Metastasis-Free Survival Is a Strong Surrogate of Overall Survival in Localized Prostate Cancer

Abstract

Purpose Adjuvant therapy for intermediate-risk and high-risk localized prostate cancer decreases the number of deaths from this disease. Surrogates for overall survival (OS) could expedite the evaluation of new adjuvant therapies. Methods By June 2013, 102 completed or ongoing randomized trials were identified and individual patient data were collected from 28 trials with 28,905 patients. Disease-free survival (DFS) and metastasis-free survival (MFS) were determined for 21,140 patients from 24 trials and 12,712 patients from 19 trials, respectively. We evaluated the surrogacy of DFS and MFS for OS by using a two-stage meta-analytic validation model by determining the correlation of an intermediate clinical end point with OS and the correlation of treatment effects on both the intermediate clinical end point and OS. Results Trials enrolled patients from 1987 to 2011. After a median follow-up of 10 years, 45% of 21,140 men and 45% of 12,712 men experienced a DFS and MFS event, respectively. For DFS and MFS, 61% and 90% of the patients, respectively, were from radiation trials, and 63% and 66%, respectively, had high-risk disease. At the patient level, Kendall's τ correlation with OS was 0.85 and 0.91 for DFS and MFS, respectively. At the trial level, R² was 0.86 (95% CI, 0.78 to 0.90) and 0.83 (95% CI, 0.71 to 0.88) from weighted linear regression of 8-year OS rates versus 5-year DFS and MFS rates, respectively. Treatment effects-measured by log hazard ratios-for the surrogates and OS were well correlated ( R², 0.73 [95% CI, 0.53 to 0.82] for DFS and 0.92 [95% CI, 0.81 to 0.95] for MFS). Conclusion MFS is a strong surrogate for OS for localized prostate cancer that is associated with a significant risk of death from prostate cancer.

Fig 1.

(A-D) Kaplan-Meier estimates of end points (A) overall survival (OS) and disease-free survival (DFS), (B) disease-specific survival (DSS) and time to disease recurrence (TDR), (C) OS and metastasis-free survival (MFS), and (D) DSS and time to metastasis (TTM). Median follow-up was 10 years.

Fig 2.

Estimated hazard across times (A) overall survival (OS) and disease-free survival (DFS), (B) disease-specific survival (DSS) and time to disease recurrence (TDR), (C) OS and metastasis-free survival (MFS), and (D) DSS and time to metastasis (TTM).

Fig 3.

Overall survival (OS) or disease-specific survival (DSS) rate at 8-year versus surrogate end points at 5 years: (A) 8-year OS versus 5-year disease-free survival (DFS), (B) 8-year disease-specific survival (DSS) versus 5-year time to disease recurrence (TDR), (C) 8-year OS versus 5-year metastasis-free survival (MFS), (D) 8-year DSS versus 5-year time to metastasis (TTM). All rates were Kaplan-Meier estimates by trial and treatment arm. Circle size and regression were weighed by inverse variance of the 5-year rate estimate for the surrogates.

Fig 4.

Treatment effects (hazard ratio [HR]) on overall survival (OS) or disease-specific survival (DSS) versus treatment effects on surrogates: (A) OS HR versus disease-free survival (DFS) HR, (B) DSS HR versus time to disease recurrence (TDR) HR, (C) OS HR versus metastasis-free survival (MFS) HR, (D) DSS HR versus time to metastasis (TTM) HR. HRs were estimated from Cox proportional hazards regression model for each study, and values were natural logarithm transformed. Circle size and regression were weighed by inverse variance of log(HR) estimates for surrogates. STE, surrogate threshold effect.

Fig 5.

Total study duration required in the study designs using metastasis-free survival (MFS) hazard ratios (HRs) and testing surrogate threshold effect (STE; solid line) or using predicted overall survival (OS) HRs from weighted linear regression (dashed line). MFS would be the preferred primary end point for HR(OS) of < 0.70, whereas OS would be the preferred primary end point for HR(OS) of > 0.72 (vertical solid lines). Design assumptions include 5-year MFS and OS rates of 0.79 and 0.84 (hazard, 0.04714 and 0.03487 under exponential distribution), respectively, 5 years of accrual period, and type I error of 0.025 (one-sided) and type II error of 0.20.