Restricted Mean Survival Time for Analysis and Interpretation of Clinical Trials for Heart Failure Devices

Abstract

Introduction Treatment effects in heart failure (HF) clinical trials are typically summarized as hazard ratios (HR). However, a HR alone without a reference value of the hazard from the control arm is difficult to interpret and is valid only when hazards are proportional throughout the trial. The restricted mean survival time (RMST) describes the mean time free from an event and does not require any model assumption. Comparison of RMSTs between treatment and control arms, either by differences or ratios, is a more intuitive summary of treatment effects than HR. Objective Compare RMST with HR in HF device clinical trials. Methods Time-to-event data was reconstructed from published Kaplan-Meier curves from landmark HF trials. We calculated the RMST-differences and restricted mean time loss (RMTL) ratios between treatment groups as measures of treatment effect, and compared statistical testing results to HR. Results We analyzed 4 clinical trials with a total of 5,186 patients. Statistical significance tests were concordant between HR and RMST or RMTL for all outcomes (Table 1). However, RMST analyses provided additional clinically useful information: for instance, in COAPT, percutaneous mitral valve repair extended each patient's life 1.7 months on average compared to medical therapy alone, over 2 years follow up. In MADIT II, implantable cardioverter defibrillator (ICD) implantation extended each patient's life on average by 2.9 months compared to medical therapy, over the 4 years of the trial, and in MADIT-CRT, cardiac resynchronization therapy added on average 3.2 months to each patient's life compared to ICD alone, over 4 years of follow up. Conclusions RMST achieved similar statistical testing results as HR but provides an intuitive estimate of treatment effect. RMST-based analysis may better communicate treatment effects to patients, to assist in patient-preference discussions and shared decision-making.