APPLYING DATA SCIENCE AND MACHINE LEARNING TO UNDERSTAND HEALTH CARE TRANSITION FOR ADOLESCENTS AND EMERGING ADULTS WITH SPECIAL HEALTH CARE NEEDS

A problem of classification places adolescents and emerging adults with special health care needs among the most at risk for poor or life-threatening health outcomes. This preliminary proof-of-concept study was conducted to determine if phenotypes of health care transition (HCT) for this vulnerable population could be established. Such phenotypes could support development of future studies that require data classifications as input. Mining of electronic health record data and cluster analysis were implemented to identify phenotypes. Subsequently, a machine learning concept model was developed for predicting acute care and medical condition severity. Three clusters were identified and described (Cluster 1, n = 36,075; Cluster 2, n = 74,438; Cluster 3, n = 6,742). Baseline metrics indicated that the machine learning models performed better than naïve forecasting (MASE [Model 1, 0.5], [Model 2, 0.7]). An analysis of state and federal policies using a dynamics framework was then conducted to identify policy interdependencies, including policy actors, institutional structures, and intergovernmental agreements relevant to HCT. Policy components identified were in the categories of insurance/payer implications and innovations, HCT surveillance, HCT services and workforce, and digital innovations and interoperability. These included waiver authorities under the Social Security Act; policy institutions and actors associated with HCT surveillance and workforce policy, including the US Health Resources and Services Administration (HRSA) and the Centers for Medicare and Medicaid Services (CMS); and policy institutions and actors associated with data innovation and interoperability, including CMS and its Interoperability and Patient Access Rule (2020). Insights from the policy analysis and the HCT phenotype study, including cluster demographics data, were applied in development of recommendations for policy collaboratives and policy research. These include the a