Multi-View Deep Learning Framework for Predicting Patient Expenditure in Healthcare
Xianlong Zeng, Simon Lin, Chang Liu
Abstract
Accurately predicting patient expenditure in healthcare is an important task with many applications such as provider profiling, accountable care management, and capitated medical payment adjustment. Existing approaches mainly rely on manually designed features and linear regression-based models, which require massive medical domain knowledge and show limited predictive performance. This paper proposes a multi-view deep learning framework to predict future healthcare expenditure at the individual level based on historical claims data. Our multi-view approach can effectively model the heterogeneous information, including patient demographic features, medical codes, drug usages, and facility utilization. We conducted expenditure forecasting tasks on a real-world pediatric dataset that contains more than 450,000 patients. The empirical results show that our proposed method outperforms all baselines for predicting medical expenditure. These findings help toward better preventive care and accountable care in the healthcare domain.