CAUSAL MACHINE LEARNING FOR HEALTH INSURANCE COVERAGE EFFECTS: FROM INDIVIDUAL TREATMENT EFFECT ESTIMATION TO POLICY OPTIMIZATION USING REAL-WORLD DATA

Author(s)

Niki Kiriakidou, PhD1, Ioannis E. Livieris, PhD1, Thodoris Kotsilieris, PhD1, Charalampos Tzanetakos, MSc2, George Gourzoulidis, PhD3.
1Department of Business and Organizations Administration, University of the Peloponnese, Kalamata, Greece, 2Health Through Evidece, Athens, Greece, 3Health Through Evidence, Athens, Greece.
OBJECTIVES: To estimate the causal effect of health insurance coverage on annual medical expenditure among working-age U.S. adults using causal machine learning, and to evaluate the transferability of this framework to health systems seeking to optimize benefit design and resource allocation.
METHODS: Data from the Medical Expenditure Panel-Survey 2022 were used to construct a causal inference framework targeting 7,290 working-age adults (25-64), with positivity enforced through age restriction and propensity score trimming. Causal assumption plausibility was assessed through IPW-weighted covariate balance diagnostics, placebo permutation testing, and E-value sensitivity analysis. Individual-level CATEs were estimated using C-XGBoost and were aggregated to derive the population Average Treatment Effect (ATE) and an IPW-weighted policy-value. Higher expenditure among the insured was interpreted as utilization gains from removal of financial barriers, not wasteful spending.
RESULTS: Insurance coverage was associated with an ATE of +$818/person (95% CI: $531-$1,065) in annual medical expenditure. Under the coverage expansion policy, the IPW-weighted expected annual expenditure was $5,694 (95% CI: $4,579-$7,187). Substantial heterogeneity was identified (ITE range: −$73,377 to +$46,600; extreme values reflect variation in baseline health need and unmet demand), with the top quintile showing 4.65× greater gains than the bottom ($8,336 vs. $1,794). Causal assumption plausibility was corroborated through IPW-weighted SMD=0.085, placebo permutation ratio of 19.1×, E-value of 3.11 and sensitivity analysis results robust through Γ=3.0.
CONCLUSIONS: This study demonstrates a reproducible pipeline transforming observational data into individualized causal estimates with direct policy implications. Although US-derived, the proposed framework is directly applicable to European payers and HTA bodies evaluating coverage expansion and risk-stratified reimbursement. The pronounced heterogeneity challenges uniform policy design, demanding benefit-guided resource allocation. Operationalized through CausalCare, a publicly accessible tool generating patient-specific estimates from routine inputs, this framework offers a transferable template for value-based coverage design and comparative effectiveness research.

Conference/Value in Health Info

2026-11, ISPOR Europe 2026, Vienna, Austria

Value in Health, Volume 29, Issue 12S

Code

MSR282

Topic

Methodological & Statistical Research, Real World Data & Information Systems

Topic Subcategory

Artificial Intelligence, Machine Learning, Predictive Analytics

Disease

No Additional Disease & Conditions/Specialized Treatment Areas

Your browser is out-of-date

ISPOR recommends that you update your browser for more security, speed and the best experience on ispor.org. Update my browser now

×