Robust policy evaluation from large-scale observational studies.

Area of Science:

• Health Services Research
• Econometrics
• Data Science

Background:

• Current policy decision-making relies on identifying causal relations between interventions and outcomes, often using matching methods.
• One-to-one matching algorithms can produce different causal conclusions from the same data due to multiple assignment options.
• This uncertainty is amplified in large-scale observational studies, hindering reliable policy evaluation.

Purpose of the Study:

• To develop computationally efficient algorithms for causal inference testing with binary outcomes in large-scale observational studies.
• To address the scalability limitations of existing integer programming models for robust causal inference.
• To reduce the computational burden through a proposed robustness condition leveraging optimization model structure.

Main Methods:

• Proposed novel, computationally efficient algorithms for causal inference with binary outcomes.
• Introduced a robustness condition to decrease computational complexity in optimization models.
• Validated algorithms using the Medicare Hospital Readmission Reduction Program (HRRP) and California hospital discharge data (2010-2014).

Main Results:

• The Medicare Hospital Readmission Reduction Program (HRRP) shows a causal relationship with increased non-index readmissions.
• The proposed algorithms demonstrate high scalability for causal relation testing in large observational datasets.
• The new methods significantly reduce computational expense compared to existing integer programming approaches.

Conclusions:

• The developed algorithms provide a scalable and efficient solution for causal inference in large-scale observational studies.
• This research enhances the reliability of policy decision-making by reducing uncertainty in causal effect estimation.
• Findings confirm a causal link between HRRP and increased non-index hospital readmissions, informing healthcare policy.