Power Priors for Leveraging Historical Data: Looking Back and Looking Forward

• Published in: Journal of Data Science Vol. 23; no. 1; pp. 1 - 30
• Main Authors: Chen, Ming-Hui, Guan, Zhe, Lin, Min, Sun, Max
• Format: Journal Article
• Language: English
• Published: 中華資料採礦協會 01.01.2025
• Subjects: discounting parameters; propensity score based power priors; partial borrowing power priors; Bayesian design of clinical trials; borrowing-by-parts power priors; informative priors
• ISSN: 1683-8602; 1680-743X; 1683-8602
• DOI: 10.6339/24-JDS1161

Historical data or real-world data are often available in clinical trials, genetics, health care, psychology, environmental health, engineering, economics, and business. The power priors have emerged as a useful class of informative priors for a variety of situations in which historical data are available. In this paper, an overview of the development of the power priors is provided. Various variations of the power priors are derived under a binomial regression model and a normal linear regression model. The development of software on the power priors is also briefly reviewed. Throughout this paper, the data from the Kociba study and the National Toxicology Program study as well as the data from the Alzheimer's Disease Neuroimaging Initiative (ADNI) study are used to demonstrate the derivations of the power priors and their variations. Detailed analyses of the data from these studies are carried out to further demonstrate the usefulness of the power priors and their variations in these real applications. Finally, the directions of future research on the power priors are discussed.