Automated frailty scores: towards clinical implementation

Abstract

Frailty is a state of physiological decline across multiple systems. It is common among older adults and is a strong predictor of mortality and other adverse outcomes independent of age. Given its public health burden especially during the COVID-19 pandemic, increasing efforts have been made in recent years to develop frailty assessment tools that could allow early identification and management of frail individuals. Although there is still no consensus on how we can best measure frailty, the Clinical Frailty Scale (CFS) is one of the most frequently adopted frailty measures in clinical settings. It is a simple, rapid, and accurate assessment tool based on clinical evaluation on several domains such as diseases, functioning, and cognition [1]. However, the need of in-person evaluation makes the CFS possibly prone to interrater bias and not always a priority in settings that need to add resources for bedside assessment. Alternatively, automated frailty scores based on readily available electronic health records (EHRs) or administrative claims data are increasingly used as frailty screening tools. Examples include the Hospital Frailty Risk Score (HFRS) calculated based on 109 International Classification of Diseases, Tenth Revision (ICD-10) codes [2], and electronic frailty indices (eFIs) constructed based on the widely validated deficit accumulation model [3, 4]. These scores are generally proven to be valid prognostic tools for predicting mortality, yet they are usually limited to country-specific settings (e.g., the eFI by Clegg et al. is calculated based on the Read codes used in the UK primary care [3], which may not be applicable to other health systems). Testing whether and how automated frailty scores can be applied in other populations and health systems is therefore essential before they can be widely implemented.