Assessing Heterogeneity in Treatment Effects (HTE)

• Personalized treatment decisions, which hold great potential for improving health outcomes for patients of varying characteristics, involve the assessment and estimation of individualized or conditional treatment effects.

• However, using data to make these decisions in a reliable, precise, and generalizable way is challenging when sample sizes are low and/or are not representative of the target population. Leveraging the information provided by multiple studies with unconfounded treatment assignment allows for combining datasets to improve the power to estimate heterogeneous treatment effects.

• In this website, we provide a repository of resources, including relevant literature and newly developed tools, designed to examine treatment effect heterogeneity when data from multiple sources is available, leveraging recent machine learning methods.

Literature

Articles we have published, developing and demonstrating novel methods to assess treatment effect heterogeneity when combining data from multiple sources, include:

• Brantner, C. L., Nguyen, T. Q., Tang, T., Zhao, C., Hong, H., & Stuart, E. A. (2024). Comparison of methods that combine multiple randomized trials to estimate heterogeneous treatment effects. Statistics in medicine, 43(7), 1291-1314.
• Brantner, C. L., Chang, T. H., Nguyen, T. Q., Hong, H., Di Stefano, L., & Stuart, E. A. (2023). Methods for integrating trials and non-experimental data to examine treatment effect heterogeneity. Statistical science: a review journal of the Institute of Mathematical Statistics, 38(4), 640.
• Brantner, C. L., Nguyen, T. Q., Parikh, H., Zhao, C., Hong, H., & Stuart, E. A. (2025). Precision Mental Health: Predicting Heterogeneous Treatment Effects for Depression through Data Integration. arXiv preprint arXiv:2509.04604.
• Brantner, C. L., Yu, W., Zhao, C., Jeon, K., Ringlein, G. V., Wang, Q., … & Hong, H. (2025). The challenges of integrating diverse data sources: A case study in major depression. Health Services and Outcomes Research Methodology, 1-23.
• Parikh, H., Nguyen, T. Q., Stuart, E. A., Rudolph, K. E., & Miles, C. H. (2025). A Cautionary Tale on Integrating Studies with Disparate Outcome Measures for Causal Inference. arXiv preprint arXiv:2505.11014.
• Wang, Q., & Hong, H. (2024). Bayesian hierarchical models with calibrated mixtures of g-priors for assessing treatment effect moderation in meta-analysis. arXiv preprint arXiv:2410.24194.
• Hua, K., Wang, X., & Hong, H. (2025). Network Meta‐Analysis of Time‐to‐Event Endpoints With Individual Participant Data Using Restricted Mean Survival Time Regression. Biometrical Journal, 67(1), e70037.
• Hua, K., Wojdyla, D., Carnicelli, A., Granger, C., Wang, X., & Hong, H. (2025). Network Meta‐Analysis With Individual Participant‐Level Data of Time‐to‐Event Outcomes Using Cox Regression. Statistics in Medicine, 44(5), e70027.

Software (R)

Software tools we have developed include:

• Multicate: An R package designed to estimate the conditional average treatment effect (CATE) across multiple studies, and to predict the CATE in a target population of interest. This package is mainly designed to give various estimation and aggregation methods when combining multiple studies and assess heterogeneous treatment effects across studies with diverse visualization tools.

  • Background
  • Application
     

• ShinyApp: A visualization tool that summarizes all simulation results from scenarios discussed in Tang et al., 2025 (in preparation). The app compares multiple methods for estimating causal effects and identifying effect modifiers, including ridge, lasso, adaptive lasso, linear mixed-effects model, Bayesian linear mixed-effects model, Bayesian stochastic search variable selection, Bayesian lasso, Bayesian mixed-effects model with a horseshoe prior, and Bayesian mixed-effects model with a regularized horseshoe prior. Performance is evaluated using metrics such as prediction MSE, treatment effect MSE, effect modifier MSE, main effect MSE, bias, and standard error.

  • ShinyApp github code
     

• CATEDisplays: A new R package designed to seamlessly 1) estimate conditional average treatment effect (CATE) coming from one or more sites and/or trials, 2) test for treatment effect heterogeneity, 3) visualize results. This package is primarily designed for researchers evaluating evidence from multiple trials and interested in potentially differential treatment effects across different subpopulations, and includes a sample tutorial and comprehensive documentation.

• CATE MultRCT: A public repository including R code for performing machine learning methods that estimate the conditional average treatment effect (CATE) by combining data from multiple RCTs.