Confounder Adjustment Using the Disease Risk Score: A Proposal for Weighting Methods.

Area of Science:

• Epidemiology
• Biostatistics
• Causal Inference

Background:

• Propensity score analysis is standard for confounding in nonrandomized studies but relies on strict assumptions.
• The disease risk score (DRS) offers an alternative, relaxing some assumptions.
• Traditional DRS methods like matching have arbitrary choices and computational demands.

Purpose of the Study:

• Introduce novel weighting methods for the disease risk score (DRS).
• Evaluate the performance of these new methods against traditional matching techniques.
• Demonstrate the practical application of DRS weighting in real-world case studies.

Main Methods:

• Developed two weighting approaches for DRS: inverse probability weighting and target distribution weighting.
• Compared weighting methods to matching using bias, efficiency (mean squared error), and computational speed metrics.
• Applied methods to case studies on multiple sclerosis and stroke patient data.

Main Results:

• Weighting methods demonstrated comparable bias reduction to matching.
• Weighting approaches significantly outperformed matching in efficiency and computational speed (up to >870x faster).
• Successful implementation illustrated in multiple sclerosis and stroke case studies.

Conclusions:

• Weighting methods provide a computationally efficient and effective alternative to DRS matching for causal inference.
• These new methods relax assumptions and reduce arbitrary choices in confounding adjustment.
• The presented techniques enhance the utility of the DRS for analyzing observational data.