A Meta-Learner Framework to Estimate Individualized Treatment Effects for Survival Outcomes.

Area of Science:

• Biostatistics
• Machine Learning
• Precision Medicine

Background:

• Precision medicine requires understanding patient-specific treatment effects (ITE) to tailor therapies.
• Large-scale genetic and clinical data enable more accurate ITE estimation.
• Machine learning in counterfactual frameworks shows promise for analyzing complex health data.

Purpose of the Study:

• To extend meta-learning approaches for estimating individualized treatment effects (ITE) with survival outcomes.
• To evaluate the performance of T-learner and X-learner meta-learning algorithms combined with various machine learning models.
• To identify patient risk factors contributing to treatment heterogeneity using ITE estimates.

Main Methods:

• Utilized T-learner and X-learner meta-learning algorithms.
• Integrated machine learning models: random survival forest, Bayesian accelerated failure time model, and survival neural network.
• Employed the Boruta algorithm for risk factor identification and simulations for performance comparison.

Main Results:

• Evaluated meta-learning algorithms for ITE estimation in survival data.
• Provided practical guidelines for applying these methods in randomized clinical trials (RCTs).
• Demonstrated application in an age-related macular degeneration (AMD) trial.

Conclusions:

• Meta-learning approaches effectively estimate individualized treatment effects for survival outcomes.
• The methods can guide personalized treatment recommendations in precision medicine.
• Identified key risk factors influencing treatment heterogeneity in patient populations.