The lay abstract featured today (for Using Propensity Score Weighting to Enhance the Operating Characteristics of Power Prior in Leveraging External Data to Augment a Traditional Clinical Study by Heng Li, Wei-Chen Chen, Chenguang Wang, Nelson Lu, Changhong Song, Ram Tiwari, Gregory Alexander, Yunling Xu, Lilly Q. Yue) is from Pharmaceutical Statistics with the full article now available to read here.

How to Cite

Li, H., Chen, W.-C., Wang, C., Lu, N., Song, C., Tiwari, R., Alexander, G., Xu, Y. and Yue, L.Q. (2025), Using Propensity Score Weighting to Enhance the Operating Characteristics of Power Prior in Leveraging External Data to Augment a Traditional Clinical Study. Pharmaceutical Statistics, 24: e2471. https://doi.org/10.1002/pst.2471

Lay Abstract

Clinical studies often need large amounts of data to produce reliable results, but collecting new data can be expensive and time-consuming. One promising approach is to use data from previous studies or real-world sources, such as patient registries, to supplement ongoing studies. However, differences between patients in the current study and those in the external data can lead to biased conclusions if not properly addressed.

This paper introduces a new statistical method—called the propensity score-weighted power prior—that improves how external data can be combined with new study data. The method builds on the “power prior,” a well-known Bayesian approach that allows researchers to control how much influence external data have on study results. It also incorporates propensity score weighting, a common technique used to adjust for differences in patient characteristics between data sources.

By weighting patients in the external data according to how similar they are to those in the current study, this new method greatly reduces bias and the chance of prior-data conflict, and improves the accuracy of study findings. Through computer simulations, we show that the proposed approach performs better than the traditional power prior, especially when the two data sources differ in key characteristics. The new method also provides credible intervals (the Bayesian counterpart to confidence intervals) with very reliable coverage probabilities.

We illustrate the use of the method in a realistic example, where external data are used to strengthen the control arm of a randomized clinical trial for a medical device. The proposed approach is simple to implement and can help make better use of existing data while maintaining the scientific integrity of new clinical studies.

Finally, we point out that the scope of proposed method is not limited to clinical studies. It can be applied in any scenario where the method of power prior can be used, and covariate information is available.