Project Summary
Patients’ health problems are increasingly complex as a result of comorbid conditions, socioeconomic disparities, and behavioral and demographic dissimilarities. It is imperative to treat complex patients effectively, as many are high users of healthcare services. With increasing numbers and types of treatments for many conditions, patients likely experience differential responses to any particular treatment, or heterogeneity of treatment effectiveness (HTE). A key focus of PCOR methods is to match patients with the most effective treatments to improve treatment efficacy and adherence in the presence substantial HTE.
To reach this goal of recommending better treatments for patients, popular approaches model or predict patient outcomes under the comparative treatments and then match treatments for patients according to the outcome model. While this is one feasible way, it is not very effective. The main reason is that outcomes are affected by both prognostic variables and treatment moderators. Therefore, outcome modeling needs to model both types of variables correctly to avoid poor prediction—even though, in the end, only treatment moderators are used to guide treatment selection. The outcome modeling approaches require correct modeling of prognostic effects from hundreds of variables that may not be of interest for discovering treatment moderators. We have developed award-winning novel methods that focus on estimating only covariate-treatment interactions. By disengaging the covariate main effects from covariate-treatment interactions, our methods can lead to more parsimonious models and, hence, more robust results. Our method amounts to a weighted classification approach for treatment choices, where patient outcomes act as weights instead of as modeling targets. Consequently, our approach can be handled by well-known statistical and mainstream machine learning tools.
Both our preliminary work and existing SGA methods have a narrow focus on one study setting. First, current publications focus on single-time-point outcomes. Second, current methods hinge on using secondary data sets from individual clinical trials. Our proposal intends to enhance the descriptive SGA framework in each of these aspects. Aim 1 of the proposal investigates treatment scoring systems development for multiple and longitudinal outcomes, using data from individual clinical trials. Aim 2 of the proposal investigates treatment scoring systems development, using data sets from multiple clinical trials. Aim 3 of the proposal investigates treatment scoring systems development, using large-scale observational studies. In all aims, we will seek inputs from the stakeholder panel, which includes patients, a patient advocate, a caregiver, clinicians, and a researcher. We have a detailed engagement and dissemination and implementation plan to outreach our methodology and results to stakeholders and increase its impact.
Chen S, Tian L, Cai T, et al., A general statistical framework for subgroup identification and comparative treatment scoring, Biometrics (February 2017).
Zhao SG, Chang SL, Spratt DE, et al., Development and validation of a 24-gene predictor of response to postoperative radiotherapy in prostate cancer: a matched, retrospective analysis, The Lancet Oncology (November 2016).
Loh WY, Fu H, Man M, et al., Identification of subgroups with differential treatment effects for longitudinal and multiresponse variables, Statistics in Medicine (November 2016).
Project Details
Information