This article updates with more person-years of follow-up our previously reported finding of no reduction in mortality from prostate cancer in the intervention arm compared with the control arm to 10 years, with no indication of a reduction in prostate cancer mortality to 13 years. We observed a statistically significant 12% relative increase in the incidence of prostate cancer and a non-statistically significant decrease in the incidence of high-grade prostate cancer in the intervention arm. There was no apparent prostate cancer mortality interaction of trial arm with age, baseline comorbidity (defined by a modified Charlson index), and pretrial PSA testing.
In contrast, the ERSPC trial reported a 20% reduction in prostate cancer mortality in their core age group (men aged 55–69 years) that largely occurred after 10 years of follow-up, although a non-statistically significant reduction of 15% was noted in all men (aged 50–74 years) randomly assigned (2
). There were major differences between the PLCO and ERSPC trials. One relates to the extent of opportunistic PSA screening that occurred in the control arms. In the PLCO trial, 45% of those randomly assigned had at least one PSA test in the 3 years preceding randomization, and PSA screening in the control arm was estimated to be 52% during the time period of the last round of screening in the intervention arm (1
). In a more detailed analysis, the intensity of PSA screening in the control arm was estimated to be approximately half of that in the intervention arm (10
). Nevertheless, the level of screening in the intervention arm was substantially greater than that in the control arm throughout the trial screening period. In the ERSPC trial, the degree of contamination was probably less, although details have only been reported from one center (11
A possible reason for the difference in the mortality results between the PLCO and ERSPC trials is differences in the application of treatment for prostate cancer. In a recent publication from the ERSPC trial investigators, it was reported that men in the screened arm who were diagnosed with prostate cancer were more likely to be treated at an academic center than men who were diagnosed in the control arm (12
). The difference in place of treatment favored the screening arm to the extent that outcomes after major surgery are better in major referral centers than in community hospitals. Furthermore, trial arm was associated with treatment choice, especially in men with high-risk localized prostate cancer. Thus, a control arm subject with high-risk prostate cancer was more likely than a screened arm subject to receive radiotherapy (odds ratio [OR] = 1.43, 95% CI = 1.01 to 2.05), expectant management (OR = 2.92, 95% CI = 1.33 to 6.42), or hormonal treatment (OR = 1.77, 95% CI = 1.07 to 2.94) instead of radical prostatectomy (12
). These differences are potentially important given a recent report that radical prostatectomy is associated with improved mortality in young men with aggressive cancers (13
). In contrast, the policy in the PLCO trial not to mandate specific therapies after screen detection resulted in substantial similarity in initial treatment by stage between the two arms (). A planned combined evaluation of the PLCO and ERSPC trials, using mathematical models specifically developed with regard to prostate cancer (14
), may help to resolve some of the current uncertainties.
Improvements in prostate cancer treatment are probably at least in part responsible for declining prostate cancer mortality rates (17
). Even if life is only prolonged by therapy, the opportunities for competing causes of death increase, especially among older men.
A statistically significant interaction of trial arm by comorbidity status has recently been reported by Crawford et al. (4
) using the PLCO prostate mortality data through 10 years. The reported hazard ratio (intervention vs control arms) was 0.56 (22 deaths in the intervention arm vs 38 deaths in the control arm) among those with no comorbidity vs 1.43 (62 deaths in the intervention arm vs 42 deaths in the control arm) among those with a comorbidity. The primary explanation for the difference in analysis between Crawford et al. (4
) and this study lies in the definitions of comorbidity used. Crawford et al. (4
) used an expanded definition of comorbidity, which included, in addition to the Charlson conditions, obesity and hypertension (plus a few other comorbid conditions). Hypertension was included as a comorbid condition whether or not it was well controlled. This increased the proportion of subjects classified as having a comorbidity to 64% (4
), from the 30% based on the modified Charlson index used in this study. With the 13-year data, the relative risks of mortality in the intervention arm vs control arm were 0.73 and 1.26 for those without and with comorbidity as defined in Crawford et al. (4
), ratios both closer to unity than their reported values of 0.56 and 1.43, although the interaction was still statistically significant (P
= .03). Furthermore, to validate our observations, we used only the additional deaths and follow-up time through 13 years that were not included in the analysis by Crawford et al. (4
). A total of 123 additional prostate cancer deaths were available for this analysis since the 164 deaths reported in the original analysis (4
). We found no evidence of interaction between screening arm and comorbidity status because the prostate cancer mortality relative risks comparing trial arms were very similar (RR = 0.99 in the no comorbidity group and RR = 1.06 in the comorbidity group).
In a more detailed exploration of the 13-year data, the highest relative risk of 1.40 was observed for those men with comorbidity according to the definition by Crawford et al. (4
) but with a modified Charlson score of 0. This compares to the relative risk of 0.73 for men with no comorbidity according to the definition by Crawford et al. and a relative risk of 1.11 for men with comorbidity according to the definitions of both Crawford et al. and Charlson. Combining the rates from which the 1.40 and the 0.73 rate ratios for the Charlson scores were calculated, 0 men gives the observed ratio of 1.0 for these men. In any case, the biological plausibility of the interaction reported by Crawford et al. (4
) seems questionable, because the cohort was relatively healthy to start with (18
) and because those men in the 1.40 rate ratio group primarily reported obesity and/or hypertension, which would seem to convey minimal extra risks associated with treatment and minimal differences in treatment options. Thus, the interaction between screening effect and baseline comorbidity is sensitive to the definition of comorbidity. Further, Bach and Vickers (19
) concluded that the data do not support the notion that an elevated degree of comorbidity attenuates the benefit of PSA screening in the PLCO study. They advised caution in the interpretation of the analysis of Crawford et al. (4
A report of follow-up through 14 years of the Goteborg component of ERSPC included findings from some subjects who were not part of the ERSPC analysis (20
). Comparing the earlier ERSPC report (2
) with this report, it seems that 60% of the Goteborg cohort was included in the core age group (55–69 years) of ERSPC. Of the 122 deaths from prostate cancer reported in the Goteborg trial, 109 (89%) occurred in those aged 55–69 years at entry. Although the extent of the overlap in deaths is unclear, it seems reasonable to assume that most or all of these 109 were included in the core group analysis of the ERSPC. Indeed, the overall ERSPC result without the Goteborg (Swedish) component did not quite reach statistical significance (RR = 0.84, 95% CI = 0.70 to 1.01) (2
). Thus, we conclude that the major finding of the Goteborg study (20
) concerning reduction in prostate cancer mortality from screening seems largely derived from previously reported data from the ERSPC trial. Furthermore, as the control group in the Goteborg trial was followed passively, it is possible that differences in treatment had an impact on the reported results.
This study has certain limitations. For example, the borderline statistically significant lower all-cause (excluding PLCO cancers) mortality in the intervention arm compared with the control arm raises the question as to whether a reduction in prostate cancer mortality in the present analysis has somehow been masked by problems in death attribution, the sticking diagnosis effect (21
), even though a death review process has been in operation throughout the trial (6
). There is no single cause of death that could account for the difference in all-cause mortality. The deaths in excess in the control arm are from cerebrovascular accidents, other circulatory diseases; respiratory illnesses; infectious diseases; endocrine, nutritional, metabolic and immune diseases; diseases of the nervous system; accidents; and other causes. The deaths in excess in the intervention arm are from non-PLCO neoplasms, ischemic heart disease, and digestive diseases. As anticipated, the majority of those diagnosed with prostate cancer died of other causes. Of the 4250 prostate cancer case patients diagnosed in the intervention arm, 455 (10.7%) had died of causes other than prostate, lung, and colorectal cancer by 13 years, the corresponding numbers for the usual care arm being 3815 prostate cancer case patients and 377 deaths (9.9%), respectively. Thus, a higher percentage of deaths from other causes rather than a deficit occurred among the prostate cancer patients diagnosed in the intervention arm, an indication of the overdiagnosis associated with PSA screen detection (14
). We conclude that error in cause of death attribution does not account for the excess in prostate cancer deaths in the intervention arm; random errors and bias cannot be ruled out as explanations for the discrepant outcomes.
A systematic review and meta-analysis of all randomized screening trials for prostate cancer has been reported (22
), including the data from PLCO (1
) and ERSPC (2
) trials in 2009, data from the Goteborg and French components that were not part of the ERSPC 2009 report, and the earlier Quebec and Norwegian trials. In this meta-analysis, there was no statistically significant effect of screening on prostate cancer mortality (RR = 0.88, 95% CI = 0.71 to 1.09) and no effect on overall mortality (RR = 0.99, 95% CI = 0.97 to 1.01).
We plan to update the mortality findings from the prostate component of the PLCO when follow-up data through 15 years are available. In PLCO, the screening that occurred in the usual care arm was not enough to eliminate the expected impacts of the annual screening in the intervention arm, such as earlier diagnosis and a persistent excess of cases. Therefore, the trial was evaluating the effect of adding an organized component of annual screening to the opportunistic screening already in place, and as far as the follow-up has continued to date (13 years), there is no evidence of a benefit. Indeed, there is evidence of harms, in part associated with the false-positive tests, but also with the overdiagnosis inseparable from PSA screening, especially in older men.
Caution is required in determining whether the efficacy of screening is influenced by comorbidity in the relatively healthy population of men that is generally targeted for screening. Using an approach to classify comorbidity as has been previously applied in the trial, we did not find any evidence for an interaction of comorbidity with trial arm, in contrast to what has been recently reported.