Generalizability of Clinical Trial Results: Value of Individual Data versus Aggregate Data

Application Number
Lay Summary

There is growing interest in improving the generalizability of randomized clinical trial results to real-world populations. Due to stringent inclusion/exclusion criteria in clinical trials, there may remain overt differences in some important characteristics between the trial participants and the real-world population for whom the treatment would be indicated (i.e., the target population). These characteristics may affect treatment effects, leading to problems in generalizing trial results to the target population. Several methods of generalizing trial results have been proposed to overcome this limitation; however, these methods require all data of every individual in both the trial and the target population (i.e., individual-level data). Obtaining individual-level data for the trial and the target population may be a resource-intensive undertaking, whereas aggregate-level data (i.e., data at the level of subgroup, such as the total numbers of male and female patients, respectively) can often be easily retrieved from the literature. Little is known about these generalizing methods using aggregate-level data, thus this study aims to evaluate different approaches to generalize trial results to target populations using aggregate-level data, using the JUPITER trial and CPRD data. This study provides new information on use of individual-level versus aggregate-level data in these approaches of generalizing trial results.

Technical Summary

The objective of this study is to explore different approaches to generalize clinical trial results to the target population when only aggregate data or subgroup-specific results are available in either clinical trial or real-world data. We use data from the JUPITER trial and the CPRD, and examine the following three methods of generalizing the trial results. The first method is weighting methods based on predicted probabilities from multivariable models. For a given data source with aggregate data, we simulate a hypothetical cohort which is consisted of fake individual data, and apply these methods to standardize the JUPITER trial results to the CPRD population. We compare the estimates of generalized trial results based on individual data to the estimates based on aggregate data. The second is to estimate the absolute risk reduction for anticipated treatment effect in the target population. We calculate the cardiovascular risk in the CPRD patients without receiving statins, and multiply it by the relative risk reduction observed in the JUPITER trial. The third is to estimate the statin effect by computing a weighted average of the subgroup-specific treatment effect estimates in the JUPITER trial with weights from the population distributions of effect modifiers in the CPRD.

Health Outcomes to be Measured

The primary end point is the occurrence of a first major cardiovascular event, defined as nonfatal myocardial infarction, nonfatal stroke, hospitalization for unstable angina, an arterial revascularization procedure, or confirmed death from cardiovascular causes.


Til Stürmer - Chief Investigator - University of North Carolina at Chapel Hill
Til Stürmer - Corresponding Applicant - University of North Carolina at Chapel Hill
Jin-Liern Hong - Collaborator - University of North Carolina at Chapel Hill
Michelle Jonsson Funk - Collaborator - University of North Carolina at Chapel Hill
Nze Shoetan - Collaborator - Astra Zeneca Inc - USA
Robert LoCasale - Collaborator - Astra Zeneca Inc - USA
Sara Dempster - Collaborator - Astra Zeneca Inc - USA
Stephen Cole - Collaborator - University of North Carolina at Chapel Hill