Prediction model related to 6-year risk of frailty in older adults aged 65 years or above in China
10.3760/cma.j.cn112338-20231205-00333
- VernacularTitle:中国65岁及以上老年人6年虚弱发生风险预测模型研究
- Author:
Jinhui ZHOU
1
;
Li QI
;
Jun WANG
;
Sixin LIU
;
Wenhui SHI
;
Lihong YE
;
Zhenwei ZHANG
;
Zenghang ZHANG
;
Xi MENG
;
Jia CUI
;
Chen CHEN
;
Yuebin LYU
;
Xiaoming SHI
Author Information
1. 国家癌症中心/国家肿瘤临床医学研究中心/中国医学科学院北京协和医学院肿瘤医院，北京 100021
- Keywords:
Frailty;
Older adults;
Key factors;
Prediction model
- From:
Chinese Journal of Epidemiology
2024;45(6):809-816
- CountryChina
- Language:Chinese
-
Abstract:
Objective:To develop a prediction tool for 6-year incident risk of frailty among Chinese older adults aged 65 years or above.Methods:Data from the Chinese Longitudinal Healthy Longevity Survey from 2002 to 2018 was used, including 13 676 older adults aged 65 years or above who were free of frailty at baseline. Key predictors of frailty were identified via the least absolute shrinkage and selection operator (LASSO) method, and were thereafter used to predict the incident frailty based on the Cox proportional hazards regression model. The model was internally validated by 2 000 Bootstrap resamples and evaluated for the performance of discrimination and calibration using the area under the receiver operating characteristic curve (AUC) and calibration curve, respectively. The net benefit of the developed prediction tool was evaluated by decision-curve analysis.Results:The M( Q1, Q3) age and follow-up time of the participants were 81.0 (71.0, 90.0) years and 6.0 (4.1, 9.2) years, respectively. A total of 4 126 older persons (30.2%) were recorded with frailty incidents during the follow-up, with the corresponding incidence density of 41.8/1 000 person-years. A total of 15 key predictors of frailty were selected by LASSO, namely, age, sex, race, education years, meat consumption, tea drinking, performing housework, raising domestic animals, playing cards or mahjong, and baseline status of visual function, activities of the daily living score, instrumental activities of the daily living score, hypertension, heart disease, and self-rated health. The prediction model was internally validated with an AUC of 0.802, with the max Youden's index of 0.467 at a risk threshold of 19.0%. The calibration curve showed high consistency between predicted probabilities and observed proportions of frailty events. The decision curve indicated that higher net benefits could be obtained via the prediction model than did strategies based on intervention in all or none participants for any risk threshold less than 59%, and the model-based net benefit was estimated to be 0.10 at a risk threshold of 19.0%. Conclusions:The herein developed 6-year incident risk prediction model of frailty, based on easily accessible questionnaires and physical examination variables, has good predictive performance. It has application potential in identifying populations at high risk of incident frailty.
