We analyzed population-based data on prostate cancer incidence and found that an important majority of excess cases diagnosed in the first few years after the introduction of PSA testing occurred in men older than 60. We then estimated the risk at age 60 for prostate biopsy-detectable cancer in two different PSA-screened population-based cohorts and compared this risk to that of long-term metastasis and death from prostate cancer in an unscreened population. The risk that a biopsy detectable cancer would not lead to cancer-related morbidity or mortality increased exponentially as PSA fell below conventional biopsy thresholds.
There are two major clinical implications of our findings. First, more selective screening of men older than 60 is justified. We found that a clear majority of excess cases are diagnosed in men older than 60, yet there is randomized evidence from the European trial  that screening reduces mortality for men in their 60s. As such, it would appear unwise to recommend that screening be terminated at age 60 for all men. An alternative would be to restrict screening to men with PSA levels above 1 ng/ml (WHO calibration), close to the median. Men with PSA <1 ng/ml – approximately 50% of the population – can be told that if they continue to be screened, any prostate cancer thereby detected is unlikely to harm them and that if they elect to be treated they will likely be subjecting themselves to overtreatment. The ERSPC found evidence that screening is not of benefit for men who start at age 70 or older, with the lower bound of the 95% CI excluding the central estimate for risk reduction for men younger than 70 ; a subsequent modeling study reported that any decreases in mortality associated with screening men older than 70 were offset by overdiagnosis . Restricting screening in men in their 70s to a small group with excellent health and above average PSA would likely reduce overdiagnosis considerably without any substantive effect on mortality.
The second implication of our findings is that it becomes hard to justify prostate biopsy in men with PSA below typical thresholds for biopsy, such as 3 or 4 ng/ml. It has been estimated that about one in seven diagnoses occur in men with PSA below 4 ng/ml , constituting about 35,000 cases a year. Although use of a lower PSA threshold may be justifiable in younger men, such as those below 50 years old, in clinical practice, older men with low PSA are often subject to biopsy because of a positive digital rectal exam, rapid increase in PSA, low ratio of free-to-total PSA or family history [11, 12]. Such indications would only be justified if they dramatically raised the risk of aggressive cancer. There is no clear evidence that this is the case.
Several lines of evidence from the literature support our overall findings. First, the strong association between overdiagnosis and age is supported by consideration of life expectancy data. In the studies with appropriately long follow-up, lead time has been estimated to average around 12 years [15, 16]. For instance, in the Malmö cohort used for this paper, the mean time to clinical diagnosis was 11.8 years among men who were subsequently diagnosed with cancer and who had a baseline PSA ≥3 ng/ml at age 60 . From the Social Security Life Tables it can be calculated that the probability of death within 12 years is 21% for a 60 year old but 45% for a 70 year old. This means that, for a group of 200 men with screen-detected cancer, half 60 years old and half 70 years old, 66 men would die before they would be expected to be clinically diagnosed. Of these, 45, close to 70%, would be in the older age group. These data can also be used to support our finding that, in 60-year-old men with PSAs above biopsy thresholds, most screen-detected cancers would eventually lead to a clinical diagnosis, as close to 80% of men survive longer than the mean lead time. Moreover, the Malmö cohort does not stand alone in finding that PSA is strongly predictive of prostate cancer mortality in unscreened populations. Numerous other studies have shown associations between baseline PSA and long-term prostate cancer outcomes [17–22].
PSA screening can only reduce mortality in that it leads to curative treatment. Two trials have compared surgery with conservative management for prostate cancer [23, 24] and both report a decreased effect of treatment in older men. In one analysis , differences between surgery and conservative treatment started to decrease around age 65 with little benefit for men older than 70. These studies indicate that the effects of treatment diminish with age, suggesting that men older than 70 should only be considered for screening if they are at higher than average risk for prostate cancer mortality and lower than average risk for other cause mortality. Evidence that a risk stratification approach, based on PSA, would improve screening outcomes is provided by a reanalysis of the ERSPC data. Van Leeuwen et al. evaluated the effects of PSA-based screening on men in reference to the PSA-level measured at their first screen. Their results demonstrate that had men with a PSA <2 ng/ml at baseline been excluded from further screening, the number of men needing to be screened and diagnosed to prevent one death would be reduced by 90% and 50%, respectively. Results for a PSA cut-off of 1 ng/ml are not reported. These findings are of particular relevance to our recommendations, as the median age in the screened group was 61 .
One study using SEER data came to a quite different conclusion from the current paper. Welch and Albertsen reported that excess incidence 'which must represent overdiagnosis’ was 'particularly dramatic for younger men’ . There are two major problems with this conclusion. First, it focuses on relative rather than absolute increase in diagnoses. So, for example, the reported seven-fold increase in men younger than 50 constitutes only 8 additional cases per 100,000. A relative seven-fold increase sounds large, suggesting that public health efforts might focus on reducing screening in this age group, whereas the absolute increase demonstrates that such efforts would not have an important impact at the population level. Second, the authors look at an extended period of PSA screening, leading to age-related artifacts. For instance, they state that in men older than 80, 'incidence declined dramatically between 1986 and 2005’. This is because some men who would have been clinically diagnosed in their 80s in the years 2000 to 2005 were screen-detected at an earlier age. It would be entirely unsound to use this finding to suggest that use of PSA in older men does not lead to overdiagnosis.
It is worth considering differences both between the different cohorts within the study and between these study cohorts and contemporary patients. First, the biopsy cohorts and the SEER population sample predominately involved 6-core biopsy rather than the more extended biopsy schemes typical in current practice. This suggests that we may have underestimated the risk of screen-detected cancer. Second, prostate cancer mortality has historically been higher in Sweden – around 5% of male deaths compared to less than 3% in the US . Moreover, recent advances in treatment have led to improvements in survival leading to a lower risk of death . This suggests that we may have overestimated the risk of death. The effect of any such misestimation – underestimating cancer incidence, overestimating mortality risk - would be to exacerbate the difference in risk by PSA levels: men with a low PSA would be even more likely to have cancer detected and even less likely to die without screening.
A clear limitation of our paper is that we are measuring excess incidence, which is only a correlate of overdiagnosis. Yet, overdiagnosis by its very nature is a counterfactual – that a patient would die without a diagnosis if not for screen detection – and, therefore, cannot be directly observed. Moreover, an increase in incidence, even if only temporary, is a prerequisite of screening: a screening test associated with zero overdiagnosis would still raise incidence after clinical implementation, even though incidence rates would subsequently fall below prior levels. That said, a persisting increase in incidence, as seen in the US, is clear evidence of overdiagnosis; comparably, the gross disparities we have seen in the relative number of excess cases by age clearly suggests that restricting PSA screening in older men would importantly shift overdiagnosis rates.