Impacts of a population-based prostate cancer screening programme on excess total mortality rates in men with prostate cancer: a randomized controlled trial

Abstract

Objectives

To assess the effect of screening in terms of excess mortality in the European Randomized Study of Screening for Prostate Cancer (ERSPC).

Methods

A total of 141,578 men aged 55–69 were randomized to systematic screening or usual care in ERSPC sections in Finland, Italy, the Netherlands and Sweden. The excess number of deaths was defined as the difference between the observed number of deaths in the prostate cancer (PC) patients and the expected number of deaths up to 31 December 2006. The expected number was derived from mortality of all study participants before a diagnosis with PC adjusted for study centre, study arm and study attendance. The excess mortality rates were compared between the two study arms.

Results

The PC incidence was 9.25 per 1000 person-years in the intervention arm and 5.49 per 1000 person-years in the control arm, relative risk (RR) 1.69 (95% confidence interval [CI] 1.62–1.76). The excess mortality among men with PC was 0.29 per 1000 person-years in the intervention arm and 0.37 per 1000 person-years in the control arm; the RR for excess mortality was 0.77 (95% CI 0.55–1.08). The absolute risk reduction in the excess mortality was 0.08 per 1000 person-years. The overall mortality was not significantly different between the intervention and the control arms of the study: RR 0.99 (95% CI 0.96–1.01).

Conclusions

Although the reduction in excess mortality was not statistically significant, the between-arm reduction in excess mortality rate was in line with the previously reported 20% reduction in the disease-specific mortality. This finding indicates that the reduction in PC mortality in the ERSPC trial cannot be due to a bias in cause of death adjudication.

INTRODUCTION

The European Randomized Study of Prostate Cancer Screening (ERSPC) has shown that prostate specific antigen (PSA)-based screening for prostate cancer (PC) reduces PC mortality in men aged 55–69 at invitation for screening.^1–3 The effect was demonstrated by comparing the number of deaths from PC in the trial populations assigned to screening and usual care. The cause of death for all men diagnosed with PC who subsequently died was labelled as either ‘death from’ or ‘death with’ PC by independent committees in each of the participating countries. ⁴ No effect of screening was found on the all-cause mortality.

There have been debates on whether disease-specific death should be the endpoint of a cancer screening trial.^5–7All-cause mortality is potentially a more robust outcome than disease-specific mortality, because all-cause mortality indicates the impact on total life-years (longevity) including potential adverse effects of the intervention. For example, a screening programme can reduce the disease-specific mortality, whereas the indirect effects of the screening procedure (anxiety, depression or even the treatment for the disease being screened for itself) cause an increased risk of death from other causes. In such a case, the disease-specific mortality decreases, while the overall mortality may increase. In clinical trials, such effects are known from fibrates for cholesterol reduction, ⁸ antiarrhythmics following myocardial infarction^9,10 and liberal red cell transfusion in the critically ill. ¹¹

However, for evaluation of a screening trial, disease-specific mortality remains an appropriate endpoint, because screening trials would require unrealistic sample sizes to have the statistical power to detect a minor reduction in all-cause mortality. In a recent study, we have presented an alternative to assess the total effect of screening on the mortality in a randomized trial. ¹² It comprises a comparison of the estimates of the excess mortality rate in the cancer patients in both arms of the study. These excess mortality rates are based on the observed number of cancer patient deaths in excess of the number expected on the basis of a cohort of cancer-free individuals per unit of time. To evaluate excess mortality as an indicator of effectiveness of PC screening we used data of the four largest centres of the ERSPC.

METHODS

Study design

The ERSPC was designed as a randomized, multicentre trial of screening for PC, with mortality from PC as the primary outcome. ¹³ As complementary approach to the primary outcome, it was planned to calculate the ‘excess mortality’ in PC patients (i.e. patients diagnosed after randomization). ¹⁴

In the present study, data from the four largest ERSPC centres are analysed to obtain robust estimates of the excess mortality. The procedure of recruitment and randomization differed among centres. In Finland, Italy and Sweden, men underwent randomization before written informed consent was provided by those allocated to the screening arm (population-based effectiveness trial). In the Netherlands, men provided informed consent before randomization (efficacy trial). We randomized men aged 55–69: 80,377 men in Finland, 14,517 in Italy, 11,852 in Sweden and 34832 in the Netherlands. In Italy, Sweden and the Netherlands the study was performed based on allocation in a 1:1 ratio to the screening group or the control group. In Finland, the size of the screening group was fixed with a resulting allocation ratio to the screening versus control arm of approximately 1:1.5.

Screening tests and indications for biopsy

In Sweden, a PSA cut-off value of 3.0 ng/mL was used as an indication for biopsy. ¹⁵ In Finland, a PSA value of 4.0 ng/mL or more was defined as positive test, by which men were referred for biopsy; those with a value of 3.0–3.9 ng/mL underwent an ancillary test (digital rectal examination [DRE] until 1998 and determination of the ratio of the free PSA value to the total PSA value [biopsy indication if FT ratio <0.16] starting in 1999) and were referred for biopsy if either of the two tests was positive. ¹⁶ In Italy, a PSA value of 4.0 ng/mL or more was defined as positive, but men with a PSA value of 2.5–3.9 ng/mL also underwent ancillary tests (DRE and transrectal ultrasonography [TSU]). ¹⁷ In the Netherlands, up to February 1997, a combination of DRE, TSU and PSA testing (with a cut-off value of 4.0 ng/mL) was used for screening; in 1997, this combination was replaced by PSA testing only (biopsy indication if PSA ≥3 ng/mL). ¹⁸ Centres used sextant biopsies guided by TSU for diagnostic examinations. In June 1996, lateralized sextant biopsies were recommended. In Italy, transperineal sextant biopsies were used. In Finland, a biopsy procedure with 10–12 biopsy cores was adopted in 2002 as a general policy similar for the two study groups. The screening interval in Finland, Italy and the Netherlands was four years; Sweden used a two-year interval. Treatment of PC was performed according to local policies and guidelines. Earlier analysis showed similar treatment approaches in the two study groups according to tumour stage. ¹⁹

Follow-up data

Cancer incidence data were obtained by linking the trial database with the regional (Italy, Sweden) or national (the Netherlands, Finland) Cancer Registry. Mortality data of participants who died in the period up to 31 December 2006 were obtained by linking the trial database with the National Causes of Death Registry. Linkage was possible by using the personal administrative number of each participant as a linkage key. Causes of death were evaluated in a blinded fashion and according to a standard algorithm or, after validation, on the basis of official causes of death. ⁴ Patients were determined to have died from PC if they were classified as either ‘definitely PC death’, as ‘probable PC death’ or as ‘PC intervention-related death’.

Statistical analysis

Follow-up for diagnosis and mortality began at randomization and ended at death, emigration or common closing date (31 December 2006), whichever came first. Men who were randomized to the intervention arm were classified as attendees or non-attendees using the following definition: after randomization all men in the intervention arm were considered as non-attendees until the date that they participated for the first time in the study by a PSA test, DRE or TRUS. At this date their status switched from non-attendee to attendee. Men who were not screened during the study were considered as non-attendees during the complete follow-up.

For all participants the total follow-up period was subdivided into yearly intervals until death, emigration or censoring. For each interval, the attained age equalled the age in the previous interval for the same individual plus one year (age at the initial interval = age at randomization). For each attained age, for all episodes in which no cancer was diagnosed, the number of deaths were added. By dividing the total number of deaths by the respective number of person-years, crude estimates of the all cause mortality rates for each attained age were obtained. For each centre, these crude estimates were smoothed by fitting a Poisson model to the number of deaths with attained age centred at 65 years as a sole predictor. The logarithm of the number of person-years at risk was added as an offset to the model constant. The expected number of deaths in attendees and non-attendees randomized to the screening arm, and in men randomized to the control arm, was calculated by multiplying the person-years in cancer patients for each attained age with the age, attendance and arm-specific crude mortality of men without cancer. For all calculated expected deaths, for each attained age the excess number of deaths was calculated as the difference of the observed and expected deaths in cancer patients. For all calculated excess deaths, for each attained age, the excess mortality rates were calculated by dividing the excess deaths by the total number of person-years (i.e. in all participants irrespective of disease status) for that attained age. Thus the expected number of deaths was calculated based on the assumption that the PC patients would have had the same age-specific mortality as the study population taking into account the centre and attendance status (attendees and non-attendees in the intervention arm and all men in the control arm before any diagnosis of PC).

For each study arm, the sum of the number of life-years of all participants from the randomization until either censoring or death was calculated. This sum was divided by the number of men randomized to that arm, which yields the average number of life-years until either censoring or death per study arm per participant. Subtracting the number of life-years per participant in the control arm from the number of life-years per participant in the screening arm yields the average life-time gained per participant.

Poisson regressions were used to calculate all cause-specific mortality and overall mortality and all PC incidence rate ratios, confidence intervals (CIs) of these ratios and their associated P values based on the intention-to-screen principle. ²⁰ Separate models were used for each centre with the arm as the only covariate. The excess mortality hazard rate ratio was derived as shown in Appendix A with its associated 95% CI based on the delta method (accounting for attendance in the intervention arm and difference in randomization scheme in Finland). A two-sided P value <0.05 was considered statistically significant. All analyses were performed with the commercially available STATA package: Data Analysis and Statistical Software, version 10.

RESULTS

Baseline characteristics

A total of 141,578 men aged 55–69 years at randomization were included in the study. Of these men, 62,578 were assigned to the intervention arm and 79,000 to the control arm. The PC incidence rate in the intervention arm was 1.69-fold higher than in the control arm, i.e. 5206 (cumulative incidence 8.3%) men were diagnosed with PC in the intervention arm and 3871 (4.9%) men were diagnosed with PC in the control arm (Table 1). Up to the end of 2006, 9673 (15.5%) men died in the intervention arm and 12,309 (15.6%) died in the control arm. This resulted in an all-cause mortality that was not significantly different between the two arms: relative risk (RR) 0.99 (95% CI 0.96–1.01, P = 0.27), Table 2.

Table 1

Number of subjects, deaths and results of screening according to study group, attendees and non-attendees

	Intervention	Attendees	Non-attendees	Controls	All
Number of men	62,578	50,242	12,336	79,000	141,578
Age at randomization
Median	59	60	59	59	59
Mean	59.8	59.9	59.6	59.7	59.8
Total PY	562,521	437,709	124,813	705,264	1,267,786
Number of deaths	9,673	6,100	3,573	12,309	21,982
Deaths per total PY, rate	17.20	13.94	28.63	17.45	17.34
Number of PC	5,206	4,681	525	3,871	9,077
PC per total PY, rate	9.25	10.70	4.21	5.49	7.16
PY PC patients	27,260	25,117	2,143	14,208	41,468
Total deaths in PC patients	705	574	131	597	1,302
Number of PC specific deaths	188	126	62	296	484
PC specific deaths per total PY, rate	0.33	0.29	0.50	0.42	0.38

Attendees: men randomized to intervention arm who attended screening; non-attendees: men randomized to intervention arm who never attended screening; PY: person years; deaths: number of all cause deaths; PC: prostate cancer; rates are expressed as number per 1000 person-years. Note that it is not possible to calculate the excess deaths directly from the table. It illustrates the importance of using attained age-specific mortality rates. For example, for the intervention arm, the excess number of deaths derived from the person-years and deaths reported in this table equals 705 − (9673/562,521) × 27260 = 236.2 (considerably higher than the actually excess mortality of 164.5)

Table 2

Effect of screening on prostate cancer incidence, overall mortality, prostate cancer specific mortality and excess mortality based on the intention-to-screen principle

	Intervention*	Controls*	Relative risk	95% CI	P value
PC rates	9.25	5.49	1.69	1.62–1.76	<0.01
Death rates	17.20	17.50	0.99	0.96–1.01	0.27
PC death rate	0.33	0.42	0.79	0.66–0.96	0.01
Excess mort rate in men with PC	0.29	0.37	0.77	0.55–1.08	0.13

*All rates are expressed as number per 1000 person-years. The intervention arm includes both attendees and non-attendees

Disease-specific mortality

Up to the end of 2006, 188 men in the intervention arm and 296 men in the control arm died from a PC-related cause according to the causes of death committee (CODC). The corresponding disease-specific mortality rate was 0.33 per 1000 person-years in the intervention arm and 0.42 per 1000 person-years in the control arm. This resulted in a statistically significant reduction in PC mortality of 21% in the intervention relative to the control arm: RR 0.79 (95% CI 0.66–0.95, P = 0.014; Table 2).

Excess mortality

Up to the end of 2006, the excess mortality rate in men with PC was 0.29 men per 1000 person-years in the intervention arm and 0.37 men per 1000 person-years in the control arm. This resulted in a non-significant reduction in the excess mortality rate of 23% in the intervention population relative to the control population: RR 0.77 (95% CI 0.55–1.08, P = 0.132; Table 2). This equalled with an absolute risk reduction in the excess mortality of 0.075 per 1000 person-years. The life time gained per participant was approximately 23 days after a median follow-up of nine years.

DISCUSSION

The latest publications by the ERSPC trial^1–3 have added to the evidence that PSA-based screening for PC can reduce mortality from the disease. Some issues remain unresolved, most notably the cost-effectiveness of screening. ²¹ There has been little debate on the validity of PC-specific mortality as an endpoint of a PC screening trial. Nevertheless, in the ERSPC trial, screening might have reduced the number of deaths from PC, while adverse effects of the screening procedure (anxiety, depression over-treatment) might have caused an increase in the death from other causes. Up to the present study, these screening effects remained unknown. The present study showed a reduction in the excess mortality in men with PC that is consistent with the reduction in the rate in disease-specific mortality as an effect of population-based screening. This is an important observation because, by definition, disease-specific mortality is incorporated in excess mortality, but is not identical to it.

Disease-specific mortality refers to mortality due to only one particular cause of death such as disease progression, while treatment related mortality and excess mortality measures both the direct and indirect mortality due to the cancer of interest, e.g. including cachexia, uraemia, suicide, depression and loss of interest in life. Although in this study the excess mortality in men with PC did not reach statistical significance, the results corroborate the finding that PSA-based screening reduces mortality related to PC and strongly supports the view that PSA-based screening reduces the death from PC. The support provided by this analysis complements the earlier findings, because the current analysis does not rely on human expertise to determine causes of death.

After a median of nine years follow-up, the ERSPC as a whole showed an alpha spending adjusted RR for death from PC in the intervention group of 0.80 (95% CI 0.65–0.98; P = 0.04) based on data including 162,243 men randomized in seven different countries. ¹ Data in the subsequent study showed a reduction in the PC specific mortality rate of 21%, RR 0.79 (95% CI 0.68–0.91; P = 0.001). ³ The causes of death were assigned based on the consensus of the CODC that evaluated all deceased cases in a blinded fashion according to a standard algorithm. ⁴ Nevertheless, in practice it was impossible to blind the arm of the study entirely before the committee reviewed the records of deceased men. The records were for example not blinded for stage at diagnosis, although this is very suggestive for the arm of the study. Therefore, it remained possible that the lack of adequate blinding of the records with respect to study arm had resulted in a bias in the cause of death ascertainment. Based on the findings in the present study, it appears that this potential bias did not influence the study outcomes.

The results are not completely in line with a previous study that compared the effect of screening on the disease-specific with excess mortality in the ERSPC section Rotterdam only. ¹² The previous study showed in contrast to the disease-specific mortality rates, an increased difference in the excess mortality rates between the two study arms. The main difference between the excess mortality and disease-specific mortality rate was observed for the older age group at randomization, i.e. 70–74 years, men who were not included in the present study. The results for men aged 55–64 at randomization in the present study were in line with the results presented before. ¹² The differences in results by age are in agreement with excess mortality studies on mammography screening for breast cancer which also observed a more pronounced discrepancy between the excess mortality and disease-specific mortality analysis for the older ages.^22,23 In line with suggestions made in the previous publication, a possible explanation for this is an increase in the uncertainty of individual cause of death ascertainment in older age groups, when overall mortality is increasing.

Excess mortality analyses are often limited by the validity of the expected mortality. Typically, in an excess mortality analysis, the expected mortality is estimated from nationwide population life tables stratified by age, sex and period. However, in a screening setting, the general population does not seem to be suitable as a reference population because the people who participate are healthier and of higher socioeconomic classes, i.e. subject to healthy screenee bias.^12,24 For this reason, in the present study, the expected mortality was based on all participants before a diagnosis of PC and adjusted for study arm and study participation. These adjustments were needed, shown in detail by Kranse et al. ²⁵ This study showed that if the expected mortality were to be estimated without a correction for attendance status, an excess mortality analysis on PC screening would overestimate the effect of screening because men who participate and attend in a screening trial have been shown to be healthier and have a decreased risk of death from other causes compared with men who do not attend. Additional adjustment by estimating the expected mortality with correction for non-compliance among men participating in the intervention arm with a biopsy indication might have optimized the excess mortality because men in the intervention arm who had a raised PSA but were not biopsied were likely to have higher all-cause mortality than those with biopsy. Although this seems to have a very small effect on the final outcomes, it might be incorporated in future excess mortality analyses.

Population-based screening entails an intervention in a healthy population and, therefore, should have a favourable harm–benefit trade-off, which at the moment is not yet proven. Therefore, although this study corroborates that PSA-based screening reduces mortality, population-based PC screening cannot be recommended at the moment. The PC incidence in the intervention arm was 1.69-fold higher than in the control arm. Because PC is a leading chronic condition affecting men and the incidence of overdiagnosed PC is increasing rapidly, there is an urgent need to resolve these issues. Therefore additional studies are needed. The present study showed that the estimation of excess mortality can be used in future studies to evaluate the effects of screening on PC mortality. This is useful because disease-specific mortality requires reliably ascertained causes of death and assumes that cancer mortality is independent of competing risk mortality, an assumption which is especially for PC only approximately true.

In conclusion, PSA-based screening reduces the excess mortality among men with PC. The estimated excess mortality reduction of 23% is in line with our earlier results, implying that PSA-based screening is effective in reducing the mortality related to PC.