A Comparison of Network Meta Analysis Using Hazard Ratios and Rmst in Clinical Data
TL;DR
Few of the alternate approaches to Hazard Ratio based Network Meta-Analysis are demonstrated by firstly reconstructing the time-to-event dataset from Kaplan Meier curves from results of clinical trials available in public domain using relevant software and then recomputing the restricted mean survival time (RMST) estimates as well as parametric estimates assuming exponential, Weibull and Gompertz distributions to be used respectively.
Abstract
Hazard Ratio based Network Meta-Analysis (NMA) has gained popularity in evidence synthesis in data with time-to-event outcomes. However, the most commonly used Cox-Proportional Hazard ratio suffers due to lack of baseline information utilization. The proportionality assumption, which assumes that ratio of the hazards between any two treatment arms or strata is constant over time for the entire duration of study, is also violated in most Cox-regression models. In the current work we demonstrate few of the alternate approaches by firstly reconstructing the time-to-event dataset from Kaplan Meier curves from results of clinical trials available in public domain using relevant software such as ‘IPDfromKM’ and ‘DigitizeIt’ and then recomputing the restricted mean survival time (RMST) estimates as well as parametric estimates assuming exponential, Weibull and Gompertz distributions to be used respectively, each in a separate NMA. The methods are compared and the advantages of each approach is highlighted. For illustration purpose, Kaplan Meier based results from four trials of Advanced Renal Cell Cancer, are used for analysis. The results of Cox-Proportional Hazard Ratio were compared with parametric distribution-based Hazard Ratio obtained in our analysis. The RMST estimates-based analysis provided additional insights. Based on the NMA, the comparative treatment landscape is provided in terms of a treatment ranking of the included combination therapies.