Lauren Parlett, Qinli Ma, Abiy Agiro, and Kori Magallanez
Tags Medicare | Healthcare Quality | Prediction | Machine Learning | Factor Analysis
Research Objective: The Centers for Medicare & Medicaid Services (CMS) administers the Medicare Health Outcomes Survey (HOS) to members in order to calculate star ratings. Although HOS scores account for 10% of star ratings, HOS scores are the most fluctuating components of MA plan star ratings. The star ratings are important to value-based purchasing agreements, accreditation, and Medicare Advantage (MA) plan reputation. We sought to identify plausibly modifiable factors that influenced health scores (mental and physical), falling, bladder control, and physical activity. Study Design: Members of MA plans completed a baseline HOS in 2012 and 2013 and follow-up surveys in 2014 and 2015, respectively. The member responses were linked with multi-dimensional data (claims, benefit design, prior authorizations, program/rewards participation). We employed machine-learning techniques, factor analysis, and regressions to select variables and to identify factors associated with improved HOS scores and to assess the relationships between member characteristics and HOS measures. Population Studied: MA members that completed a baseline survey in the years included in the study were included in cross-sectional analyses (N=8,063). Longitudinal analyses of physical and mental health scores required completion of the follow-up survey (N=5,592). Principal Findings: On average, members were 75 years old and female (60%). Between baseline and follow-up, 48% and 53% of members maintained or increased their physical or mental health scores, respectively. At baseline, among those who had fallen, 34% had spoken with their doctor about preventing falls; among those who had urine leakage, 34% received treatment; and among those who talked to their doctor about physical activity, 75% were advised to increase or maintain their physical activity. Factors associated with improved HOS scores included “not talking with doctor about physical activity, bladder control, or risk for fall”, “difficulties with daily living” and management of depression, pain, and high blood pressure. Mental and physical health were correlated. Conclusions: HOS measures are affected by member characteristics that fall into three domains: healthcare utilization, lifestyle difficulties, and medical conditions. There are shared predictors among the measures, and physical and mental health affect each other. Implications for Policy or Practice: Targeting improvement for one HOS measure could result in positive results for other measures, especially if the intervention involves communication with a provider or addresses the difficulty with daily tasks and mobility.