Using Extreme Gradient Boosting in Claims Data to Predict Future Costs in a Health First Colorado Population

This white paper presents a machine learning approach utilizing extreme gradient boosting (XGBoost) to predict future healthcare costs within the Health First Colorado Medicaid population. By analyzing approximately 3,700 claims-based predictors and demographic characteristics, the study achieved an R² value of 0.786, indicating strong predictive capability. Significant findings suggest that prior costs are the most crucial predictors of future healthcare expenses, outweighing other claims-based factors. The research underscores the potential of machine learning in improving the management of high-utilizing Medicaid members while using minimal computational resources.