Clinical Factors and Social Determinants of Health Predict Elder Mistreatment at Two Years Among Older Adults: A...

Low rates of screening for elder mistreatment preclude early detection and harm reduction. No validated tools currently exist to identify elder mistreatment with high sensitivity. We developed, internally validated, and tested a preliminary prediction model to improve elder mistreatment screening of older adults. This retrospective observational study used 20% national Medicare Fee-for-Service claims data for the logistic regression and machine learning approaches (random forest, gradient boosted decision tree, and multilayer perceptron classifier). We included beneficiaries aged 66 and older with no elder mistreatment diagnosis from January 1, 2015 through December 31, 2016; continuously enrolled in Medicare Parts A, B, and D with no HMO coverage; and followed through 2018 (n = 2 261 166). The primary outcome was elder mistreatment diagnosis between January 1, 2017 and December 31, 2018. Predictors included demographic characteristics, comorbidities, health symptoms, and medical and social factors (ie, social determinants of health). Data analyzed were from June 23, 2022 to November 4, 2024. The sample was diverse (eg, 60.9% female, 15.2% racial and ethnic minorities, 14.6% Medicaid dual eligible, 8.9% disability prior to age 65, 72.6% pre-frail, 11.4% dementia, 64.6% hypertension). Overall, elder mistreatment diagnosis was low at 0.2%. An elder mistreatment prediction model with the best model performance was identified (c-statistic: 0.7253; 95% Confidence Interval [95% CI]: 0.712-0.738; sensitivity = 0.801, specificity = 0.467, positive predictive value = 0.002, negative predictive value = 0.999). Primary patient/caregiver support challenges (odds ratio [OR], 3.55; 95% CI: 2.36-5.34), housing and income problems (OR, 2.70; 95% CI: 1.59-4.61), learning disabilities (OR, 2.35; 95% CI: 1.15-4.78), and being tested for sexually transmitted infections (OR, 2.25; 95% CI: 1.44-3.53) were the top 4 predictors associated with increased risk of elder mistreatment diagnosis. Our preliminary elder mistreatment prediction model may guide the development of a risk assessment tool to facilitate clinician screening practices and implementation of interventions to better protect older adults from mistreatment.