[Skip to Navigation]
Comment & Response
March 2, 2022

Age-Adjusted Survival Extrapolations—Results May Differ From Those Generated by the Weibull Model—Reply

Author Affiliations
  • 1British Heart Foundation Cardiovascular Research Centre, University of Glasgow, Glasgow, United Kingdom
JAMA Cardiol. 2022;7(5):570. doi:10.1001/jamacardio.2021.6032

In Reply Although Messori and colleagues estimated a smaller increase in event-free survival than we did, the Weibull-based approach they used they has several limitations. They tried to extract data from published figures, which is not as reliable as using original data, and small differences may have an important impact on long-term projections.1 Furthermore, relying on overall summary data, and not individual patient data, means that age-specific information and the corresponding variations in rates over patient lifetimes are not used. As we have previously described,2 our method uses individual patient data for each age, with survival curves estimated for each age, and these combined to give the estimates presented. Thus, we do not extrapolate but rather use the observed data to make the calculations. In our article,3 we used Kaplan-Meier methods to estimate age-specific event rates for patients in each treatment arm and then used age-specific rates to construct estimated survival times over patient lifetime. We have previously shown that our estimates using this method are very close to the actual values that have been observed in long-term trial extension studies.2-4 For example, using data from the period of double-blind randomized treatment in the Studies of Left Ventricular Dysfunction (SOLVD) trial (median follow-up, 3.2 years), we estimated that the mean survival time at 12 years would be 5.7 years (95% CI, 5.2-6.3) for a patient aged 61 years (the mean age of patients randomized was 61 years).2,5 In the extended 12-year follow-up of this trial, the actual median (IQR) survival time observed in the enalapril group was 5.5 (2.4-10.8) years.5 These estimates were also robust to several other assumptions. While no 2 methods of extrapolation will agree, our approach has been externally validated for other treatments using observed data.2-4,6

Add or change institution