Tomaschitz, R. (2026). Time-dependent hazard ratios in breast cancer clinical trials, Next Research 9, 101713, DOI: 10.1016/j.nexres.2026.101713
Abstract ScienceDirect
Multiply broken power-law densities are employed as cumulative hazard functions for the analytic modeling of Kaplan-Meier (KM) data sets from oncology randomized clinical trials. These adaptive densities are applied to KM data sets for disease-free survival (DFS) from the IBCSG 22-00 trial, which evaluated maintenance therapy with cyclophosphamide and methotrexate for HER2-negative early breast cancer. Also studied are KM data sets for overall survival (OS) and progression-free survival (PFS) from the CLEOPATRA trial, which examined combination therapy with pertuzumab, trastuzumab, and docetaxel for HER2-positive metastatic breast cancer. The multiparameter survival functions of the experimental and control trial arms are regressed and compared with Weibull fits. The uniform local accuracy of the regressed distributions over the entire empirical time range determined by the uncensored event times is demonstrated by residual plots. Confidence bands of the survival and cumulative hazard functions are calculated as well as 95% confidence intervals for selected percentiles. The time evolution of the cumulative-hazard ratio of the trial arms is compared with the constant hazard ratio from the semiparametric Cox regression model.
description: Roman Tomaschitz (2026) Time-dependent hazard ratios in breast cancer clinical trials, Next Res. 9, 101713.
Keywords: Multiparametric survival functions; Cumulative hazard functions and varying hazard ratios; Confidence bands and time-dependent p-values; Nonlinear multiparameter regression; Kaplan-Meier data sets
Highlights
Adaptive multiparameter distributions are employed for the continuum modeling of Kaplan-Meier (KM) data sets from randomized clinical trials in oncology.
The cumulative hazard functions are structured as multiply broken power-law densities depending on a variable number of parameters.
The survival functions of the experimental and control arms of two breast cancer trials are regressed, based on KM data for disease-free, progression-free and overall survival.
Time-dependent hazard ratios of the trial arms are calculated from the regressed survival functions, including confidence bands and p-values.
Survival rates and selected percentiles are obtained, and analytic extrapolation of the survival functions and their confidence bands is attempted.
download full-text article (PDF)