OBJECTIVES : Subgroup analyses are vital components of health technology assessments but randomized controlled trials (RCTs) do not commonly report survival distributions for subgroups. This study developed an analytical framework to elicit unreported subgroup-specific survival information from aggregate RCT data.

METHODS : Assuming exponentially-distributed subgroup survival times, we developed an optimization model which aimed to approximate the restricted mean survival time (RMST) for the overall population via weighted average of the RMSTs of two mutually exclusive and exhaustive subgroups in each arm. Reported hazard ratios from the forest plots between the arms were used to enforce the relationship among subgroups’ hazard rates in the model. The performance of the model was tested in a case study consisting of 8 RCTs in advanced stage gastrointestinal tumors (3 in hepatocellular carcinoma, 3 in gastric/gastroesophageal junction/esophageal cancer and 2 in colorectal cancer) which also reported Kaplan-Meier (KM) curves for overall survival (OS) for 40 subgroups in total. For each subgroup, predicted median survival, OS rates and the RMSTs were compared against the actual OS rates, median survival and the RMSTs as well as their 95% confidence intervals (CIs).

RESULTS : Predicted median survivals and RMSTs were within the 95% CIs of the reported values in 32 and 35 out of 40 subgroups, respectively. In 8 of the subgroups, the gap between the estimated RMSTs from the predicted and the reported survival curves was less than 5%. In 6 of the subgroups, predicted survival curves laid within the 95% CIs of reported KM-curves in at least 90% of the time. Across all subgroups, on average, predicted survival curves laid within the 95% CIs of reported KM-curves in 71% of the time.

CONCLUSIONS : Our study offers a useful and reliable approach for extracting subgroup-specific survival from aggregate RCT data to better inform economic and comparative effectiveness analyses for subgroups.