate a moderate-to-severe level of inconsistency, and results of such an analysis should be interpreted with caution.4

Only two studies in the systematic review presented an upfront power calculation.1 5 The power of the other four studies is unknown. Furthermore, in the Quebec study, only 23.6% of participants in the intervention group were screened and 7.4% in the control group were screened.2 As a result, that study cannot be interpreted as a randomised trial of screening versus no screening.

Similar observations hold for the PLCO trial6 where (based on only 7 years of follow-up) the effect on prostate cancer mortality was the result of screening versus somewhat less screening and although contamination was rated using the GRADE approach, already 44% of men randomised to the PLCO trial were prescreened. In addition, the sample size was limited and compromised the power of the study to detect any difference. In our opinion the meta-analysis by Djulbegovic and colleagues has used eligibility criteria for inclusion of studies that did not meet sufficient quality criteria. Therefore, their meta-analysis could be regarded as seriously flawed when including studies of insufficient quality and/or insufficient follow-up.

The results coming from two better designed and conducted prostate cancer screening trials and having sufficient follow-up demonstrated that organised PSA screening can result in a relevant reduction in prostate cancer mortality.1 5 Lower quality trials or trials designed with a different goal should not have been combined with these studies in the meta-analysis. Instead, evidence on which to make recommendations regarding prostate cancer screening should rely on high-quality randomised controlled trials. The review by Djulbegovic and colleagues does point out that the quality of studies of prostate cancer screening has been limited though some higher-quality studies do exist. In that context, we agree with the conclusion of Djulbegovic and colleagues stating that routine use of screening for prostate cancer is not yet supported. However, we agree because, currently, the available screening and diagnostic tests are not optimal in selectively identifying those men who will benefit of early detection. This results in unnecessary testing and diagnosis and treatment of prostate cancers that, without screening, would never have caused any harm. The balance between benefit and harm urgently needs to be established. To reach this goal we need, besides studies on new biomarkers and more accurate methods for risk stratification, longer follow-up of the ongoing high-quality screening studies and studies estimating the possible harm measured as quality-adjusted life years gained by screening together with cost-benefit and cost-effectiveness analyses.