Population health informatics, a path to health equity for Medicaid patients enrolled in the patient centered medical home (PCMH)

Background: The Medicaid population is burdened with social and environmental factors that impede care and result in poor health outcomes.1 Currently, the collection of social determinants of health (SDH) in electronic databases in the United States is fragmented and inadequate resulting in health gaps leading to substandard care. The integration of SDH with medical information will help achieve health equity for the Medicaid population and determine how SDH affect adherence to physician recommendations.2 Methods: Supervised and unsupervised ML algorithms were used to mine data and develop prediction models to determine if social determinants of health were risk factors for appointment compliance for Medicaid patients enrolled in the Patient-Centered Medical Home model from July 1, 2017 – June 30, 2019. This retrospective study analyzed records for 911 patients, ages 18-65 years and 12,118 encounters. Results: k-nearest neighbors had the best performance (AUC=0.743). Support vector machines (AUC = 0.656), logistic regression (0.644) and random forest had the lowest performance (0.599). Tobacco-related disorders, nutritional anemia, age, and gender predicted preventative health appointments. Conclusion: SMOTE oversampling technique can be used to balance a minority class to improve risk predictions significantly compromising performance scores (AUC, precision, recall, F1 score).Ph.D.Includes bibliographical reference