Projecting prostate cancer mortality in the PCPT and REDUCE chemoprevention trials

Authors


Abstract

BACKGROUND.

Two recent chemoprevention trials demonstrated significant reductions in overall prostate cancer incidence. However, a possible increase in high-grade disease has raised concerns that the harms of the drugs, including mortality because of high-grade disease, may outweigh the benefits. The authors attempted to estimate the effect of these drugs on prostate cancer mortality to be able to better evaluate the cost-benefit tradeoff.

METHODS.

The authors analyzed prostate cancer incidence in the Prostate Cancer Prevention Trial (PCPT) and Reduction by Dutasteride of Prostate Cancer Events (REDUCE) trial, which evaluated finasteride and the related compound dutasteride, respectively (both vs placebo). They used 13-year prostate cancer survival data from the Prostate, Lung, Colorectal and Ovarian (PLCO) trial to project prostate cancer mortality from incidence patterns; survival rates were applied to incident cancers according to prognostic strata, which were defined by Gleason score, prostate-specific antigen level, and clinical stage. For PCPT, the analysis was performed using both original trial results and previously published adjusted analyses that attempted to account for artifacts related to the drugs' effect on prostate volume.

RESULTS.

For the PCPT trial, the estimated relative risk (RR) for prostate cancer mortality was 1.02 (95% confidence interval [95% CI], 0.85-1.23) using the original trial results and 0.87 (95% CI, 0.72-1.06) and 0.91 (95% CI, 0.76-1.09) based on the adjusted PCPT analyses. For the REDUCE trial, the RR for prostate cancer mortality was 0.93 (95% CI, 0.80-1.08).

CONCLUSIONS.

Projecting a mortality outcome of the PCPT and REDUCE trials as an approach to weighing benefits versus harms suggests at most a small increase in prostate cancer mortality in the treatment arms, and possibly a modest decrease. Cancer 2013. © 2012 American Cancer Society.

INTRODUCTION

Chemoprevention is one approach to lessening the burden of prostate cancer. In 2003, the Prostate Cancer Prevention Trial (PCPT) reported a 25% reduction in prostate cancer period prevalence associated with the 5α-reductase inhibitor finasteride compared with placebo in men with baseline prostate-specific antigen (PSA) levels < 3 ng/mL.1 In 2009, the Reduction by Dutasteride of Prostate Cancer Events (REDUCE) trial, which compared the related 5α-reductase inhibitor dutasteride against placebo in men with a PSA > 3 ng/mL and a prior negative biopsy reported a 23% reduction in overall prostate cancer period prevalence.2

However, the results of the PCPT and REDUCE trials raised concerns regarding high-grade disease; in both trials, an increase was observed in the incidence of Gleason score 8 to 10 disease in the active drug treatment arm.1, 2 Although the increase was not found to be statistically significant in either study, the numbers of Gleason score 8 to 10 cases were small and there was relatively little power to identify a significant difference. The biological plausibility of the increase in high-grade disease has been debated, with some evidence of a possible mechanism.3, 4 Recently, a US Food and Drug Administration (FDA) panel voted overwhelmingly against the approval of both finasteride and dutasteride for a cancer prevention indication.5 The primary reason for the disapproval was the apparent increase in the incidence of Gleason score 8 to 10 tumors in the patients in the active drug arm that was observed in the 2 trials.

At the FDA panel, and elsewhere, the question arose as how to balance a clear decrease in Gleason score 2 to 6 disease, and possibly Gleason score 7 disease, with a possible increase in Gleason score 8 to 10 disease. The absolute number of Gleason score 2 to 6 tumors prevented by the drugs overwhelms the observed absolute number of additional Gleason score 8 to 10 tumors, because Gleason score 2 to 6 disease is so much more common. However, Gleason score 8 to 10 disease has a considerably worse prognosis than Gleason score 2 to 6 (or even 7) disease, and therefore it is not clear how the trade-off would be evaluated. Furthermore, especially in the PCPT, in which many of the Gleason score 2 to 6 tumors were diagnosed only on mandated end-of-study biopsy performed in men with low PSA levels, the clinical significance of these tumors has been questioned.5

One common metric that could be used to assess the benefits and harms of an increase in one type of prostate cancer versus a decrease in another is disease-specific mortality. Specifically, one could project from the incidence results, using appropriate survival statistics, the prostate cancer-specific mortality rate over an extended period of follow-up for the drug and placebo arms (note that neither the REDUCE nor PCPT trial planned for an extended mortality follow-up). This approach would effectively weight each cancer according to its risk of disease-related death and use prostate cancer mortality as the common metric with which to compare the risks and benefits of the drugs. The recently published US Preventive Services Task Force recommendation statement regarding screening for prostate cancer, which recommended against such screening, suggested, under its discussion of research needs and gaps, that more study was needed to determine the effect of 5α-reductase inhibitors on the mortality from prostate cancer.6 This exercise then comprises a modeling-based component of such research.

Several prior analyses have attempted to estimate the survival benefit associated with finasteride by applying prostate survival statistics to observed prostate cancer incidence rates in the PCPT trial.7-9 These analyses all used survival rates obtained either from cohorts of men with symptomatically detected disease or from the Surveillance, Epidemiology, and End Results (SEER) database, which represents a mixture of screen-detected and symptomatically detected disease. However, because of frequent PSA screening and routine scheduled (by protocol) biopsies in both the PCPT and REDUCE trials, the cancers diagnosed in these trials were overwhelmingly detected in the absence of symptoms. Because of lead time and overdiagnosis, which are both substantial in prostate cancer, survival rates would be expected to be substantially higher in screen-detected cases than in symptomatically detected cases, even if screening had no effect on prostate cancer mortality.10 Therefore, to estimate the mortality effect in these 2 chemoprevention trials as accurately as possible, we used survival rates obtained from the intervention arm of the Prostate, Lung, Colorectal and Ovarian (PLCO) cancer screening trial, in which men were receiving annual screening with PSA.

MATERIALS AND METHODS

PCPT Trial

The PCPT trial randomized 9423 men to treatment with finasteride (at a dose of 5 mg/day) and 9459 men to placebo.1 Eligibility criteria included a baseline PSA level ≤ 3.0 ng/mL. Subjects were treated and followed over a 7-year period, during which they were monitored for compliance and screened annually with PSA and digital rectal examination (DRE). Subjects with an elevated PSA or suspicious DRE were referred for biopsy. In addition, all subjects without a cancer diagnosis at the end of the study were asked to return for an end-of-study biopsy. All biopsies were reviewed centrally by pathologists blinded to the study arm. Among men with at least 1 biopsy, a 25% reduction in prostate cancer period prevalence was reported for men in the finasteride treatment arm compared with the placebo arm.

An issue in the PCPT trial was a possible bias in cancer detection and Gleason grading on biopsy because of the effect of finasteride in shrinking the prostate gland. Several research groups attempted to correct for this bias using various approaches, including extrapolating the Gleason scoring from the subset of cases with radical prostatectomy and correcting for differences in prostate volume across treatment arms.11-13 The mortality projection was performed in multiple ways for the PCPT trial, using both the observed (biopsy) Gleason scoring and adjusted scoring.

REDUCE Trial

The REDUCE trial tested dutasteride at a dose of 0.5 mg/day against placebo for prostate cancer chemoprevention.2 Eligibility criteria included a baseline PSA of 3.0 ng/mL to 10 ng/mL (2.5-10 ng/mL for men aged < 60 years) and a history of a single negative biopsy. Subjects were treated and followed for 4 years and received PSA tests every 6 months. Biopsies were performed at year 2 and year 4 or for cause. All positive biopsies were reviewed centrally by a pathologist who was blinded to the treatment arm. A total of 8231 men were enrolled in the trial. Among men who underwent prostate biopsy in years 1 to 4, an overall reduction of 20% in the prostate cancer period prevalence was observed for the dutasteride treatment arm compared with the placebo arm.

PLCO Trial

The PLCO trial is a multicenter, randomized controlled trial designed to test the efficacy of screening for 4 types of cancer in persons aged 55 years to 74 years at baseline.14-16 The methods have been described elsewhere. Briefly, randomization to a screened or control arm took place between 1993 and 2001, with 154,900 persons (and 76,685 men) enrolled at 10 centers. Men in the screening arm received PSA and DRE at baseline (year 0) and then annually through year 3, and received PSA only at years 4 and 5. Exclusion criteria included history of a PLCO cancer; removal of the entire prostate; and, beginning in 1995, having undergone > 1 PSA blood test within the past 3 years. A PSA result of > 4 ng/mL was considered to be positive. Prostate cancer cases and deaths were ascertained through routine follow-up of positive screens and through the use of an annual study update questionnaire supplemented with National Death Index searches. Certified tumor registrars at each screening center abstracted, using a standardized protocol, clinical and pathologic (if available) T, N, and M classification characteristics, biopsy Gleason score, and Gleason score from radical prostatectomy (if performed). There was no centralized pathology review of Gleason scoring.

The major characteristics of all 3 trials are summarized in Table 1.

Table 1. Major Characteristics of PLCO and the 2 Chemoprevention Trials
CharacteristicPLCOPCPTREDUCE
  • Abbreviations: PCPT, Prostate Cancer Prevention Trial; PLCO, Prostate, Lung, Colorectal and Ovarian cancer screening trial; PSA, prostate-specific antigen; REDUCE, Reduction by Dutasteride of Prostate Cancer Events trial; SD, standard deviation.

  • a

    Intervention (screening) arm only.

  • b

    All trials excluded men with a prior diagnosis of prostate cancer.

  • c

    Approximately 10% of men were randomized before the time (April 1995) that this criterion went into effect.

  • d

    PSA of 3.0 to 10.0 ng/mL for men aged >60 years.

No. randomized38,350a18,8828231
Y enrolled1993-20011994-1997Not reported
Mean age (SD), y62 (5)63 (6)63 (6)
Baseline eligibility criteriab<2 prior PSA testscPSA <3 ng/mLSingle negative biopsy; PSA 2.5-10 ng/mLd
PSA testing scheduleAnnual (for 6 y)Annual (for 7 y)Every 6 mo (for 4 y)
BiopsyFor causeFor cause and end of studyFor cause, y 2, and y 4

Statistical Methods

To estimate mortality from prostate cancer from the chemoprevention trial data, we simulated a mortality endpoint trial by “virtually” following up all diagnosed prostate cancer cases. Because there are 13 available years of PLCO survival data, to use all that survival information, we virtually followed subjects with prostate cancer for 13 years from diagnosis. However, because subjects were diagnosed at different study times, this necessitates that (virtual) follow-up time from enrollment varied across cases. Cases diagnosed at the latest time in the PCPT, year 7, were followed for 20 years from enrollment whereas those diagnosed earlier, at year “n,” were followed for only n +13 years, and were effectively “censored” for the last (7-n) years of a virtual 20-year mortality trial.

However, this virtual censoring should not bias the results, because the time of diagnosis was similar across treatment arms in both the PCPT and REDUCE trials. In the PCPT trial, 23% of cancers in the finasteride treatment arm were diagnosed through year 3 and 46% were diagnosed at end of study biopsy; the corresponding percentages for the placebo arm were 23% and 50%, respectively. In the REDUCE trial, 66% of cancers in the dutasteride arm versus 68% of cancer in the placebo arm were diagnosed at study years 1 through 2. Thus, virtual censoring was approximately equal across arms for both trials.

We divided prostate cancer cases into prognostic strata, and for each stratum computed expected deaths as the number of cases times 1 minus the appropriate survival rate, in which the latter was the 13-year prostate-specific survival rate in the corresponding stratum among PLCO cases in the intervention arm as computed using the Kaplan-Meier method. For each stratum, the same survival rate applied for each trial arm. We did not account for other-cause mortality, because it would be expected to be similar for each arm. Total expected deaths were calculated by summing up expected deaths over all prognostic categories, with death rates defined as expected deaths over total subjects with incidence endpoint data (ie, with biopsy). The relative risk (RR) for mortality was computed as the ratio of expected death rates.

For the PCPT trial, because the majority of Gleason score 2 to 6 cancers were diagnosed at the time of the end-of-study biopsy (ie, not for cause) and were generally low risk, we stratified these further according to PSA level (≤ 4 ng/mL, > 4 ng/mL) and T classification (T1a-T1c, T2a-T2c) for clinically localized cancers, and by TNM stage (stage III, stage IV) for nonclinically localized cancers. Because finasteride reduces PSA levels, for the drug arm we used the same adjustments to PSA as were used in the PCPT trial to determine for-cause biopsy (ie, multiplying measured PSA by a factor of 2.0 through year 3 and by a factor of 2.3 afterward).1 Gleason score 7 and score 8 to 10 cancers were stratified according to TNM stage (I/II, III, and IV); for stages III and IV, because of small numbers, cases in the control and intervention arms of the PLCO trial combined were used to derive survival rates. In addition, for stage IV, Gleason score 2 to 6 and Gleason score 7 cases were grouped together.

In addition to the standard (“unadjusted”) analysis of PCPT data, we performed 2 “adjusted” analyses based on adjustments for the biopsy grading artifact mentioned earlier. The first adjusted analysis is based on the method of Pinsky et al, which involved extrapolating radical prostatectomy (RP) Gleason results to all cases; the second is based on the analysis of Cohen et al, which modeled Gleason score-specific risk corrected for differences in volume and the number of biopsy cores across treatment arms.12, 13

For the first adjusted analysis, the method derived Gleason category-specific misclassification probabilities of biopsy grading that were differential by arm; applying these produced the adjusted RRs for each Gleason category reported by Pinsky et al.12 For the current analysis, we applied these misclassification probabilities assuming that they were independent of PSA and clinical stage. This produced adjusted counts of cases in each Gleason category, with the categories stratified similarly as described earlier by PSA and clinical stage.

For the second adjusted analysis, the reported odds ratios (ORs) from Cohen et al were converted to revised counts by Gleason category in the placebo arm; counts in the finasteride arm were assumed unchanged.13 This (prostate volume-adjusted) method, unlike the RP extrapolation method, changes not only the Gleason distribution of cases but also the total number of cases, because more cancers are likely to be missed in the placebo arm because of larger volume. As noted earlier, conversions of Gleason category were assumed independent of PSA and clinical stage; in addition, all additional placebo arm Gleason score 2 to 6 cancers were assumed to be of the lowest risk (PSA < 4 ng/mL and T1a-T1c classification).

Unlike the PCPT, we did not have the access to the raw data in the REDUCE trial, and therefore we were only able to use as strata the reported Gleason categories (scores 2-6, 7, and 8-10). Note that the majority of men in the REDUCE trial had a PSA > 4 ng/mL (mean baseline PSA, 5.9; standard deviation, 1.9) so it is not as critical to stratify by PSA (or T classification).

A bootstrapping approach was used to calculate 95% confidence intervals (95% CIs) on the RR for mortality. Specifically, we performed bootstrapping with replacement (n = 1000 runs) to randomly generate replicate trial data sets and then applied the process described earlier to each replicate, including necessary modifications for the adjusted PCPT analyses. Variability in the PLCO-derived survival rates was accounted for by sampling from the distribution of survival rate estimates, which were assumed to be normal with the standard error as computed from the Kaplan-Meier analysis.

RESULTS

Table 2 summarizes the findings of the PCPT and REDUCE trials with respect to prostate cancer period prevalence by treatment arm and Gleason score.5, 12, 13 For the PCPT trial, the data for the observed (biopsy) Gleason categories are shown as well as the results of the adjusted analyses. For cases in the PCPT trial with a Gleason score of 7, the period prevalence RR was slightly above 1 (1.08) in the unadjusted and below 1 (RR, 0.79 and 0.89, respectively) in the adjusted analyses. RRs were below 1 for Gleason score 2 to 6 cases (range, 0.47-0.64) and above 1 for Gleason score 8 to 10 cases (range, 1.44-1.61) in all the PCPT analyses. In the REDUCE trial, period prevalence RRs were lower than 1 for Gleason score 2 to 6 cases (RR, 0.73) and Gleason score 7 cases (RR, 0.92), and greater than 1 (RR, 1.60) for Gleason score 8 to 10 cases.

Table 2. Period Prevalence by Gleason Score in the PCPT and REDUCE Trials
PCPTFinasteride (N = 4958 With Biopsy)Placebo (N = 5223 With Biopsy)RR (95% CI)
Unadjusted (biopsy Gleason score)Period prevalence  
2-610.0%16.9%0.59 (0.50-0.63)
7-105.8%4.7%1.23 (0.97-1.36)
73.9%3.6%1.08 (0.84-1.25)
8-101.9%1.2%1.61 (1.1-2.1)
Unknown0.8%1.2%0.64
Adjusted based on RP Gleason scorea   
2-69.3%14.6%0.64 (0.51-0.71)
7-107.2%8.3%0.87 (0.68-1.05)
75.8%7.3%0.79 (0.60-0.99)
8-101.4%1.0%1.44 (0.8-2.5)
Adjusted based on prostate volumeb  OR (95% CI)
2-610.0%19.0%0.47 (0.41-0.54)c
7-105.8%5.6%1.03 (0.84-1.26)
73.9%4.4%0.89 (NR)
8-101.9%1.2%1.51 (1.01-2.26)
Unknown0.8%1.2%0.64 (NR)
REDUCEDutasteride (N = 3299 With Biopsy)Placebo (N = 3407 With Biopsy)RR (95% CI)
  • Abbreviations: 95% CI, 95% confidence interval; NR, not reported; OR, odds ratio; PCPT, Prostate Cancer Prevention Trial; REDUCE, Reduction by Dutasteride of Prostate Cancer Events trial; RP, radical prostatectomy; RR, relative risk.

  • a

    Data derived from Pinsky PF, Parnes H, Ford L. Estimating rates of true high-grade disease in the Prostate Cancer Prevention Trial. Cancer Prev Res. 2008;1:182-186.12

  • b

    Data derived from Theoret MR, Ning Y, Zhang J, Justice R, Keegan P, Pazdur R. The risks and benefits of 5α-reductase inhibitors for prostate-cancer prevention. N Engl J Med. 2011;365:97-995; and Cohen YC, Liu KS, Heyden NL, et al. Detection bias due to the effect of finasteride on prostate volume: a modeling approach for analysis of the Prostate Cancer Prevention Trial. J Natl Cancer Inst. 2007;99:1366-1374.13

  • c

    The 95% CI is for the OR of Gleason score 2 to 6 cancer versus not cancer; approximates the 95% CI for the OR of Gleason score 2 to 6 cancer versus not Gleason score 2 to 6 cancer.

Biopsy Gleason score   
2-613.2%18.1%0.73 (0.65-0.83)
75.8%6.3%0.92 (0.76-1.1)
8-100.88%0.55%1.60 (0.9-2.8)

Table 3 shows 1-survival rate estimates (and standard errors) at 13 years derived from the PLCO cohort for different prognostic strata. These rates were used to compute the mortality projections for the 2 trials.

Table 3. 1-Survival Rates (Prostate-Cancer Specific) at 13 Years From PLCO Cohort
 Gleason Score 2 to 6Gleason Score 7Gleason Score 8 to 10Unknown Gleason Score
 Rate (SE)Rate (SE)Rate (SE)Rate (SE)
  • Abbreviations: NA, not applicable; PCPT, Prostate Cancer Prevention Trial; PLCO, Prostate, Lung, Colorectal and Ovarian cancer screening trial; PSA, prostate-specific antigen; REDUCE, Reduction by Dutasteride of Prostate Cancer Events trial; SE, standard error.

  • Based on the PLCO intervention arm cases only except for stage III/IV disease, for which control arm cases were also used.

  • a

    Rate for PLCO Gleason score 2 to 6 cases was used.

  • b

    PLCO Gleason score 2 to 6 and Gleason score 7 cases were combined to produce the rate.

SubstratumRates used for PCPT 
PSA ≤4 ng/mL, T1a-c0.018 (0.013)   
PSA >4 ng/mL, T1a-c0.024 (0.008)   
PSA ≤4 ng/mL, T2a-c0.026 (0.016)   
PSA >4 ng/mL, T2a-c0.058 (0.023)   
Any PSA, T1/2 0.135 (0.038)0.298 (0.065)0.071 (0.04)
T3 (TNM stage III)0.128 (0.070)0.230 (0.131)0.293 (0.098)0.128 (0.070)a
T4/N1/M1 (TNM stage IV)b0.547 (0.10)0.547 (0.10)0.840 (0.06)0.547 (0.01)a
 Rates used for REDUCE 
All0.038 (0.009)0.141 (0.036)0.360 (0.056)NA

Table 4 displays the results of the projected mortality analysis. For unadjusted analysis in the PCPT trial, expected death rates were similar in each treatment arm (14.6 [finasteride] and 14.2 [placebo] per 1000 men; RR, 1.02 [95% CI, 0.85-1.23]). The death rate due to Gleason score 2 to 6 tumors was greater in the placebo arm, whereas the death rate due to Gleason score 8 to 10 tumors was greater in the finasteride arm. The adjusted analyses showed RRs of 0.87 (95% CI, 0.72-1.06) and 0.91 (95% CI, 0.76-1.09) for the RP Gleason score and prostate volume methods, respectively. In the RP Gleason-adjusted analysis, the death rate differential (finasteride vs placebo) for Gleason score 7 disease was −2.0 (8.6 vs 10.6), compared with 0.2 (5.4 vs 5.2) in the unadjusted analysis; the differential also decreased for Gleason score 8 to 10 cases (from 2.1 to 1.2). In the prostate volume-adjusted analysis, the death rate differential also decreased for Gleason score 7 disease (compared with the adjusted analysis). In this (volume) analysis, deaths rates for Gleason score 7 to 10 disease were essentially equal, reflecting the adjusted RR of 1.03 for Gleason score 7 to 10 disease, with minor adjustments for small clinical stage differences between the treatment arms; however, the death rate due to Gleason score 2 to 6 disease was lower in the finasteride arm, resulting in the mortality RR of 0.91.

Table 4. Projected Mortality by Treatment Arm in the PCPT and REDUCE Trials
PCPTFinasteride ArmPlacebo Arm 
 Death Rate per 1000 (% of Deaths)aDeath Rate per 1000 (% of Deaths)aRR (95 % CI)
Original biopsy Gleason score   
2-62.7 (18)4.3 (30) 
75.4 (37)5.2 (37) 
8-105.9 (40)3.8 (27) 
Unknown0.6 (4)1.0 (7) 
Total14.614.21.02 (0.85-1.23)
Adjusted (RP Gleason score)   
2-62.5 (16)4.0 (23) 
78.6 (56)10.6 (60) 
8-104.2 (27)3.0 (17) 
Total15.317.60.87 (0.72-1.06)
Adjusted (prostate volume)   
2-62.7 (18)4.7 (29) 
75.4 (37)6.5 (40) 
8-105.9 (40)4.0 (25) 
Unknown0.6 (4)1.0 (6) 
Total14.616.10.91 (0.76-1.09)
REDUCEDutasteride ArmPlacebo Arm 
Biopsy Gleason scoreDeath Rate per 1000 (% of Deaths)aDeath Rate per 1000 (% of Deaths)aRR (95 % CI)
  • Abbreviations: 95% CI, 95% confidence interval; PCPT, Prostate Cancer Prevention Trial; REDUCE, Reduction by Dutasteride of Prostate Cancer Events trial; RP, radical prostatectomy; RR, relative risk.

  • a

    Death rates based on denominator of all biopsied men.

2-65.0 (30)6.9 (39) 
78.2 (50)8.9 (50) 
8-103.2 (20)2.0 (11) 
Total16.417.70.93 (0.80-1.08)

For the REDUCE trial, the death rates were 16.4 (dutasteride) versus 17.7, for an RR of 0.93 (95% CI, 0.80-1.08 ). The death rate was higher for Gleason score 8 to 10 cases in the dutasteride arm, but lower for the Gleason score 2 to 6 and the Gleason score 7 cases.

DISCUSSION

Similar to most chemoprevention trials, neither the PCPT nor the REDUCE trial had a cancer mortality endpoint, but rather an incidence endpoint. For a slow-growing tumor such as prostate cancer, the length of follow-up required for a mortality endpoint would be considerable, on the order of 15 years to 20 years, making such trials logistically prohibitive under most circumstances. For the PCPT and REDUCE trials, even if the cohorts were followed for a long period, as this analysis has done in a virtual manner, the sample size was still too small to have adequate power with a reasonable hypothesized mortality reduction. Based on the current analyses, assuming a true mortality reduction of 15% and 15 years to 20 years of follow-up from randomization, the power of the study would only be within the range of 20% to 30%.

Realistically, chemoprevention trials with a mortality endpoint are therefore unlikely to be performed for prostate cancer and we are left with using cancer incidence as a surrogate endpoint. However, with this surrogate endpoint unclear in these trials, due to a possible increase in high-grade cancer along with an overall incidence decrease, a modeling exercise that projects future mortality is one approach in the attempt to move from the surrogate endpoint (incidence) to the endpoint of arguably greater interest (mortality).

However, any attempt to extrapolate a mortality effect from an incidence one must be interpreted cautiously. In the current analysis of the PCPT trial, due to widely expressed concerns about the preponderance of cancers (primarily Gleason score 2 to 6 tumors) that were not detected for cause and that may have limited clinical significance, we made efforts to carefully stratify cases by significant prognostic variables, T classification and PSA, in addition to Gleason score, when applying survival rates from the PLCO trial. In the PCPT trial, 78% of finasteride treatment arm and 77% of placebo arm Gleason score 2 to 6 cancers were classified as T1a to T1c, which was modestly higher than the comparable percentage in the PLCO trial (68%). Because of this stratification, the survival rates we applied should reasonably approximate the true rates for the PCPT cohort. In addition, we estimated the RR of mortality, which is less sensitive to survival rate fluctuations than the absolute mortality rates. For the REDUCE trial, because men had generally high (> 4 ng/mL) baseline PSA levels, concerns about clinically insignificant disease were lower, and without stage or PSA data available to us, we stratified survival based only on Gleason score. To assess the potential effect of additional stratification by PSA and tumor stage, we repeated the PCPT analysis using only the Gleason categories (scores 2-6, 7, and 8-10) as survival strata. Although absolute mortality rates were 10% to 20% higher, the RRs were very similar: 0.99 (vs 1.02) for the unadjusted analysis and 0.87 and 0.88 (compared with 0.87 and 0.91), respectively, for the 2 adjusted analyses. Thus further stratification for the REDUCE trial would likely have resulted in little change in the estimated RR for mortality.

Although, we stratified survival by Gleason score, PSA, and disease stage for the PCPT trial, there are other prognostic factors that we were unable to account for because they were not collected in PLCO. The Cancer of the Prostate Risk Assessment (CAPRA) risk assessment tool, which has been validated as a predictor of prostate cancer-specific mortality, uses percentage positive core needle biopsies in addition to Gleason score, PSA, and disease stage (and patient age) in determining a 10-point prognostic index; specifically, 1 point is allotted for percentage positive core needle biopsies > 33%.17 Analysis of the PCPT indicated that for cases of Gleason score 2 to 6 low-risk (T1/T2) disease, there was an approximately 5% differential by treatment arm in cases meeting this threshold, with higher rates noted in the placebo arm. In addition, a 1-point increase in the CAPRA score in low-risk men results in an approximately 2% decrease in the 10-year prostate-specific survival.17 Using these numbers, we were able to effectively further stratify the PLCO survival strata for Gleason score 2 to 6 disease by percentage positive core needle biopsies. The result was essentially no change in the mortality RRs: 1.02 and 0.87, respectively, as before for the unadjusted and adjusted RP Gleason score analysis and 0.90 versus 0.91 for the adjusted volume analysis.

There is a general consensus in the research community that some bias in prostate cancer detection and/or biopsy grading occurred in the PCPT trial, either of which would favor the finasteride treatment arm; however, the magnitude of such a bias and the appropriate methods for adjusting for it are less clear. Using 2 previously published methods for such an adjustment, we obtained RR estimates that were very similar (0.87 and 0.91). Another adjusted analysis, which also used the RP Gleason score results, produced lower incidence RR estimates for Gleason score 7 to 10 and 8 to 10 disease (RRs of 0.73 and 1.25, respectively) than did the RP Gleason score method used in the current study (RRs of 0.87 and 1.44, respectively).9 Applying this analysis to our mortality projections would result in a lower mortality RR than those noted herein. A similar artifact with respect to biopsy grading and the detection of prostate cancer may have also occurred in the REDUCE trial. An analysis of the effects of dutasteride on prostate volume estimated that cancer detection rates were increased by 11% to 17% in men treated with dutasteride compared with placebo.18 Adjusting for such artifacts would reduce the projected mortality RR estimate for dutasteride from that computed in the current study.

Several prior analyses used prostate cancer survival rates to assess the benefits versus harms of finasteride, with life-years saved used as the metric.7-9 Each produced a range of estimates depending on whether Gleason score differences were deemed artifactual as well as other considerations. Grover et al demonstrated per person life-years gained of between 0.02 to 0.2 years and Lotan et al reported similar life-years gained of between 0.35 to 3 months (0.03-0.25 years); the greatest gains were achieved under the assumption that all Gleason score differences were artifactual whereas the lowest were achieved assuming the score differences were as observed and that only cancers detected during for-cause biopsy contributed to mortality.7, 8 Using SEER survival rates, Zeliadt et al demonstrated gains of between 0.006 to 0.04 life-years per person, with the latter derived assuming Gleason score differences were artifactual.9 However, as mentioned earlier, the survival rates used in these analyses were not derived from the experience of highly screened cohorts. For example, Lotan et al used a 15-year cause-specific survival rate for Gleason score 5 to 6 disease of 85% (for men aged 60-64 years) based on a case series of clinically detected cancers, whereas in the current analysis (Table 3), the 13-year cause-specific survival rate for Gleason score 2 to 6 cases (92% of which were Gleason score 5-6 tumors) was 96.2%. Analysis of PLCO and SEER data indicates that for Gleason score 5 to 7 cases the death rate (ie, 1-cumulatitve survival rate) was approximately one-third lower in the PLCO trial compared with what would be expected based on SEER survival rates.

One can convert the deaths rates shown in Table 4 to life-years saved, at least for the 13-year follow-up period used in the current study. For cases with Gleason scores 2 to 6, 7, and 8 to 10, the average life-years lost during this period for a man dying of prostate cancer were 4.0, 4.4, and 6.1 life-years, respectively, based on the PLCO survival rates. Taking the adjusted RP Gleason analysis results as an example, the differential death rate per 1000 (placebo minus finasteride) of 1.5 (ie, 4.0-2.5) for Gleason score 2 to 6 disease translates into a relative life-years gained for men treated on the finasteride arm of 6.0 per 1000 (1.5*4.0). A similar analysis for Gleason score 7 and Gleason score 8 to 10 cases demonstrated relative life-years gained for men treated on the finasteride arm of 8.8 (2.0*4.4) and −7.3 (-1.2*6.1) per 1000, respectively. This sums to 0.0075 life-years gained per person (or 7.5 per 1000), which is 5 to 33 times less than the most optimistic figures from the above-mentioned 3 studies.7-9 These divergent results are explained by the higher survival rates, especially for Gleason score 2 to 6 tumors, used in the current analysis, as well as by differences in how the possible grading artifact is dealt with. Although our RP Gleason score analysis adjusted for bias by treatment arm in grading, it retained differences in the percentage incidence reductions by Gleason score; in contrast, in the above-mentioned studies, percentage incidence reductions were assumed to be constant across Gleason score categories when grade differences were assumed to be artifactual. Differences in the follow-up time periods used could also lead to some differences in the estimates.

As a thought experiment, consider an actual mortality endpoint trial that demonstrated incidence trends similar to those noted in the PCPT and REDUCE trials (ie, overall decrease but possible increase in high-stage disease while receiving the drug) but demonstrated either: 1) a statistically significant mortality reduction; 2) no significant mortality reduction but a point estimate for the mortality RR < 1 and an upper bound only slightly > 1; or 3) a point estimate modestly > 1.0 (regardless of whether it was statistically significant or not). Under the first scenario, despite harm to some individual men, it is likely that most would conclude that the benefits outweigh the harms, whereas for the third scenario, the consensus would likely be strongly against using the drugs. In the second scenario, the benefit is essentially confined to reductions in cancer incidence only, which, given the frequent and serious side effects of standard prostate cancer treatments, is substantial in itself. However, for this to outweigh harms, one would need confidence that there was no true mortality increase or that it was at most negligibly small. With the current modeling exercise, we have in effect scenario 2 (assuming one uses the adjusted PCPT analyses), except that there is further uncertainty due to the finding that it is only a “virtual” trial, with all of the caveats of modeling and extrapolation. Nonetheless, the analysis is useful in attempting to quantify the effects of these drugs on prostate cancer mortality, especially because they are currently approved for the treatment of benign prostatic hyperplasia and are used for that purpose by many men.

Conclusions

The PCPT and REDUCE trials of 5α-reductase inhibitors each demonstrated a significant overall decrease in prostate cancer incidence for the treatment arm along with a trend toward an increase in high-grade disease. Projecting a mortality outcome of these trials as an approach to weighing benefits versus harms demonstrates that there likely would have been at most a small increase in prostate cancer deaths in the treatment arms, and possibly a modest decrease.

FUNDING SUPPORT

No specific funding was disclosed.

CONFLICT OF INTEREST DISCLOSURES

Dr. Andriole is an advisor/consultant for Amarex Clinical Research; Amgen, Inc; Augmenix, Inc; Bayer; Bristol-Myers Squibb; Cambridge Endo; Caris; GlaxoSmithKline; Janssen Biotech, Inc; Myriad Genetics, Inc; Steba Biotech; Ortho Clinical Diagnostics; and Viking Medical. He also acts as an investigator for Johnson & Johnson; Medivation, Inc; and WILEX, Inc, and is an investor in Envisioneering Medical Technologies. Dr. Grubb acts as an investigator for GlaxoSmithKline.

Ancillary