MP08-03: Progression-free survival as surrogate endpoint in high-risk non-muscle invasive bladder cancer studies: results from a machine learning-based analysis of a large multi-institutional database
Introduction: Progression-free survival (PFS) is often used as surrogate endpoint in clinical trials investigating treatment effectiveness in patients affected by high-risk non-muscle invasive bladder cancer patients (HR-NMIBC). However, the goodness of such surrogate endpoint and its association with overall survival (OS) is challenged by conflicting results showing lack of OS advantage, even in presence of PFS improvement. We aimed to test the effect of time-to-progression on OS in a large cohort of HR-NMIBC treated with intravesical BCG. Methods: We included patients at first diagnosis of T1 HR-NMIBC after TURB. All procedures were performed at 18 different tertiary institutions between 2002 and 2012. Progression was defined as the time form HR-NMIBC to muscle invasive bladder cancer (MIBC) development. All patients who experienced a progression underwent radical cystectomy. Machine learning approach, based on random survival forest (RSF) was used to rank covariates based on OS prediction. Then survival tree data mining technique was used to obtain classes based on time-to-progression. Finally, Cox-regression models were deployed. Results: Overall, 1510 patients were included. During a median follow up of 49 months, 485 patients progressed to MIBC, while 163 patients died. The median time to progression was 82 months. In RSF time-to-progression and age were the most predictive covariates of OS. The C-index resulted to be 0.814 on the training set and 0.726 on the test set. Survival tree including these two covariates defined 5 groups of risk: PFS = 62.5 months and age < 77.5 years (reference), PFS = 62.5 months and age = 77.5 years (HR: 9.0, p<0.001), PFS between 10.5 and 62.5 months and age < 79.5 years (HR: 14.3, p<0.001), PFS <10.5 months and age <79.5 (HR: 44.9, p<0.001) and PFS < 62.5 month and age = 79.5 years (HR:58.7, p<0.001). In multivariable Cox’s regression models accounting for progression status as time-dependent covariate, longer PFS (as continuous covariate) was associated with longer OS (HR: 0.9, p<0.001). Results were virtually the same after PFS stratification (PFS= 10.5 months as reference): PFS between 10.5 and 62.5 months (HR: 0.4, p<0.001) and PFS =62.5 months (HR: 0.2, p<0.001). Conclusions: Our results suggest that PFS represent a good surrogate endpoint of OS considering its strong independent predictor status and the high accuracy. Future studies evaluating the efficacy of HR-NMIBC treatments can safely rely on PFS as surrogate of OS. SOURCE OF Funding: .
