Prognostic factors of prostate cancer mortality in a Finnish randomized screening trial

Transcription

1 International Journal of Urology (2018) 25, doi: /iju Original Article: Clinical Investigation Prognostic factors of prostate cancer mortality in a Finnish randomized screening trial Subas Neupane, 1 Ewout Steyerberg, 2 Jani Raitanen, 1,3 Kirsi Talala, 4 Juho Pylv al ainen, 1 Kimmo Taari, 5 Teuvo LJ Tammela 6,7 and Anssi Auvinen 1 1 Faculty of Social Science, Health Sciences, University of Tampere, Tampere, Finland, 2 Department of Public Health, Erasmus University Medical Center, Rotterdam, the Netherlands, 3 UKK Institute for Health Promotion Research, Tampere, 4 Finnish Cancer Registry, 5 Department of Urology, University of Helsinki and Helsinki University Hospital, Helsinki, 6 Department of Urology, Tampere University Hospital, and 7 School of Medicine, University of Tampere, Tampere, Finland Abbreviations & Acronyms CCI = Charlson Comorbidity Index CI = confidence interval ERSPC = European Randomized Study of Screening for Prostate Cancer HR = hazard ratio IQR = interquartile range PCa = prostate cancer PSA = prostate-specific antigen Correspondence: Subas Neupane Ph.D., Faculty of Social Science, Health Sciences, University of Tampere, FI Tampere, Finland. subas.neupane@uta.fi Received 11 May 2017; accepted 7 November Online publication 10 December 2017 Objectives: To identify the prognostic factors of prostate cancer death among patients enrolled in a Finnish prostate cancer screening trial. Methods: Data on TNM stage, Gleason score, serum prostate-specific antigen at diagnosis, comorbidity and primary treatment were collected from medical records, as well as date and cause of death from Statistics Finland. Four prognostic risk groups were defined based on TNM stage, Gleason score and prostate-specific antigen at diagnosis. Hazard ratios and their 95% confidence intervals for prostate cancer death were calculated using Cox regression and competing-risk analysis with follow up from randomization. The differences in the effects of prognostic factors were assessed using interaction terms. Results: The 15-year survival was significantly lower among cases in the control arm compared with the screening arm (0.90 vs 0.92). However, the survival advantage was limited to screen-detected cases (0.94 vs 0.91 in cases detected outside screening). The prognostic risk group was the strongest factor predicting survival in the control arm, but weaker in screen-detected cases. Advanced disease was associated with substantially poorer outcome in cases detected outside screening than in screen-detected disease. Primary treatment had a similar effect in all groups. Comorbidity had a small prognostic effect in the control arm only. Conclusions: Prognostic factors had a different effect on the outcome of cases detected through screening as those diagnosed otherwise. A high diagnostic prostatespecific antigen and advanced disease carried a poor prognosis, especially among the cases detected outside screening, even when lead-time was eliminated. This shows that the screening resulted in earlier treatment among the cases in the screening arm. Key words: prognostic factors, prostate cancer, randomized trial, screening, survival. Introduction PSA-based screening has been shown to reduce PCa mortality in a European screening trial, 1 and recent modeling studies suggested that the US Prostate, Lung, Colorectal, and Ovary trial would also have shown a mortality benefit, if the contamination had been lower and biopsy compliance higher. 2,3 Even if population-based PCa screening programs have not been launched, PSA testing is widely used in several countries. This has led to substantial improvement in PCa survival, which likely reflects partly a true screening effect, and partly an artefactual increase in survival as a result of lead-time (earlier diagnosis) and overdiagnosis (detection of cases that would not have been detected without screening during a man s lifetime). Optimal management of PCa is challenging because of a wide spectrum of disease behavior, from indolent to highly aggressive. Prognostic stratification of patients with PCa for the choice of optimal treatment remains a major issue. Optimal management should avoid both overtreatment; that is, excessively aggressive treatment modalities in patients who are not at high risk of disease progression, and undertreatment; that is, ineffective management of aggressive disease leading to treatment failure and development of metastatic disease The Japanese Urological Association

2 Prognostic factors of PCa mortality It is unclear whether the major prognostic factors exert a similar effect on the outcome of cases detected through screening as those diagnosed otherwise. Here, we compared the impact of established prognostic classifications among men with screen-detected cancers and other cases in a Finnish PCa screening trial. The results are expected to inform prognostic prediction and hence allow optimized choice of treatment modalities to match the likely outcome of the disease. This issue has not been addressed in earlier Finnish Randomized Study of Screening for Prostate Cancer analyses, which have evaluated the differences primarily between the trial arms to assess the effect of screening on PCa mortality at the population level. Methods The Finnish Randomized Study of Screening for Prostate Cancer is the largest component in the multicenter ERSPC. 1 Briefly, men (aged years) were identified from the Finnish Population Registry. 4 During , a random sample of 8000 men was annually allocated to the screening arm, and the remaining men formed the control arm without any intervention. Men in the screening arm were invited for screening based on serum PSA. Screen-positive men (either PSA 4.0 ng/ml or PSA ng/ml with the proportion of free PSA <0.16) were referred to diagnostic examinations including transrectal ultrasound-guided biopsy. The second screening round was carried out 4 years later and the final one after 8 years. Men aged >71 years, those diagnosed with PCa and men who had emigrated from the study area were no longer invited. All men diagnosed with PCa between randomization and the end of 2013 were included in this analysis. The follow up started at randomization to avoid lead-time bias, and ended at death, emigration or the common closing (31 December 2014). Death from PCa was the end-point, with causes of death obtained from Statistics Finland. The method of PCa detection was divided into screendetected (n = 1633) and other cases (n = 6315), where the other cases included the patients of the control arm (n = 4475), as well as interval cases and cancers among nonparticipants (n = 1840) from the screening arm (Fig. 1). Information on TNM stage and Gleason score was abstracted from the medical records. PSA at diagnosis (not screening) was used for all men. A modified version of the CCI was constructed based on hospital inpatient episodes up to the year 2010, obtained from the nationwide hospital discharge registry and categorized into no versus any comorbidity (score 0 vs 1 8) based on the distribution of CCI score. 5 The prognostic risk group at diagnosis was classified as low, moderate, high and advanced, according to European Association of Urology criteria. 6 Low-risk PCa was defined as stage T1 T2 with Gleason score 6 and PSA <10 ng/ml; moderate risk as T1 T2 with either Gleason 2 6 and 10 PSA < 20 or Gleason 7 with PSA <20; high risk was either stage T1 T2 with either Gleason 8 10 or 20 PSA 100, or stage T3 with any Gleason score. Advanced PCa was defined as the presence of at least one of the following: T4, M1, N1 or PSA >100 ng/ml. (n = 1633) Screening arm (n = ) PCa diagnosed (n = 3473) Primary treatment data were retrieved from hospital records and classified as radical prostatectomy, curative radiation therapy (external beam radiation or brachytherapy), endocrine therapy (luteinizing hormone releasing hormone agonist or antagonist, antiandrogen, or both, or surgical castration) or observation (watchful waiting or active surveillance). Statistical analysis Randomized (n = ) No PCa (n = ) Non-participant, interval and post screening cancers (n = 1840) (n = ) PCa diagnosed (n = 4474) Fig. 1 A CONSORT-style flow diagram showing the study groups and participants. The differences in clinical characteristics at presentation were assessed using Pearson s v 2 -tests and v 2 -tests for trend. Frequencies of missing values in the prognostic factors varied from 1.5% in primary treatment to 18% in PSA at diagnosis (Table 1). PSA and other variables were statistically imputed for cases with a single value in the key prognostic variables. We assumed that any data were missing at random, as supported by patterns of missing value analysis. We used the multiple imputation by chained equations algorithm to assign the most likely values predicted based on correlation with other variables. 7,8 There was no essential difference in findings using complete case analysis (n = 7227) and imputed data (n = 7948). To assess statistical significance of the differences between the groups in the impact of various prognostic factors (trial arm or method of detection), interaction of the predictor and indicator of the patient group was tested. An interaction term was added to a model with both main effects. Statistical significance was assessed using the likelihood ratio test. Cox proportional hazards analysis was used to estimate HRs of PCa death and their 95% CIs. Variables were selected for multivariable models using a backward stepwise selection 2017 The Japanese Urological Association 271

3 S NEUPANE ET AL. Table 1 Characteristics of the PCa cases and PCa deaths by trial arm (and method of detection in the screening arm) PCa cases PCa death Comparison of screen-detected and not screen-detected groups with controls Comparison of screen-detected and not screen-detected groups with controls Characteristics (n = 4475) Screening arm (n = 3473) Control (n = 1633) Not screen-detected (n = 456) screening arm (n = 278) (n = 1840) Control (n = 93) Not screen-detected (n = 185) Age at randomization (years), n (%) (23.1) 829 (23.9) 1035 (23.1) 358 (21.9) 471 (25.6) 67 (14.7) 41 (14.8) 67 (14.7) 13 (14.0) 28 (15.1) (25.8) 897 (25.8) 1154 (25.8) 458 (28.1) 439 (23.9) 102 (22.4) 65 (23.4) 102 (22.4) 18 (19.4) 47 (25.4) (25.7) 925 (26.6) 1152 (25.7) 474 (29.0) 451 (24.5) 119 (26.1) 76 (27.3) 119 (26.1) 27 (29.0) 49 (26.5) (25.7) 822 (23.7) 1134 (25.7) 343 (21.0) 479 (26.0) 168 (36.8) 96 (34.5) 168 (36.8) 35 (37.6) 61 (33.0) Median age at 69 (55 83) 67 (55 83) 69 (55 83) 67 (55 72) 69 (55 83) 68 (55 81) 67 (55 81) 68 (55 81) 67 (55 72) 68 (58 81) diagnosis (IQR) Study center, n (%) Helsinki 3200 (71.5) 2503 (72.1) 3200 (71.5) 1185 (72.6) 1318 (71.6) 336 (73.7) 200 (71.9) 336 (73.7) 64 (68.8) 136 (73.5) Tampere 1275 (28.5) 970 (27.9) 1275 (28.5) 448 (27.4) 522 (28.4) 120 (26.3) 78 (28.1) 120 (26.3) 29 (31.2) 49 (26.5) Median PSA at diagnosis, 9.8 ( ) 8.0 ( ) 9.8 ( ) 5.6 (2 123) 9.6 ( ) 40.3 ( ) 32.8 ( ) 40.3 ( ) 9.1 ( ) 37.9 ( ) ng/ml (IQR) PSA at diagnosis (ng/ml), n (%) <0.001 < < <6 939 (21.0) 879 (35.3) 939 (21.0) 471 (28.8) 408 (22.2) 33 (7.2) 23 (8.3) 33 (7.2) 10 (10.8) 13 (7.0) (28.9) 721 (20.8) 1292 (28.9) 228 (14.0) 493 (26.8) 53 (11.6) 38 (13.7) 53 (11.6) 17 (18.3) 21 (11.4) (25.0) 554 (16.0) 1118 (25.0) 114 (7.0) 440 (23.9) 77 (16.9) 32 (11.5) 77 (16.9) 3 (3.2) 29 (15.7) (20.5) 405 (11.7) 919 (20.5) 53 (3.3) 352 (19.1) 276 (60.5) 130 (46.8) 276 (60.5) 17 (18.3) 113 (61.1) Missing 207 (4.6) 914 (32.3) 207 (4.6) 767 (47.0) 147 (8.0) 17 (3.7) 55 (19.8) 17 (3.7) 46 (49.5) 9 (4.9) Biopsy Gleason sum, n (%) <0.001 < < (45.5) 2071 (59.6) 2034 (45.5) 1195 (73.2) 876 (47.6) 78 (17.1) 64 (23.0) 78 (17.1) 38 (40.9) 26 (14.1) (33.0) 877 (25.3) 1478 (33.0) 318 (19.5) 559 (30.4) 113 (24.8) 75 (27.0) 113 (24.8) 31 (33.3) 44 (23.8) (18.7) 464 (13.4) 836 (18.7) 117 (7.2) 347 (18.9) 230 (50.4) 126 (45.3) 230 (50.4) 24 (25.8) 102 (55.1) Missing 127 (2.8) 61 (1.8) 127 (2.8) 3 (0.2) 58 (3.2) 35 (7.7) 13 (4.7) 35 (7.7) 0 13 (7.0) CCI, n (%) <0.001 < (77.1) 2755 (79.3) 3449 (77.1) 1382 (84.7) 1371 (74.5) 373 (81.8) 226 (81.3) 373 (81.8) 83 (89.3) 143 (77.3) (22.9) 718 (20.7) 1026 (22.9) 249 (15.3) 469 (25.5) 83 (18.2) 52 (18.7) 83 (18.2) 10 (10.8) 42 (22.7) Risk group, n (%) <0.001 < < Low 1160 (25.9) 995 (28.6) 1160 (25.9) 520 (31.8) 475 (25.8) 17 (3.7) 9 (3.2) 17 (3.7) 6 (6.5) 3 (1.6) Intermediate 1398 (31.2) 764 (22.0) 1398 (31.2) 200 (12.3) 564 (30.7) 40 (8.8) 33 (11.9) 40 (8.8) 14 (15.1) 19 (10.3) High 1155 (25.8) 651 (18.8) 1155 (25.8) 215 (13.2) 436 (23.7) 121 (26.5) 80 (28.8) 121 (26.5) 31 (33.3) 49 (26.5) Advanced 520 (11.6) 238 (6.9) 520 (11.6) 35 (2.1) 203 (11.0) 262 (57.5) 121 (43.5) 262 (57.5) 15 (16.1) 106 (57.3) Missing 242 (5.3) 825 (23.7) 242 (5.3) 663 (40.6) 162 (8.8) 16 (3.5) 35 (12.6) 16 (3.5) 27 (29.0) 8 (4.3) Primary treatment, n (%) <0.001 < <0.001 < Radical prostatectomy 826 (18.5) 899 (25.9) 826 (18.5) 566 (34.7) 333 (18.1) 21 (4.6) 29 (10.4) 21 (4.6) 20 (21.5) 9 (4.9) Radiation 583 (13.0) 595 (17.1) 583 (13.0) 361 (22.1) 234 (12.7) 26 (5.7) 30 (10.8) 26 (5.7) 23 (24.7) 7 (3.8) Hormonal 2143 (47.9) 1130 (32.5) 2143 (47.9) 327 (20.0) 803 (43.6) 366 (80.3) 186 (66.9) 366 (80.3) 41 (44.1) 145 (78.4) Observation 843 (18.8) 807 (23.2) 843 (18.8) 375 (23.0) 432 (23.5) 32 (7.0) 26 (9.4) 32 (7.0) 9 (9.7) 17 (17.2) Missing 80 (1.8) 42 (1.2) 80 (1.8) 4 (0.2) 38 (2.1) 11 (2.4) 7 (2.5) 11 (2.4) 0 7 (3.8) P-value shown above the respective screening arm column is for the comparison with controls The Japanese Urological Association

4 Prognostic factors of PCa mortality (a) 1.00 (b) 1.00 Proportion event-free Proportion event-free Fig. 2 Survival curve for PCa from the date of randomization to the last follow-up time for (a) control and screening arm, and (b) screendetected and other cases. Control 0.00 Screening Time since randomization (years) Number at risk Control Screening Others Time since randomization (years) Number at risk Others method with P = 0.10 as the cut-off. In the competing risk analysis, the risk of PCa death was estimated allowing for competing causes of death (i.e. death from other causes than PCa among PCa cases). 9 Statistical analyses were carried out using Stata version 13.0 (StataCorp, College Station, TX, USA) and SPSS 23 (IBM Corp., Armonk, NY, USA). Ethical issues Helsinki and Tampere University Hospital Ethics committees reviewed the study protocol. Permission to use cancer registry data was obtained from the National Institute for Health and Welfare. Written, informed consent was obtained from the participating men in the screening arm. Results Of the men randomized (after exclusion of 282 men who had pre-randomization cancer or death), PCa was diagnosed in 3473 men of in the screening arm, and in 4475 men of in the control arm (Fig. 1). The median length of follow up was 16 years in both arms, regardless of the method of detection. The median age at diagnosis was higher (69 years vs 67 years, P < 0.001) in the control arm than in the screening arm (Table 1). The participants in the control arm had more frequent comorbidities (age-standardized prevalence 10.5% vs 8.7%, P-value for the trend <0.001), statistically significantly higher PSA levels at diagnosis (9.59 ng/ml vs 7.20 ng/ml, P < 0.001), more frequent high-risk (25.8% vs 18.8%) or advanced tumors (11.6% vs 6.9%) and received hormonal therapy as the primary treatment (47.9% vs 32.5%, P < 0.001). The characteristics of the cases in the control arm were comparable with those detected outside screening in the screening arm. There were 734 deaths from PCa. Cause-specific survival from randomization was significantly higher among the cases in the screening arm compared with the control arm (Kaplan Meier survival estimates 0.96 vs 0.95 at 10 years, and 0.92 vs 0.90 at 15 years; age-adjusted HR 0.79, 95% CI over the entire follow up; Fig. 2a). PCa mortality in the two arms commenced to diverge after 7 8 years, and the difference increased over time (Fig. 3). Most prognostic factors had a comparable effect in both arms (Table 2). PSA 6 10 appeared to carry a poorer prognosis than PSA <6 in the screening arm, but not in the control arm. A biopsy Gleason score of 8 10 had a stronger effect in the screening arm compared with the control. The presence of comorbidity was associated with poorer survival only in the screening arm, though the difference between the arms was not significant. Risk group was a very strong prognostic determinant in both arms, with some suggestion of poorer outcome for advanced disease in the control arm (survival proportion at 15 years in the screening arm 0.60 vs 0.51 in the control arm). A survival advantage was also seen for screen-detected cases compared with other cancers (Kaplan Meier mortality estimates 0.03 vs 0.05 at 10 years and 0.06 vs 0.09 at 15 years; HR 0.58, 95% CI for the entire follow up) with 33% lower risk of PCa mortality among screendetected cases (Fig. 2b). Age had a slightly stronger effect in the screen-detected cases, whereas PSA >20 was associated with a poorer prognosis in the cases diagnosed outside screening (HR 9.7, 95% CI vs HR 3.9, 95% CI , interaction P = 0.005), with some difference also at PSA Nelson-Aalen cumulative hazard Fig Control Screening Time since randomization (years) Cumulative risk of PCa mortality from the date of randomization The Japanese Urological Association 273

5 S NEUPANE ET AL. Table 2 Unadjusted and multivariable-adjusted hazard ratios for PCa death during follow up from the date of randomization in the control and screening arm of the Finnish Randomized Study of Prostate Cancer Screening HR (95% CI) Screening arm Characteristics Unadjusted Multivariable Unadjusted Multivariable P-value for interaction term Age at randomization (years) ( ) 1.32 ( ) 1.57 ( ) 1.31 ( ) ( ) 1.46 ( ) 1.83 ( ) 1.62 ( ) ( ) 1.95 ( ) 2.86 ( ) 2.02 ( ) Age 9 trial arm 0.97/0.98 PSA at diagnosis < ( ) 1.65 ( ) ( ) 1.73 ( ) ( ) 8.31 ( ) PSA 9 trial arm 0.30 Biopsy Gleason sum ( ) 2.66 ( ) ( ) 9.10 ( ) Gleason 9 trial arm 0.23 Method of detection NA NA Others NA NA 1.68 ( ) 0.92 ( ) Comorbidity index ( ) 1.05 ( ) 1.52 ( ) 0.94 ( ) Comorbidity 9 arm 0.41/0.83 Risk group Low Intermediate 2.10 ( ) 1.89 ( ) 2.55 ( ) 2.64 ( ) High 7.39 ( ) 6.17 ( ) 7.24 ( ) 6.36 ( ) Advanced ( ) ( ) ( ) ( ) Risk group 9 arm 0.08/0.16 Primary treatment Radical prostatectomy Radiation 1.68 ( ) 2.04 ( ) 1.56 ( ) 1.60 ( ) Hormonal 6.78 ( ) 1.61 ( ) 5.34 ( ) 1.93 ( ) Observation 1.43 ( ) 1.09 ( ) 1.02 ( ) 1.15 ( ) Treatment 9 arm 0.76/0.97 Multivariable analysis using backward selection method with P = 0.10 as the cut-off. P-value for interaction term derived from univariate (first part) and multivariate (second part) model. level 6 10 (Table 3). Advanced disease carried a poor prognosis, especially in cases diagnosed outside screening (interaction P < 0.001). The results of the competing risk analysis were similar to the main analysis (Table S1). Discussion PCa-specific survival was significantly lower among PCa cases in the control arm compared with the screening arm at 18 years from randomization. cases had a more favorable prognosis than cancers diagnosed outside screening. Of the specific prognostic factors, a higher risk was associated with advanced disease in the control arm compared with the screening arm, and particularly among cases detected outside screening compared with screen-detected cancers. A high diagnostic PSA was related to poor outcome, especially among the cases detected by other means than screening. PCas detected by screening had more favorable prognostic features and a better outcome than other cases. This is in line with the findings of other ERSPC centers The present finding is also consistent with the reduction in PCa mortality reported in the ERSPC, and to a smaller degree also within the Finnish trial alone. 4 However, screening tends to detect slowly-growing tumors (length bias), which could result in more favorable outcomes without a mortality reduction. In addition, screening can increase survival time also in the absence of any real mortality effect as a result of lead-time bias (earlier diagnosis without change in time of death). We avoided this caveat by using follow up from randomization instead of diagnosis, assuming that mortality risk estimates The Japanese Urological Association

6 Prognostic factors of PCa mortality Table 3 Unadjusted and multivariable-adjusted hazards ratios of PCa death for prognostic factors during follow up from the date of randomization among screen-detected and other patients in the Finnish Randomized Study of Prostate Cancer Screening HR (95% CI) Screen detected Others Characteristics Unadjusted Multivariable Unadjusted Multivariable P-value for interaction term Age at randomization (years) ( ) 1.13 ( ) 1.53 ( ) 1.34 ( ) ( ) 1.51 ( ) 1.81 ( ) 1.50 ( ) ( ) 2.86 ( ) 2.69 ( ) 1.87 ( ) Age 9 method of detection 0.31/0.35 PSA at diagnosis < ( ) 1.14 ( ) ( ) 1.84 ( ) ( ) 9.74 ( ) PSA 9 method of detection Biopsy Gleason sum ( ) 2.03 ( ) ( ) 7.97 ( ) Gleason 9 method of detection 0.16 Comorbidity index ( ) 0.59 ( ) 1.34 ( ) 1.02 ( ) Comorbidity 9 method of detection 0.19/0.48 Risk group Low Intermediate 2.45 ( ) 2.67 ( ) 2.56 ( ) 2.33 ( ) High 5.29 ( ) 5.20 ( ) 9.10 ( ) 7.45 ( ) Advanced ( ) ( ) ( ) ( ) Risk group 9 method of detection <0.001/<0.001 Primary treatment Radical prostatectomy Radiation 1.87 ( ) 1.92 ( ) 1.48 ( ) 1.72 ( ) Hormonal 3.90 ( ) 2.19 ( ) 6.69 ( ) 1.62 ( ) Observation 0.69 ( ) 0.91 ( ) 1.42 ( ) 1.18 ( ) Treatment 9 method of detection 0.038/0.81 Multivariable analysis using backward selection method with P = 0.10 as the cut-off. P-value for interaction term derived from both univariate (first part) and multivariable (second part) model. would be underestimated as a result of excess follow-up time around the time of diagnosis. Of all 3473 cases in the screening arm, 47% were screendetected, 16% were interval cases, 18% were diagnosed 4 years after screening (post-screening cases) and 13% were among non-participants. The compliance rate was 91%. The frequency of PSA tests in the screening arm after the intervention remained reasonably high (24.4%), but this also includes tests carried out for diagnostic purposes among men with urinary symptoms or cancer suspicion (result not shown). In the present study, older age at entry (67 years) was associated with a slightly higher risk of PCa death in the screen-detected cases. This finding did not persist, however, when allowing for competing causes of death. This could be due to detection of aggressive cancer at repeat screening, after harvesting those with a long lead-time at earlier rounds. Alternatively, these deaths could be due to contraindications for curative treatment, as the finding was not confirmed when allowing for other cases of death. The effect of PSA >20 ng/ml was stronger among cases detected outside of screening than in screen-detected cancers. This suggests improved treatment outcomes also for larger tumors, because the finding cannot be due to lead-time only, as the follow up started at randomization. Almost no difference between the arms was found in the association of the biopsy Gleason score with PCa mortality. Also, the differences between the methods of detection in the impact of Gleason score were non-significant. This is consistent with no major grade shift during the natural course of PCa from the preclinical to clinical phase. We used a modified Charlson score based on hospitalization data alone, but found, in line with others, that comorbidity is predictive of PCa survival. 13,14 However, men with comorbidity (CCI 1) fared worse than those with no comorbidity only in the control arm. This could be explained by 2017 The Japanese Urological Association 275

7 S NEUPANE ET AL. treatment delays among patients with comorbidity. 15 The lack of such impact in the screening group might be due to screening resulting in more active investigations and treatments. We used a dichotomous CCI (0 vs 1), restricting comparisons of men with various degrees of comorbidity. As expected, the prognostic risk group was the strongest determinant of PCa survival in both trial arms. The risk of PCa death increased with risk group in both arms, but the effect was significantly smaller in the screening arm and in screendetected cases. This suggests that screening improves outcomes also within risk group, or the screen-detected cases have more favorable features within risk group, or both. This might be due to earlier diagnosis and more active treatment for the patients in the same prognostic groups in the screening arm. The frequency of endocrine treatment was lower in the screening arm than the control arm (47.9% vs 32.5%) in accordance with earlier reports from the ERSPC. 16 Our finding is consistent with an earlier ERSPC study that showed a better survival among curatively-treated patients (prostatectomy or radiotherapy) in the screening arm compared with the control arm. 17 We found the most obvious difference in advanced disease, with substantially poorer outcomes in cases diagnosed outside of screening than screen-detected cancer. This cannot be explained by lead-time gained by screening, because it was avoided by starting follow up at randomization instead of diagnosis. This suggests that earlier detection also allows improvement in treatment outcome in advanced disease. This could be due to less advanced disease or better effectiveness of treatment, or both. Furthermore, hormonal treatment was associated with worse outcomes, especially in the control arm. Overdiagnosis leads to overestimation of survival in screen-detected cases. We were unable to identify cases detected by opportunistic screening. This very likely underestimates the differences in outcome between the screendetected and other cases. Missing data was common in some covariates, especially PSA, and it was addressed by imputation. Imputation did not materially affect the results. In summary, screen-detected cancers have a better prognosis than cases detected outside screening. The screening arm had a 20% reduced risk of PCa mortality compared with the controls. Advanced disease is associated with poorer outcomes in cases outside of screening than screen-detected cancers, even when lead-time is eliminated. Minor differences between patient groups were found for specific prognostic factors. A high diagnostic PSA was related to poor outcome, especially among the cases detected outside of screening. This indicates that the screening resulted in earlier treatment among the cases in the screening arm. Nevertheless, the prognostic risk group based on stage, Gleason score and PSA at diagnosis remains the major prognostic determinant for PCa detected by screening and other means. Conflict of interest None declared. References 1 Schr oder FH, Hugosson J, Roobol MJ et al. Screening and prostate cancer mortality: results of the European Randomized Study of Screening for Prostate Cancer (ERSPC) at 13 years of follow-up. Lancet 2014; 384: Shoaq J, Mittal S, Halpern JA, Scherr D, Hu JC, Barbieri CE. Lethal prostate cancer in the PLCO cancer screening trial. Eur. Urol. 2016; 70: Palma A, Lounsbury DW, Schlecht NF, Agalliu I. A system dynamics modelling of serum prostate-specific antigen screening for prostate cancer. Am. J. Epidemiol. 2015; 183: Kilpel ainen TP, Tammela TLJ, Malila N et al. The Finnish prostate cancer screening trial: analyses on the screening failures. Int. J. Cancer 2015; 136: Quan H, Li B, Couris CM et al. Updating and validating the Charlson comorbidity index and score for risk adjustment in hospital discharge abstracts using data from 6 countries. Am. J. Epidemiol. 2011; 173: Babjuk M, Burger M, Zigeuner R et al. EAU guidelines on non muscle-invasive urothelial carcinoma of the bladder: update Eur. Urol. 2013; 64: White IR, Royston P, Wood AM. Multiple imputation using chained equations: issues and guidance for practice. Stat. Med. 2001; 30: Rubin DB. Inference and missing data. Biometrika 1976; 63: Fine J, Gray RJ. A proportional hazard model for the subdistribution of a competing risk. J. Am. Stat. Assoc. 1999; 94: van der Cruijsen-Koeter Vis AN, Roobol MJ, Wildhagen MF, de Koning HJ, van der Kwast TH, Schr oder FH. Comparison of screen detected and clinically diagnosed prostate cancer in the European randomized study of screening for prostate cancer, section Rotterdam. J. Urol. 2005; 174: Postma R, Schr oder FH, van Leenders GJ et al. Cancer detection and cancer characteristics in the European Randomized Study of Screening for Prostate Cancer (ERSPC) Section Rotterdam. A comparison of two rounds of screening. Eur. Urol. 2007; 52: Vis AN, Roemeling S, Reedijk AM, Otto SJ, Schr oder FH. Overall survival in the intervention arm of a randomized controlled screening trial for prostate cancer compared with a clinically diagnosed cohort. Eur. Urol. 2008; 53: Scosyrev E, Messing EM, Mohile S, Golijanin D, Wu G. Prostate cancer in the elderly. Cancer 2012; 118: Guzzo TJ, Dluzniewski P, Orosco R, Platz EA, Partin AW, Han M. Prediction of Mortality after radical prostatectomy by Charlson Comorbidity Index. Urology 2010; 76: Gold HT, Do HT, Dick AW. Correlates and effect of suboptimal radiotherapy in women with ductal carcinoma in situ or early invasive breast cancer. Cancer 2008; 113: Wolters T, Roobol MJ, Steyerberg EW et al. The effect of study arm on prostate cancer treatment in the large screening trial ERSPC. Int. J. Cancer 2010; 126: Roemeling S, Roobol MJ, Gosselaar C, Schr oder FH. Biochemical progression rates in the screen arm compared to the control arm of the Rotterdam Section of the European Randomized Study of Screening for Prostate Cancer (ERSPC). Prostate 2006; 66: Supporting information Additional Supporting Information may be found in the online version of this article at the publisher s web-site: Table S1. Competing risk (fine-gray) analysis of variables associated with the risk of PCa death during follow up from the date of randomization in the control and screening arm of the Finnish Randomized Study of Prostate Cancer Screening (FinRSPC) The Japanese Urological Association