Four-hundredfold overestimation of biopsy mortality

Authors


Boniol et al. [1] report on prostate biopsy mortality and the number of life years gained and lost in prostate cancer screening. They conclude ‘screening for prostate cancer should be discouraged’ on the grounds that it leads to ‘more harm than good’. This conclusion is importantly based on the estimate that ‘two deaths [occur] per 1000 [prostate] biopsies’.

There are numerous reasons to believe that this estimate is inaccurate in the extreme. In the largest study known to date from a screening setting (the European Randomized Study of Screening for Prostate Cancer), there were zero deaths occurring as a complication of prostate biopsy upon careful review of causes of death among 11 721 biopsied men [2]. A second estimate comes from a large Japanese study that found only a single sepsis-related death among 212 065 biopsied men [3]. The upper bounds of 95% CI for these two estimates are ∼3 per 10 000 and ∼3 per 100 000, respectively. In other words, the estimate used by Boniol et al. is >400 times greater than that from the largest study on prostate biopsy mortality, and completely inconsistent with the most conservative interpretation of risk in terms of CIs.

Consideration of population statistics provides further reason to believe that the two per 1000 estimate is entirely implausible. Every year, ∼1 000 000 men in the USA undergo a prostate biopsy. If there were two deaths per 1000 biopsies, then we would expect 2000 US men to succumb to prostate biopsy annually. On intuitive grounds, we assume that these many deaths would have been noticed; however, when we reviewed the entire world literature back to the 1960s on this topic, only a handful of case reports of fatal sepsis have been described [2].

It is worth wondering why the estimate used by Boniol et al. is so far from reliable estimates. The paper they cite [4] does not give any cause-specific data. To find that two per 1000 men die within 120 days of prostate biopsy cannot be used to claim that two per 1000 men die because of prostate biopsy. This is a straightforward error of causal attribution.

We fully agree that biopsies should be performed with caution, weighing benefits and risk. Proper selection of patients to biopsy is crucial and we do not wish to downplay the risk of complications after prostate biopsy. That said, the risk of death is extremely slight.

We also agree that ‘the modelling approach to calculating the number of years of life lost and gained by screening is a good method for evaluating the overall impact’. Indeed, we ourselves have used a modelling approach [5]; however, the estimates entered into these models must be reliable. Boniol et al. might well conclude that prostate biopsy mortality plays a ‘critical role … in the number of years of life gained and lost’ for screening, given that they overestimate its incidence four-hundredfold. On the basis of more reliable data on prostate biopsy mortality [2, 3, 6, 7], we are confident that it is a minor factor, and that, as such, we should depend on published estimates of the effects of screening [5, 8, 9].

Furthermore, the authors dramatically overestimate the treatment-related mortality rate for radical prostatectomy and radiotherapy, which has been shown to be closer to 0.1% in contemporary series [10-12]. Incorporating appropriate estimates in Boniol et al.'s main analysis (biopsy mortality 1/200 000 and treatment mortality 1/1000), we recalculate life years gained from −3.6 as theauthors report to +8.6. We therefore assert that the conclusion of the paper is invalid.

Ancillary