Profiling Frailty: A parsimonious Frailty Index from health administrative data based on POSET theory
Abstract
Frailty assessment is crucial for stratifying populations and addressing healthcare challenges associated with ageing. This study proposes a Frailty Index based on administrative health data, with the aim of facilitating informed decision-making and resource allocation in population health management. The aim of this work is to develop a Frailty Index that 1) accurately predicts multiple adverse health outcomes, 2) comprises a parsimonious set of variables, 3) aggregates variables without predefined weights, 4) regenerates when applied to different populations, and 5) relies solely on routinely collected administrative data. Using administrative data from a local health authority in Italy, we identified two cohorts of individuals aged $\ge$65 years. A set of six adverse outcomes (death, emergency room access with highest priority, hospitalisation, disability onset, dementia onset, and femur fracture) was selected to define frailty. Variable selection was performed using logistic regression modelling and a forward approach based on partially ordered set (POSET) theory. The final Frailty Index comprised eight variables: age, disability, total number of hospitalisations, mental disorders, neurological diseases, heart failure, kidney failure, and cancer. The Frailty Index performs well or very well for all adverse outcomes (AUC range: 0.664-0.854) except hospitalisation (AUC: 0.664). The index also captured associations between frailty and chronic diseases, comorbidities, and socioeconomic deprivation. This study presents a validated, parsimonious Frailty Index based on routinely collected administrative data. The proposed approach offers a comprehensive toolkit for stratifying populations by frailty level, facilitating targeted interventions and resource allocation in population health management.