Self-matched learning to construct treatment decision rules from electronic health records

Area of Science:

• Health Informatics
• Biostatistics
• Precision Medicine

Background:

• Electronic health records (EHRs) offer valuable real-world data for precision medicine, complementing traditional randomized controlled trials.
• Inferring individualized treatment rules (ITRs) from EHRs is challenged by unmeasured confounding due to the absence of randomization.

Purpose of the Study:

• To propose a novel self-matched learning method to infer optimal ITRs from EHR data.
• To mitigate unmeasured time-invariant confounding inherent in observational health data.
• To enhance the development of precision medicine strategies using real-world data.

Main Methods:

• Developed a self-matched learning approach inspired by the self-controlled case series (SCCS) design.
• Implemented within-patient matching (self-controlled matching) to address time-invariant confounding.
• Constructed a within-subject matched value function for optimizing ITRs.

Main Results:

• Self-matched learning demonstrated comparable performance to existing methods in the absence of unmeasured confounders.
• The proposed method significantly outperformed alternatives when unobserved time-invariant confounders were present.
• Sensitivity analyses confirmed the robustness of the self-matched learning approach across various scenarios.

Conclusions:

• Self-matched learning effectively mitigates unmeasured confounding in EHR data for ITR inference.
• The method shows promise for advancing precision medicine by enabling more accurate ITRs.
• Application to type 2 diabetes EHRs indicated improved patient outcomes through optimized treatment decisions.