ITR

ITR generates individualized treatment rules for precision medicine by directly estimating decision rules that maximize expected clinical reward from high-dimensional clinical and observational datasets using the ITR-Forest decision tree and random forest framework.

Key Features:

• Individualized Treatment Rules (ITR): Generates personalized treatment decision trees and random forests tailored to individual patient profiles using the ITR-Forest implementation.
• Algorithmic Innovation: Implements a novel reward function and an innovative decision tree algorithm focused on maximizing expected clinical reward and directly estimating ITRs.
• Ensemble ITR Random Forests: Aggregates multiple decision trees into an ITR random forest to improve robustness and accuracy relative to single-tree rules.
• Interpretability: Produces decision-tree-based rules that provide interpretable treatment recommendations for clinical investigation.
• Soft Probability Outputs: Outputs soft probability estimates for treatment recommendations reflecting the strength of the decision rule.
• Implementation in R: Implemented in the R statistical environment.
• Simulation-based Performance Assessment: Uses simulations to assess the performance of both ITR-Forest and single-tree ITR methods.

Scientific Applications:

• Randomized Controlled Trials: Applied to an RCT of 1,385 patients with diabetes to generate individualized treatment recommendations from controlled experimental data.
• Electronic Medical Records (EMR) Cohorts: Applied to an observational EMR cohort of 5,177 diabetic patients to derive ITRs from real-world clinical data.
• High-dimensional Clinical and Observational Studies: Designed for use with large-scale, high-dimensional datasets common in precision medicine research.

Methodology:

Methodology comprises a novel reward function and decision tree algorithm to directly estimate ITRs, aggregation into ITR random forests (ensemble), and simulation-based performance assessment.