Development of a Simple Clinical Tool for Predicting Early Dropout in Cardiac Rehabilitation
Quinn R. Pack, Paul Visintainer, Michel Eid Farah, Grace LaValley, Heidi Szalai, Peter K. Lindenauer, Tara Lagu
Abstract
BACKGROUND: Nonadherence to cardiac rehabilitation (CR) is common despite the benefits of completing a full program. Adherence might be improved if patients at risk of early dropout were identified and received an intervention. METHODS: Using records from patients who completed ≥1 CR session in 2016 (derivation cohort), we employed multivariable logistic regression to identify independent patient-level characteristics associated with attending <12 sessions of CR in a predictive model. We then evaluated model discrimination and validity among patients who enrolled in 2017 (validation cohort). RESULTS: Of the 657 patients in our derivation cohort, 318 (48%) completed <12 sessions. Independent risk factors for not attending ≥12 sessions were age <55 yr (OR = 0.23, P < .001), age 55 to 64 yr (OR = 0.35, P < .001), age ≥75 yr (OR = 0.64, P = .06), smoker within 30 d of CR enrollment (OR = 0.40, P = .001), low risk for exercise adverse events (OR = 0.54, P = .03), and nonsurgical referral diagnosis (OR = 0.66, P = .02). Our model predicted nonadherence risk from 23-90%, had acceptable discrimination and calibration (C-statistics = 0.70, Harrell's E50 and E90 2.0 and 3.6, respectively) but had fair validity among 542 patients in the validation cohort (C-statistic = 0.62, Harrell's E50 and E90 2.1 and 11.3, respectively). CONCLUSION: We developed and evaluated a single-center simple risk model to predict nonadherence to CR. Although the model has limitations, this tool may help clinicians identify patients at risk of early dropout and guide intervention efforts to improve adherence so that the full benefits of CR can be realized for all patients.