Does identifying frailty from ICD-10 coded data on hospital admission improve prediction of adverse outcomes in older surgical patients? A population-based study.

BACKGROUND: frailty is a major contributor to poor health outcomes in older people, separate from age, sex and comorbidities. This population-based validation study evaluated the performance of the International Classification of Diseases, 10th revision, coded Hospital Frailty Risk Score (HFRS) in the prediction of adverse outcomes in an older surgical population and compared its performance against the commonly used Charlson Comorbidity Index (CCI). METHODS: hospitalisation and death data for all individuals aged ≥50 admitted for surgery to New South Wales hospitals (2013-17) were linked. HFRS and CCI scores were calculated using both 2- and 5-year lookback periods. To determine the influence of individual explanatory variables, several logistic regression models were fitted for each outcome of interest (30-day mortality, prolonged length of stay (LOS) and 28-day readmission). Area under the receiving operator curve (AUC) and Akaike information criterion (AIC) were assessed. RESULTS: of the 487,197 patients, 6.8% were classified as high HFRS, and 18.3% as high CCI. Although all models performed better than base model (age and sex) for prediction of 30-day mortality, there was little difference between CCI and HFRS in model discrimination (AUC 0.76 versus 0.75), although CCI provided better model fit (AIC 79,020 versus 79,910). All models had poor ability to predict prolonged LOS (AUC range 0.62-0.63) or readmission (AUC range 0.62-0.65). Using a 5-year lookback period did not improve model discrimination over the 2-year period. CONCLUSIONS: adjusting for HFRS did not improve prediction of 30-mortality over that achieved by the CCI. Neither HFRS nor CCI were useful for predicting prolonged LOS or 28-day unplanned readmission.