Predicting long-term hospital outcomes in older adults with the hospital frailty risk score: a nationwide study of repeated readmissions

The Hospital Frailty Risk Score (HFRS) is a widely used frailty detection score. High values of the score are associated with mortality and prolonged length of stay in geriatric patients. Studies in the field of geriatrics often attempt to predict a single 30-day hospital readmission, whereas healthcare trajectories encompassing several hospitalizations over a longer period may be more specific for the ascertainment of frailty, as they indicate a prolonged decrease in physiological reserves. The study of outcomes beyond the traditional 30-day readmission may lead to better decisions in the management of patient trajectories in geriatric care. Therefore, this study aimed to evaluate variables (including the HFRS) associated with a composite outcome consisting of at least two successive readmissions or in-hospital death within 6 months. Data were collected and analyzed within the French national hospitalization database from 1 January 2018 to 30 June 2023. A multivariable analysis by logistic regression modelled the probability of a composite outcome comprising at least two readmissions or in-hospital death within 6 months. Predictive performance was validated on a large independent nationwide cohort. Model performance was evaluated using the Area Under the Curve (AUC), sensitivity, and specificity. The development cohort consisted of 2,701,224 patients aged 75 and older (mean age 83.7, standard deviation 6.11). Variables associated with the composite outcome included the number of stays in the previous year, the socioeconomic status of the patient’s zone of residence, and access to primary care in the zone of residence. The model had adequate performance in the validation cohort (674,648 hospital stays), with an Area Under the Curve (AUC) of 0.718. The model was able to predict the composite outcome with 69.6% sensitivity (95% Confidence Interval 69.4 to 69.8) and 62.2% specificity (95% CI 62.1 to 62.3). Predicting healthcare trajectories after hospitalization appears to be more relevant to the management of frail patients than attempting to predict a single 30-day readmission. The model’s performance was close to the performance of widely used screening tests. In a screening setting, positive cases would require additional tests to verify if they are at risk of adverse outcomes.