Predicting Annual Length-Of-Stay and its Impact on Health
Proceedings of The First Workshop Medical Informatics and Healthcare held with the 23rd SIGKDD Conference on Knowledge Discovery and Data Mining, PMLR 69:27-34, 2017.
Avoidable hospitalizations are a source of increased health expenditures in many health systems. Prolonged length of stay is costly for providers, insurers, and patients to the extent it is associated to higher health service consumption and to the development of endangering states during the hospital stay. In this article we use machine learning techniques to predict annual patient length-of-stay in Colombia’s statutory health care system and measure its impact on health costs by estimating the potential cost savings of a hospitalization prevention program. Results from the predictive modeling show tree-based methods outperform linear approximations and achieve lower out-of-sample error rates compared to the winning model of the Heritage Health Prize. We also show that a prevention program where patient intervention is decided upon the predictions of the model can achieve significant cost savings relative to the best uniform policy (i.e, intervene all patients or no intervention). This holds for program efficacies greater than 40% and intervention costs per patient ranging between 100,000 and 700,000 Colombian pesos.