An Original Framework for Evaluating Data Science Model Equity in Nigerian Health Insurance Systems

Machine learning models deployed in health insurance systems including risk stratification, fraud detection, and claim prioritisation carry equity implications that are particularly consequential in a context where informal economy workers and rural populations are already excluded from formal health coverage. This study developed an original framework for evaluating model equity in Nigerian health insurance systems. A theory-building methodology was employed, drawing on systematic review of health insurance AI equity literature (59 publications from 2019 to 2024), empirical equity analysis of three prediction models used in NHIA claims processing using disaggregated performance metrics, and expert validation. The systematic review confirmed that health insurance AI equity frameworks were developed for US or European contexts and assumed insurance coverage populations unrepresentative of Nigerian demographics. Empirical equity analysis showed that fraud detection models had significantly higher false positive rates for rural claimants (14.8 percent) than urban claimants (6.3 percent), creating inappropriate claim delays for rural facilities. Risk stratification models underestimated costs for patients with traditional medicine co-utilisation, creating under-provisioning for 23.4 percent of enrolled members. The original Nigerian Health Insurance Model Equity Framework (NHIMEF) proposes five equity dimensions: access equity, accuracy equity across demographic groups, allocation equity, procedural equity, and redress equity. Each dimension includes measurement metrics, minimum performance thresholds, and remediation procedures. Expert review by 18 health equity and data science specialists confirmed the framework's original contribution to responsible health AI in low-income country contexts.

Keywords: model equity, health insurance, Nigeria, algorithmic fairness, NHIA