Median Survival or Mean Survival: Which Measure Is the Most Appropriate for Patients, Physicians, and Policymakers?

Abstract

Introduction Understanding the efficacy of treatments is crucial for patients, physicians, and policymakers. Median survival, the most common measure used in the outcome reporting of oncology clinical trials, is easy to understand; however, it describes only a single time point. The interpretation of the hazard ratio is difficult, and its underlying statistical assumptions are not always met. The objective of this study was to evaluate alternative measures based on the mean benefit of novel oncology treatments. Materials and Methods We reviewed all U.S. Food and Drug Administration (FDA) approvals for oncology agents between 2013 and 2017. We digitized survival curves as reported in the clinical trials used for the FDA approvals and implemented statistical transformations to calculate for each trial the restricted mean survival time (RMST), as well as the mean survival using Weibull distribution. We compared the mean survival with the median survival benefit in each clinical trial. Results The FDA approved 83 solid tumor indications for oncology agents between 2013 and 2017, of which 27 approvals based on response rates, whereas 49 approvals were based on survival endpoints (progression‐free survival and overall survival). The average improvement in median overall survival or progression‐free survival was 4.6 months versus 3.6 months improvement in the average RMST and 6.1 months improvement in mean survival using Weibull distribution. Conclusion Mean survival may supply valuable information for different stakeholders. Its inclusion should be considered in the reporting of prospective clinical trials. Implications for Practice Mean survival may supply valuable information for different stakeholders. Its inclusion should be considered in the reporting of clinical trials.

Impact of 2 Versus 1 Colostrum Meals on Failure of Transfer of Passive Immunity, Pre-Weaning Morbidity and Mortality, and Performance of Dairy Calves in a Large Dairy Herd

“…Unfortunately, however, median survival times, which are not subjected to this bias, could not be calculated for all the feeding groups and diseases as less than 50% of the population developed the condition. However, mean survival still supplies valuable information when comparing the performance of 2 groups [ 40 ].…”

Inferring median survival differences in general factorial designs via permutation tests

“…This gives rise to study other effect sizes than hazard ratios in factorial designs. In this paper, we considered MSTs as they are frequently reported, easy to understand, 13 and have therefore been propagated recently. 17 Furthermore, in addition to a mere reporting of survival chances at specific time points of clinical interest, medians add valuable information: survival medians are quantities given in units of time and they provide an estimate that is of personal interest to many patients with a severe disease.…”

“…As this can be hard to verify or simply wrong, some alternatives to hazard ratios as effect sizes are desired that provide “ clinically meaningful outcomes for (…) patients ” and for “ discussing the value of clinical trials with patients ”. 7 Proposed alternatives cover average hazard ratios, 8,9 concordance and Mann–Whitney effects, 10–12 the (restricted) mean survival, 13,14 or median survival times (MSTs). 15–17…”

“…MSTs are “ easy to understand ” 13 and therefore frequently reported in medical studies. In fact, they are “ the most common measure used in the outcome reporting of oncology clinical trials ” 13 and their application has recently been propagated. 17 Besides their use as a descriptive tool, they are also applied for inferential conclusions.…”

Endpoint surrogacy in oncology Phase 3 randomised controlled trials

“…milestone survival 7 ) and/or newer statistical methods using the landmark analysis 9 or restricted mean survival time. 10 Therefore, we suggest current validation studies to consider any possible aggregated measures or treatment effects in their evaluation of validity.…”