Comparison of three frailty measures for predicting hospitalization and mortality in the Canadian Longitudinal Study on Aging

Abstract

Abstract Background Few studies have compared different measures of frailty for predicting adverse outcomes. It remains unknown which frailty measurement approach best predicts healthcare utilization such as hospitalization and mortality. Aims This study aims to compare three approaches to measuring frailty—grip strength, frailty phenotype, and frailty index—in predicting hospitalization and mortality among middle-aged and older Canadians. Methods We analyzed baseline and the first 3-year follow-up data for 30,097 participants aged 45 to 85 years from the comprehensive cohort of the Canadian Longitudinal Study on Aging (CLSA). Using separate logistic regression models adjusted for multimorbidity, age and biological sex, we predicted participants' risks for overnight hospitalization in the past 12 months and mortality, at the first 3-year follow-up, using each of the three frailty measurements at baseline. Model discrimination was assessed using Harrell’s c-statistic and calibration assessed using calibration plots. Results The predictive performance of all three measures of frailty were roughly similar when predicting overnight hospitalization and mortality risk among CLSA participants. Model discrimination measured using c-statistics ranged from 0.67 to 0.69 for hospitalization and 0.79 to 0.80 for mortality. All measures of frailty yielded strong model calibration. Discussion and conclusion All three measures of frailty had similar predictive performance. Discrimination was modest for predicting hospitalization and superior in predicting mortality. This likely reflects the objective nature of mortality as an outcome and the challenges in reducing the complex concept of healthcare utilization to a single variable such as any overnight hospitalization.