Screening for prostate cancer could be done by means of PSA testing, but whether this would reduce mortality from the disease was not known. No randomised trials on the question had been undertaken when the study began in 1988.

Benefit from cancer screening can only be validly assessed by randomised trials which compare mortality rates over the same time period among people randomly allocated to early detection and treatment or to usual care. This is necessary because studies of cancer screening can show an apparent benefit even if screening is completely ineffective, because:

• Survival may appear better among screen-detected cases simply because the diagnosis was made earlier (“lead-time bias”) — if the time of death is not altered, cases detected by screening may be getting more “disease time” rather than more life time;

• Screening preferentially detects slower, less aggressive cancers which have a longer preclinical (but screen-detectable) phase — fast-growing cancers are more likely to be missed by screening because they are more likely to develop and progress in the interval between screenings, giving the appearance of better outcomes among screened people (“length-time bias”); and

• People who attend for screening are generally healthier and more health conscious, so their risk of death from any cause is lower.1

The major strength of this study is that it was designed as a randomised trial and therefore could potentially validly answer the question. Important methodological features of good randomised trials are:

• high-quality randomisation with allocation concealment;

• blinded assessment of outcomes;

• high follow-up rates; and

• analysis by intention to treat.2

The major (and fatal) weakness of this study is that the primary analysis was not as a randomised trial (ie, by intention to screen). The authors compared the prostate cancer mortality rate among men who were screened (15 deaths/100 000 man-years) with that among men who were not screened (48.7 deaths/100 000 man-years). However, they included in the screened group men who had not been allocated to screening, but who sought screening anyway. They also included in the non-screened group men who were allocated to screening but did not attend. They thus destroyed their randomisation, reducing the study to an observational one. Their analysis yields a statistically significant, and apparently impressive, 69% relative reduction in prostate cancer mortality. However, this analysis is almost certainly biased (by one or more of the biases listed above).

The authors also report what they originally set out to do — an intention-to-screen analysis which preserves the randomisation — as a secondary analysis. They report a 6% relative reduction in prostate cancer mortality (relative risk, 0.94) but provide no significance test or confidence interval. We have estimated these from the article’s data and found the result to be non-significant, with a 95% confidence interval of about 0.71–1.25. This means the true effect could be anywhere between about a 30% reduction to a 25% increase in prostate cancer mortality. The null result may be attributed to the poor participation in screening in the study. In the group invited to screening, only 23.1% actually attended for screening. In the control group, 6.5% of uninvited men attended for screening. While these “drop-out” and “drop-in” rates do not invalidate the study if it is analysed correctly, they inevitably cause a substantial attenuation of the intervention if it is effective. It is possible to adjust for this attenuation (due to poor participation) by methods previously described.3,4 However, in the case of this trial, adjustment will not be useful, as the adjusted estimate will also be non-significant, with a wide confidence interval.

None — this trial contributes no new valid information.