A doubly robust approach for cost-effectiveness estimation from observational data.

Area of Science:

• Health Economics
• Biostatistics
• Machine Learning in Healthcare

Background:

• Estimating cost-effectiveness measures (incremental cost-effectiveness ratio, net monetary benefit) is complex due to informative censoring and data skewness.
• Observational claims data for medical costs and survival require accounting for potential confounders.
• Existing methods may not adequately handle complex data structures and confounding in cost-effectiveness analyses.

Purpose of the Study:

• To propose a novel, doubly robust, and unbiased estimator for cost-effectiveness.
• To incorporate cost history and time-varying covariates into cost-effectiveness estimation.
• To enhance prediction accuracy using an ensemble machine learning approach for cost and propensity score models.

Main Methods:

• Development of a doubly robust estimator based on propensity scores.
• Utilizing an ensemble machine learning approach for improved parametric and non-parametric model predictions.
• Validation through simulation studies assessing performance under model mis-specification.

Main Results:

• The proposed doubly robust approach demonstrates good performance even when either the propensity score or outcome model is mis-specified.
• Simulation studies confirm the robustness and accuracy of the novel estimator.
• The method was successfully applied to a real-world cost-effectiveness analysis comparing CT vs. chest X-ray for lung cancer surveillance.

Conclusions:

• The novel doubly robust estimator provides a reliable method for cost-effectiveness analysis with complex observational data.
• The approach effectively handles informative censoring, data skewness, and confounding.
• This methodology offers improved accuracy for healthcare economic evaluations, exemplified by the lung cancer surveillance case study.