447 Background: Adolescent and young adult (AYA) cancer survivors (ages 15-39 at diagnosis) remain active in the healthcare system after completing treatment. Risk models to predict acute care events (ACEs) in this population may contribute to identification of high-risk populations and enable targeted interventions to prevent avoidable ACEs. This study aimed to develop and validate a clinical risk model to predict ACEs in AYA cancer survivors during the early post-treatment period, a vulnerable time of health care transition. Methods: Using the UNC Lineberger Cancer Information and Population Health Resource (CIPHR), which links the North Carolina Cancer Registry with insurance claims data, we identified AYAs diagnosed between 2006 and 2018 who were two to five years post-diagnosis. The primary outcome was any ACE, defined as either a hospitalization or emergency department (ED) visit. Patients were randomly assigned to training (70%) and testing (30%) cohorts. Logistic regression models were developed using stepwise inclusion of predictive variables. Model performance was evaluated using sensitivity, specificity, and C-statistics. Results: The study cohort included 7,393 patients (development = 5176, testing = 2,217). The most common cancers were breast (17%), thyroid (14%), and gynecologic (10%). Patients with ACEs (n = 3572, 48%) were more frequently female, Black, and publicly insured. Using a cut-off of the top 20% of risk scores in the test cohort to identify AYAs with an ACE, the selected model achieved a C-statistic of 0.77, positive predictive value of 0.85, specificity of 0.95 and sensitivity of 0.32. Conclusions: This is the first validated risk model to predict ACEs in AYA cancer survivors during the early post-treatment period. The model, designed for seamless EHR integration, enables early identification of high-risk patients, presenting opportunities for targeted interventions to reduce preventable acute care use. Further validation across diverse healthcare systems is planned and will further expand its clinical applicability.
Buckley et al. (Wed,) studied this question.