Developing a predictive model to evaluate the effectiveness of population health management to reduce healthcare costs: Evidence from a Saudi health insurance database
Naof F.S. Al-Ansary, Faisal Albagmi, Mohamed K AbdelhaleemAbstract:
BACKGROUND:
Population health management (PHM) programs aim to identify members who may benefit from proactive care, but inefficient targeting can limit clinical and financial value. This study developed a hybrid fuzzy logic and machine learning framework for PHM prioritization and examined healthcare cost trends in enrolled and comparison members.
MATERIALS AND METHODS:
A retrospective observational analysis used 411,142 claim-period records of 57,979 insured members observed, including 29,480 PHM members and 28,499 controls from April 2023 to April 2025. Fuzzy scores for Charlson Comorbidity Index, age, and burning cost per member per month were combined with eXtreme Gradient Boosting models. Prediction was defined at the enrollment index date, with postenrollment variables excluded from eligible predictors. Cost effects were assessed using group change ratios and a matched-pair difference-in-differences sensitivity analysis.
RESULTS:
Mean costs fell from Saudi Arabian Riyal (SAR) 9,699.4 to SAR 9,541.4 for PHM members (−1.6%) and from SAR 9,119.1 to SAR 9,087.5 for controls (−0.3%). The impact ratio was 0.987, corresponding to 1.3% additional relative improvement. Of the 18,517 complete matched pairs with observations in both periods, the paired difference-in-differences was −SAR 457.3 per member-month (95% confidence interval, −652.2 to −262.5;
CONCLUSION:
PHM enrollment was associated with a modestly more favorable cost change than matched controls; however, observational design does not establish causality. The predictive framework remains at the development stage because the supplied workbook does not contain complete discrimination, calibration, score sensitivity, or independent validation outputs.