ABSTRACT
Objectives: Care coordination is a key component of the population health management. However, the mechanism for identifying patients who may benefit the most from this model of care is unclear. The objective of study is to evaluate the performance of a risk-stratification instrument using a model of AI - Rule-based expert system (RBES) - in predicting healthcare utilization and costs. Method(s): Retrospective cohort study from beneficiaries of a health plan using administrative databases (prior authorizations claims systems): 27,539 individuals were assigned a predicted illness burden score using a case-mix adjustment system from diagnoses and health utilization data (2019 to 2021). Population was stratified according to the score into three main groups: G1) case management;G2) health support;G3) health promotion. Analysis was also performed in subgroups: prolonged hospitalization, readmission, complex medical conditions (CC), continued therapy (CT) (G1);chronic unstable (CU), post-COVID 19, high cost, high user (G2);healthy elderly, risk factor, low risk (G3). Data Science team analyzed population using algorithms which uses a set of logical rules derivatives of human specialists. Result(s): According to score 1,053 individuals stratified in G1, average age 68 years, annual cost U$11,318, 10 times more than average;G2, n=5,429;67 years;U$2,863;G3, n=21,037;53 years;U$246. The sickest population: 3.8%, 19.7% and 76.5% uses about 37%, 48% and 15% of healthcare expenses respectively. Most representative subgroups: CC, CT, and CU with average annual cost five or more times than average. Conclusion(s): Dashboard developed using RBES tools can supports healthcare management. Stratifying risk helps to address specific health care challenges, to align levels of care, to implement a value-based care approach. Also demonstrates to be the most logical and practical initial step to create a data set with labeled variables to start a machine learning using supervised training - the next phase in this project. Copyright © 2022