Development and Validation of a Nutritional Frailty Phenotype for Older Adults Based on Risk Prediction Model: Results From a Population-Based Prospective Cohort Study

J Am Med Dir Assoc. 2024 Dec 19:105425. doi: 10.1016/j.jamda.2024.105425. Online ahead of print.

Abstract

Objectives: Malnutrition is generally studied to be involved in outlining hazard frailty trajectories, resulting in adverse outcomes. In view of frailty's multidimensional nature, we aimed to assess the contribution of nutritional items in existing frailty tools to adverse outcomes, and develop and validate a nutritional frailty phenotype based on machine learning.

Design: A population-based prospective cohort study.

Setting and participants: A total of 7641 older adults from the China Health and Retirement Longitudinal Study (CHARLS) were included as the training set to develop the nutritional frailty phenotype between 2011 at baseline and 2013 at follow-up, and 8656 older adults between 2013 at baseline and 2015 at follow-up were included for temporally external validation.

Methods: The important predictors for 2-year incident adverse outcomes including all-cause mortality, disability, and combined outcomes were selected based on the least absolute shrinkage and selection operator. The nutritional frailty phenotype was developed using 2 machine learning models (random forest and eXtreme Gradient Boosting), and modified Poisson regression with the robust (sandwich) estimation of variance.

Results: Slowness (walking speed), lower extremity function (chair-stand test), and upper limb function (grip strength) were selected as important predictors for each outcome using least absolute shrinkage and selection operator. For the training set, the models for predicting all-cause mortality (area under the receiver operating characteristics curves [AUCs], 0.746-0.752; mean AUCs of the 5-fold cross validation: 0.746-0.752) and combined outcome (AUCs, 0.706-0.708; mean AUCs of the 5-fold cross validation, 0.706) showed acceptable discrimination, whereas the models for predicting incident disability had approximately acceptable discrimination (AUCs, 0.681-0.683; mean AUCs of the 5-fold cross validation, 0.681-0.684). For external validation, all models had acceptable discrimination, overall prediction performance, and clinical usefulness, but only the modified Poisson regression model for predicting incident disability had acceptable calibration.

Conclusions and implications: A novel nutritional frailty phenotype may have direct implications for decreasing risk of adverse outcomes in older adults. Weakness and slowness play a major role in the progression of nutritional frailty, emphasizing that nutritional supplementation combined with exercise may be one of the feasible pathways to prevent or delay adverse outcomes.

Keywords: Nutritional frailty; ageing; assessment; frailty; nutrition; older adults; subtypes.