Development and validation of a prediction model of frailty risk in community-dwelling older adults: From a national longitudinal survey.

Abstract

Objectives: This study aimed to develop and validate a risk prediction model for frailty in elderly using a nationally representative longitudinal survey database.

Study design: Longitudinal study based on public databases.

Methods: Three continuous cohorts of elderly aged 65 years or older from the Chinese Longitudinal Healthy Longevity Survey, with the 2008-2018 cohort as the development cohort. 2005-2014 and 2002-2011 cohort as validation sets. Frailty was assessed using the FI constructed from 46 indicators of health deficits, with FI ≥ 0.25 considered frailty. Prediction models were constructed using Cox regression model. We assessed the predictive performance of the models using the concordance statistic and calibration accuracy.

Results: 4,878 participants from the development cohort were enrolled with a median follow-up of 65 months. The prediction model contained 9 predictors: age, BMI, cognitive function, gender, ethnicity, education, natural teeth status, smoking status, and occupation. In the development cohort, the AUCs were 0.74, 0.78, and 0.80 at 36, 60, and 96 months. The AUCs were 0.68, 0.84, 0.85, and 0.70, 0.72, and 0.76 for two validation sets, respectively. Calibration performed well in the development and two validation sets, with a Brier score of <0.25. The prediction models constructed using machine learning algorithms showed similar predictive performance.

Conclusions: We developed and validated a model to predict the risk of incident frailty in elderly. The model provides insights to enable early screening and risk stratification for frailty in elderly, and to frame the development of individualized prevention of frailty.