IJC International Journal of Cancer

Transcription

1 IJC International Journal of Cancer The Finnish prostate cancer screening trial: Analyses on the screening failures Tuomas P. Kilpel ainen 1, Teuvo L.J. Tammela 2, Nea Malila 3, Matti Hakama 3, Henrikki Santti 1, Liisa M a att anen 3, Ulf-Håkan Stenman 4, Paula Kujala 5 and Anssi Auvinen 6 1 Department of Urology, Helsinki University Hospital, FI Helsinki, Finland 2 Department of Urology, University of Tampere and Tampere University Hospital, FI Tampere, Finland 3 Finnish Cancer Registry, Unioninkatu 22, FI Helsinki, Finland 4 Department of Clinical Chemistry, Helsinki University Hospital, FI Helsinki, Finland 5 Fimlab Laboratories, Department of Pathology, Tampere University Hospital, FI Tampere, Finland 6 School of Health Sciences, University of Tampere, FI Tampere, Finland Prostate cancer (PC) screening with prostate-specific antigen (PSA) has been shown to decrease PC mortality in the European Randomized Study of Screening for Prostate Cancer (ERSPC). However, in the Finnish trial, which is the largest component of the ERSPC, no statistically significant mortality reduction was observed. We investigated which had the largest impact on PC deaths in the screening arm: non-participation, interval cancers or PSA threshold. The screening (SA) and control (CA) arms comprised altogether 80,144 men. Men in the SA were screened at four-year intervals and referred to biopsy if the PSA concentration was 4.0 ng/ml, or ng/ml with a free/total PSA ratio 16%. The median follow-up was 15.0 years. A counterfactual exclusion method was applied to estimate the effect of three subgroups in the SA: the non-participants, the screen-negative men with PSA 3.0 ng/ml and a subsequent PC diagnosis, and the men with interval PCs. The absolute risk of PC death was 0.76% in the SA and 0.85% in the CA; the observed hazard ratio (HR) was 0.89 (95% confidence interval (CI) ). After correcting for non-attendance, the HR was 0.78 ( ); predicted effect for a hypothetical PSA threshold of 3.0 ng/ml the HR was 0.88 ( ) and after eliminating the effect of interval cancers the HR was 0.88 ( ). Non-participating men in the SA had a high risk of PC death and a large impact on PC mortality. A hypothetical lower PSA threshold and elimination of interval cancers would have had a less pronounced effect on the screening impact. Key words: mass screening, prostatic neoplasms, prostate-specific antigen, randomized controlled trials, mortality Abbreviations: CA: control arm; CI: confidence interval; DRE: digital rectal examination; ERSPC: European randomized study of screening for prostate cancer; F/T PSA: free/total PSA ratio; HR: hazard ratio; PC: prostate cancer; PSA: prostate-specific antigen; RR: risk ratio; SA: screening arm Conflict of interest: Dr. Kilpel ainen has received congress travel support from GlaxoSmithKline. Prof. Tammela has served as an Advisory Board member for Astellas, Amgen and Pfizer, has received consulting fees from Orion Pharma, and received lecture fees from Astellas and Amgen. Prof. Stenman is co-owner of a patent for free and complexed PSA. Dr. Santti has received lecture fees from Astellas. Funding sources have not affected the acquisition, analysis or interpretation of data, or any other aspect of scientific purity. Grant sponsor: The Academy of Finland; Grant number: ; Grant sponsors: Competitive Research Fund (Pirkanmaa Hospital District), Finnish Cancer Organizations DOI: /ijc History: Received 2 June 2014; Accepted 7 Oct 2014; Online 30 Oct 2014 Correspondence to: Dr. Tuomas P. Kilpel ainen, Department of Urology, Helsinki University Hospital, Box 580, FI Helsinki, Finland, Fax 1[ ], tuomas.kilpelainen@hus.fi Controversy over prostate cancer (PC) screening with prostate-specific antigen (PSA) continues, as the quest for a balance between harms and benefits of screening is ongoing. The European Randomized Study of Screening for Prostate Cancer (ERSPC) has shown a 20% reduction in PC mortality (0.49% vs. 0.61%) in the screening arm (SA) relative to the control arm (CA) at a median follow-up of 13 years, 1 but the ERSPC trial is still the only randomized trial that has shown a statistically significant mortality reduction. 2,3 In addition, thorough analyses regarding cost-effectiveness and quality of life effects are needed. 4,5 The main purpose of PC screening is to reduce deaths from the disease, although screening may also have other beneficial effects. Nevertheless, screening for PC always has negative consequences, such as overdiagnosis and overtreatment, and therefore cannot be readily recommended as a public health policy, as the balance of benefits and harms remains uncertain. 6 In order to improve effectiveness of screening, it is important to analyze which component of the screening protocol can be optimized. Previously, descriptive studies of the Dutch 7 and Swedish 8 sections of the ERSPC trial have demonstrated that most of the PC deaths are due to non-participation or cancers detected at first screen (prevalent PCs).

2 2438 Finnish Prostate Cancer Screening Trial What s new? It seems logical that screening for prostate cancer using prostate-specific antigen (PSA) could save lives and indeed it does, according to one large European study. However, the Finnish component of that study showed little effect with PSA screening. Why? In this paper, the authors explore the factors associated with deaths among patients receiving screening. They found that neither lowering the threshold for abnormal PSA levels, nor eliminating the cancers that arose shortly after a normal PSA test would dramatically reduce mortality; rather, the largest impact came from men opting out of PSA screening. The Finnish Prostate Cancer Screening Trial is the largest component of the ERSPC trial and the Finnish mortality results have previously been reported. 9 A relatively conservative screening protocol yielded a modest, statistically nonsignificant mortality reduction, with a hazard ratio (HR) of 0.85 (95% CI ) in favor of the SA at 11.9 years of mean follow-up. In the current analysis, we examine more closely which factors contribute most to screening failures, i.e. PC deaths in the SA. The impacts of three major factors on mortality are quantified: non-participation, a hypothetical lower PSA threshold and interval cancers. We utilize a similar counterfactual analysis that has been used to control for nonparticipation in a randomized trial. 10 This method has also been previously used in the ERSPC trial. 11,12 Material and Methods The Finnish part of the multicenter ERSPC trial comprises 80,144 men born in (aged years at entry) identified from the Finnish Population Registry. After exclusion of men with a previous PC diagnosis, a random sample of 8,000 men was allocated to the SA annually in and the remaining men formed the CA that received no intervention. Men in the SA were invited to a local clinic for the screening test, i.e. determination of the serum PSA concentration. Men with a PSA 4.0 ng/ml were referred to a local urological clinic for diagnostic examinations including digital rectal examination (DRE), transrectal ultrasound and prostate biopsy. Men with PSA levels of ng/ml were referred to an additional test, which in was DRE and since 1999 determination of the free/total (F/T) PSA ratio with a cut-off point of 16%. Men with a suspicious DRE or F/T PSA ratio 16% were referred for diagnostic examinations similar to those with PSA 4.0 ng/ml. The men in the SA were re-invited to the second and third screening rounds in a similar manner four and eight years after the first screen. Information on vital status and place of residence was obtained from the Population Registry. Men with PC and those who had emigrated from the study area were not reinvited. An interval cancer was defined as a cancer detected less than 4 years after the PSA test in a screen-negative man. Information on cancers detected outside the screening protocol (interval cancers, and those in non-participants and in the control arm) were obtained from the nationwide, population-based Finnish Cancer Registry, which has 99% coverage of all solid cancers diagnosed in Finland. 13 The follow-up started at the first of January in the year of randomization ( ) and ended at death, emigration from Finland or the common closing date. Because the accumulation of clinical cancer data to the Finnish Cancer Registry is slower than death certificate data to the Statistics Finland, we used an earlier closing date for cancer incidence analyses (December 31, 2011) than for mortality analyses (December 31, 2012) to ensure complete coverage. All randomized men were analyzed regardless of their participation (in accordance with the intention-to-screen principle). Because of logistic difficulties, 1,671 men were not invited despite having been randomized to the SA. These men were included in the SA as non-participants, as the risk for PC death was similar in these 1,671 men as in the men who were invited but chose not to participate. In Finland, all deaths are registered in the causes of death registry by Statistics Finland ( and the 10th revision of the International Classification of Diseases has been used since To validate the causes of death in our screening study, all deaths in among men diagnosed with PC (regardless of randomization arm) were reviewed by a cause of death committee. An excellent agreement (97.7%; j ) was shown between the official causes of death registry and the cause of death committee. 14 In our study, men who had PC (code C61 in ICD-10) as the underlying cause of death in the official causes of death registry were defined as PC deaths. The study protocol was approved by Helsinki and Tampere University Hospital Ethical committees. Permission to use cancer registry data was obtained from Research and Development Centre for Welfare and Health (STAKES, currently part of the National Institute of Health and Welfare). HRs were estimated for PC and overall mortality for the SA relative to the CA using the Cox proportional hazard model. The proportional hazard assumption was verified using Schoenfeld residuals. To illustrate the impact of specific factors in the SA (nonattendance, lower PSA threshold and interval cancers), a simple analysis with exclusion in the SA only was performed. This was not intended to provide a realistic estimate of the screening effect but simply to show the magnitude of these factors within the SA (i.e., the CA was left intact). In the simple analysis, we estimated HR after exclusion of the men

3 Kilpel ainen et al Figure 1. The principle of counterfactual analysis where a corresponding number of men and deaths are excluded from both the screening arm and control arm. who never attended screening from the SA. In a similar manner, the screen-negative men who had at least once a PSA level of ng/ml and a subsequent PC diagnosis (whether or not they died of it) were excluded. The men with interval cancers were also analyzed accordingly. Subsequently, a counterfactual exclusion method was applied to estimate the impact of specific subgroups in the trial population (Fig. 1). 10 This correction is valid even though the baseline risk for attending and non-attending men is different. First, we identified men in the SA who never attended screening and then multiplied the number of these men, the number of their PC diagnoses and the number of PC deaths by the factor of 1.515, which is the ratio of men in the CA vs. SA. Subsequently, these men were excluded from the SA and a corresponding number (multiplied by 1.515) of men who actually died of PC, men who were diagnosed with PC and men free of PC were randomly selected from the CA and excluded from the trial. The number of PC deaths, PC diagnoses and men free of either were matched. A similar approach was used for a hypothetical lower PSA screening threshold of 3.0 ng/ml. Finally, this method was applied to interval cancers to exclude these men and their randomly selected counterparts from the SA and CA, respectively. All statistical analyses were performed using Stata software (StataCorp, College Station, TX). All statistical tests were two-sided. A p value below 0.05 was considered statistically significant and 95% confidence intervals were used throughout this study. Results There were altogether 31,866 men in the SA and 48,278 men in the CA (Fig. 2). The mean follow-up time in mortality analyses was 13.4 years in both arms (median 15.0 years; standard deviation 4.1 years in both arms; maximum follow-up time 17.0 years). The mean follow-up time for incidence analyses was 12.0 years in the SA and 12.2 years in the CA. The age distribution was similar in both arms (median age years at entry in both arms). Of the men in the SA, 74.6% (N 5 23,771) participated in screening at least once. In the SA, 3,277 PCs were diagnosed (cumulative incidence 10.3%) as compared to 4,082 in the CA (8.5%). Thus PC incidence was higher in the SA (HR 1.24; 95% CI ). Altogether 9,251 men died in the SA during follow-up (cumulative mortality 29.0%), and of these 241 died of PC (cumulative mortality 0.76%). In the CA, there were 14,034 deaths (29.1%) of which 410 (0.85%) were caused by PC. Therefore, screening theoretically prevented 30 PC deaths. HR for the SA was 1.00 ( ) for overall death and 0.89 ( ) for PC death (Fig. 3). When the correction was applied only in the SA to demonstrate the maximum effect of non-participation on PC mortality, the cumulative mortality was 0.64% among the screened men in the SA (HR compared to the uncorrected CA was 0.71; ) (Table 1). A similar analysis regarding a hypothetical lower PSA threshold yielded PC mortality 0.69% (HR compared to the uncorrected CA 0.85; ) and a similar risk of PC death (0.69%) was observed also after exclusion of interval cancers in the SA (HR compared to the unaltered CA 0.81; ). For non-attendance, the Cuzick-corrected PC mortality was 0.64% in the SA and 0.77% in the CA yielding a HR 0.78 ( ) (Table 1). For a hypothetical lower PSA threshold (excluding from the SA the 488 men with PC including 23 PC deaths in men with PSA ng/ml in at least one screen; and from the CA 704 men including 35 PC deaths), the Cuzick-corrected PC mortality was 0.69% in the SA and 0.79% in the CA, with HR 0.88 ( ). Exclusion of 236 interval cancer cases from the SA and correspondingly

4 2440 Finnish Prostate Cancer Screening Trial Figure 2. A CONSORT-style flow chart of the Finnish Prostate Cancer Screening Trial. Figure 3. Nelson Aalen estimates of risk of dying from PC hypothetical cases from the CA resulted in a cumulative PC mortality of 0.69% in the SA and 0.78% in the CA (HR 0.88; ). Discussion The findings from this analysis indicate that in the Finnish component of the ERSPC, non-participation in the screening arm had greater impact on the mortality effect of screening than use of PSA threshold of 4.0 instead of 3.0 ng/ml, or interval cancer occurrence. This means that maximizing screening attendance is crucial for achieving a meaningful population impact. In contrast, neither adoption of a lower PSA threshold nor other measures to increase sensitivity and reduce interval cancer incidence would not have materially improved the outcome. The rationale for PC screening is to decrease mortality from PC. Thus, in population-based screening, every man who is invited to screening but subsequently dies of PC can be regarded as a failure of the screening program. Such failures cannot be avoided, as there will always be men who choose not to participate and there will always be aggressive cancers surfacing between screens. Nevertheless, it is essential to seek all possible means to avoid these failures in order to improve screening performance. We evaluated the contribution of three aspects of the Finnish PC screening trial that contribute to screening failures, non-participation, missing PC s at intermediate PSA levels and interval cancers. The rationale was to identify features of the screening regimen that would have the potential to improve the screening effect. The ERPSC trial was designed as a multicenter trial to assure that sufficient statistical power is achieved and is the largest PC screening trial so far. The ERSPC trial has shown a statistically significant 20% reduction in PC mortality between SA and CA (0.49% vs. 0.61%) after a median followup of 13 years. 1 Three largest centers (Finland, the Netherlands and Sweden) of the ERSPC have published their PC mortality results individually. First, the Swedish reported an unprecedented 44% decrease in PC mortality by screening at median follow-up of 14 years (cumulative PC-mortality 0.44% in the SA vs. 0.78% in the CA). 15 Later, the Finnish center reported a statistically non-significant 15% decrease in

5 Kilpel ainen et al Table 1. The hazard ratios after exclusion of specific subgroups from the screening arm Screening arm No. of men No. of PC deaths (%) No. of men Control arm Personyears Personyears No. of PC deaths (%) HR (95% CI) All men (intention-to-screen analysis) 31, , (0.76) 48, , (0.85) 0.89 ( ) Correcting only the SA Excluding the nonparticipants 23, , (0.64) 48, , (0.85) 0.71 ( ) Excluding men with PSA ng/ml and PC 31, , (0.69) 48, , (0.85) 0.85 ( ) Interval cancers 31, , (0.69) 48, , (0.85) 0.81 ( ) Correcting both the SA and CA Excluding the nonparticipants 23, , (0.64) 36, , (0.77) 0.78 ( ) Excluding men with PSA ng/ml and PC 31, , (0.69) 47, , (0.79) 0.88 ( ) Interval cancers 31, , (0.69) 47, , (0.78) 0.88 ( ) PC 5 prostate cancer; HR 5 hazard ratio; CI 5 confidence interval; SA 5 screening arm; CA 5 control arm. PC mortality (0.47% in the SA vs. 0.55% in the CA) at median follow-up of 12 years. 16 Finally, the Dutch trial reported a statistically significant difference of 20% (0.72% in the SA vs. 0.90% in the CA) between trial arms at median 12.8 years of follow-up. 17 The present study updates the mortality results from the Finnish trial, reporting a cumulative mortality of 0.76% in the SA vs. 0.85% in the CA at median 15 years of follow-up (HR 0.89; 95% CI ). It is noteworthy that despite fulfilling the basic requirements of the ERSPC trial, all these three trials used slightly different screening protocols (regarding e.g. screening interval, PSA threshold and age of recruits). Also, the Swedish and Finnish trials were population-based, whereas the Dutch trial was volunteer-based. In the Dutch trial, only 48% of the men invited to the trial consented to participate resulting in substantial healthy screenee selection bias. 18 Obviously, the more aggressive screening protocol by the Dutch and Swedish trials produced more overdiagnosis than the more conservative protocol of the Finnish trial. Crude estimates can be obtained as the proportion of excess cases in the SA relative to the CA (ignoring the lead-time) and such estimate would be 30% for the Finnish trial and at least twice as high for the Swedish and Dutch trials. The higher incidence of cancer in the SA is to be expected due to earlier diagnosis of clinical cancers and detection of subclinical cancers. Other differences between the screening protocols include the number of screening rounds, which was limited to a maximum of three in Finland, while in the other two centers screening has continued with several subsequent rounds. The aim of the current study is to quantify which elements of the Finnish screening protocol are the main contributors to screening failures. This was addressed using a counterfactual approach estimating the potential results under the assumption that the screening process would have been different from what it was in reality. 19 The findings can pinpoint shortcomings in the screening protocol and subsequently addressing them may improve the screening results. A counterfactual method with balanced exclusions in both trial arms was applied to gain a realistic estimate of the impact using a similar rationale as in the method proposed by Cuzick to assess the effect of intervention adjusting for non-compliance. Previous reports from the Dutch 7 and Swedish 8 centers of the ERSPC trial have described how cancers at prevalence screen (i.e., the first screening episode that detects prevalent cancers in the population) contribute to screening failures. An important group affecting screening effect is the men who choose not to participate. Such men were also identified in the Dutch 7 and Swedish 8 trials, and they differ in relation to e.g. baseline risk of PC death from those participating. 18 In the Finnish trial, the participation proportion was acceptable for a population-based trial (69 71% at each round). 9 Usually, the non-attending men present a high-risk group that needs to be adjusted for in analyses of the screening effect in attenders, 10 as we have done here. An analysis using a counterfactual approach with balanced exclusion in both arms showed that non-participation had a larger impact on mortality than interval cancers or men with intermediate PSA levels. While the problem with non-participant men is not a characteristic of the screening protocol per se, the information given to the men in the recruitment process and the ease of participation affect the participation proportion. Thus if a large effect on mortality was observed, this could direct future endeavors to improve the recruitment process. In the Finnish trial, the non-participant men had substantially higher PC mortality than men in the CA (cumulative mortality 1.09% vs. 0.85%, respectively; HR 1.58, ). For comparison, in the Dutch section of the ERSPC trial, the PC mortality was actually lower in the non-participant men

6 2442 Finnish Prostate Cancer Screening Trial compared to the men in the CA (0.74% vs. 0.81%), but since the participation was high (due to randomization after consent), the absolute number of PC deaths among nonparticipants was very low (N 5 7 of 941). 12 It can be argued that the Cuzick-corrected Finnish trial (omitting the nonparticipants in the SA and a corresponding section of the CA) is roughly comparable to the volunteer-based Dutch trial. In this regard, the Finnish mortality result (HR 0.78; ) is very close to the Dutch result (RR 0.80; ), 17 which suggests that the screening intervention itself has comparable effect, and the difference in the results of the intention to screen analysis mainly reflects adherence of the target population. Retrospectively, it is not possible to definitely analyze whether our mortality results could have been improved by using a lower PSA threshold. To correct for this limitation, we excluded all men from the SA who had at least one negative screen with PSA level of ng/ml, and were subsequently diagnosed with PC. This gives us an upper estimate of how many PC deaths could have been avoided, if the PSA threshold had been lower. The screening effect was only marginally larger (HR 0.85 vs. 0.89) even when only the SA was corrected. The effect was even more subtle when the CA was also corrected (HR 0.88 vs. 0.89).Furthermore,suchlowerbiopsythreshold would have required referral of altogether almost 3,000 additional men to biopsy ( % of screen-negative men per round). Thus, it would have doubled the number of falsely screen-positive men from 6 8% to 11 15% per round. The cumulative incidence of interval cancers was 0.74%, which is comparable to that of the Swedish section (0.73% with biennial screening), but higher than in the Dutch (0.43% with a four-year interval) component of the ERSPC trial. 20 Had there been no interval cancers and hence no PC deaths due to interval cancers, the relative mortality reduction would have been statistically significant 19% if only the SA is corrected (HR 0.81 vs. 0.89) but statistically nonsignificant if also the CA is corrected with the Cuzickmethod (HR 0.88 vs. 0.89). These crude calculations are likely to underestimate the effect of higher sensitivity of lower PSA threshold (i.e., avoiding interval cancers), as higher sensitivity would also improve the prognosis of some screen-detected cancers due to an earlier diagnosis. It is obvious that more frequent screening will reduce the incidence of advanced PC and thus mortality. 21 The screening interval probably has a greater effect on mortality reduction than age at start of screening. 22 In the simulation study by Wu et al., shortening the screening interval to two years References could have reduced the mortality by 7.6 percentage points (95% CI ) in the Finnish trial. 22 However, biennial screening adds costs to screening and is also likely to produce more adverse effects in the form of overdiagnosis, overtreatment and false positive screening results. The Finnish PC screening trial has several obvious strengths, such as population-based design and a large study population (larger than that of the Prostate, Lung, Colorectal and Ovarian cancer screening trial). 3 Also, the cancer data accumulated by the Finnish Cancer Registry and death certificate data accrued by Statistics Finland are very reliable. 14 A real limitation in our trial is that we have so far been unable to estimate contamination (opportunistic screening) in the CA. A questionnaire survey among Finnish physicians showed that 18% of the responders reported having systematically screened asymptomatic men with PSA in 1999 and 9% in In the recent mortality analysis, no major differences in the proportion of T1c cancer in the CAs between centers were seen (40 46% for Dutch, Swedish and Finnish centers) suggesting that there are no substantial differences in contamination between these three centers. 1 Nevertheless, these are crude estimates of contamination and thus the magnitude of the diluting effect of contamination remains inconclusive in the Finnish trial. The Finnish trial showed an 11% statistically nonsignificant mortality reduction with median 15 years of follow-up based on a relatively conservative screening protocol. Of the relevant subgroups in the SA, especially the nonparticipant population in the screening arm had a substantial impact on PC mortality. Avoidance of interval cancers and lower screening threshold would have also enhanced the relative mortality effect, but to a lesser extent. Despite the acceptable participation proportion achieved in the Finnish trial, special attention needs to be given to the high-risk men who tend to opt out from population-based screening programs. Acknowledgements Ethics committee approval: The study protocol was approved by Helsinki and Tampere University Hospital Ethical committees. Permission to use cancer registry data was obtained from Research and Development Centre for Welfare and Health (STAKES, currently part of the National Institute of Health and Welfare). Author Contributions Conception and Design: Hakama, Auvinen, Tammela, Stenman, Kujala, Kilpel ainen; Data analysis: Kilpel ainen, Auvinen; Drafting the article: Kilpel ainen; Critical review: Tammela, Malila, Hakama, Santti, M a att anen, Stenman, Kujala, Auvinen; Final approval: All authors. 1. Schr oder FH, Hugosson J, Roobol MJ, et al. Screening and prostate cancer mortality: results of the European Randomised Study of Screening for Prostate Cancer (ERSPC) at 13 years of follow-up. Lancet Aug 6. doi: /S (14) [Epub ahead of print] 2. Djulbegovic M, Beyth RJ, Neuberger MM, et al. Screening for prostate cancer: systematic review and meta-analysis of randomised controlled trials. BMJ 2010;341:c Andriole GL, Crawford ED, Grubb RL III, et al; PLCO Project Team. Prostate cancer screening in the randomized Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial: mortality results after 13 years of follow-up. JNatlCancerInst2012;104: Heijnsdijk EA, Wever EM, Auvinen A, et al. Quality-of-life effects of prostate-specific antigen screening. N Engl J Med 2012;367: Booth N, Rissanen P, Tammela TL, et al. Healthrelated quality of life in the finnish trial of

7 Kilpel ainen et al screening for prostate cancer. Eur Urol 2014;65: Moyer VA; on behalf of the U.S. Preventive Services Task Force. Screening for prostate cancer: U.S. Preventive Services Task Force Recommendation Statement. Ann Intern Med 2012;157: Zhu X, van Leeuwen PJ, Bul M, et al. Identifying and characterizing escapes -men who develop metastases or die from prostate cancer despite screening (ERSPC, section Rotterdam). Int J Cancer 2011;129: Bergdahl AG, Aus G, Lilja H, et al. Risk of dying from prostate cancer in men randomized to screening: differences between attendees and nonattendees. Cancer 2009;115: Kilpel ainen TP, Auvinen A, M a att anen L, et al. Results of the three rounds of the Finnish Prostate Cancer Screening Trial the incidence of advanced cancer is decreased by screening. Int J Cancer 2010;127: Cuzick J, Edwards R, Segnan N. Adjusting for non-compliance and contamination in randomized clinical trials. Stat Med 1997;16: Roobol MJ, Kerkhof M, Schr oder FH, et al. Prostate cancer mortality reduction by prostatespecific antigen-based screening adjusted for nonattendance and contamination in the European Randomised Study of Screening for Prostate Cancer (ERSPC). Eur Urol 2009;56: Bokhorst LP, Bangma CH, van Leenders GJ, et al. Prostate-specific antigen-based prostate cancer screening: reduction of prostate cancer mortality after correction for nonattendance and contamination in the Rotterdam section of the European Randomized Study of Screening for Prostate Cancer. Eur Urol 2014;65: Teppo L, Pukkala E, Lehtonen M. Data quality and quality control of a population based cancer registry. Experience in Finland. Acta Oncol 1994; 33: M akinen T, Karhunen P, Aro J, et al. Assessment of causes of death in a prostate cancer screening trial. Int J Cancer 2008;122: Hugosson J, Carlsson S, Aus G, et al. Mortality results from the G oteborg randomised population-based prostate-cancer screening trial. Lancet Oncol 2010;11: Kilpel ainen TP, Tammela TL, Malila N, et al. Prostate cancer mortality in the Finnish randomized screening trial. J Natl Cancer Inst 2013;105: Roobol MJ, Kranse R, Bangma CH, et al.; ERSPC Rotterdam Study Group. Screening for prostate cancer: results of the rotterdam section of the European randomized study of screening for prostate cancer. Eur Urol 2013;64: Zeliadt SB, Etzioni R, Ramsey SD, et al. Trends in treatment costs for localized prostate cancer: the healthy screenee effect. Med Care 2007;45: Greenland S. Causal analysis in health sciences. J Am Stat Assoc 2000;95: Roobol MJ, Grenabo A, Schr oder FH, et al. Interval cancers in prostate cancer screening: comparing 2- and 4-year screening intervals in the European Randomized Study of Screening for Prostate Cancer, Gothenburg and Rotterdam. J Natl Cancer Inst 2007;99: van Leeuwen PJ, Roobol MJ, Kranse R, et al. Towards an optimal interval for prostate cancer screening. Eur Urol. 2012;61: Wu GH, Auvinen A, Yen AM, et al. The impact of interscreening interval and age on prostate cancer screening with prostate-specific antigen. Eur Urol 2012;61: Pogodin-Hannolainen D, Juusela H, Tammela TL, et al. Prostate cancer screening: a survey of attitudes and practices among Finnish physicians in 1999 and J Med Screen 2011;18: 46 9.