JMIR Publications

ABSTRACT

Background:

Frailty is extremely common in older people and is often considered a sign of "old age" and therefore overlooked; Timely identification - screening out the "frail" state of the elderly, conducting comprehensive assessment of the elderly, and giving effective intervention, which has positive practical significance for improving the quality of life of the elderly and prolonging the quality of life. At present, there is no clear definition benchmark for elderly frailty, and the industry is generally based on some frailty assessment scales such as the Fried scale to judge the frailty of the elderly.

Objective:

This study aims to investigate the current situation of frailty of the elderly population of Mongolian and Han ethnic groups aged 55 and above in Inner Mongolia, select features through machine learning algorithms, use ensemble learning algorithms, establish models and use corresponding evaluation indicators to evaluate the accuracy of the model, and analyze the risk factors, so as to provide more convenient help for the prevention of frailty of the elderly in this region.

Methods:

A total of 1186 cases of debilitating cases over 55 years old were randomly selected from the data set of people over 55 years old in Inner Mongolia Autonomous Region as the survey subjects, and 82 characteristic indicators were collected to model them as feature matrices, and the 1186 investigators were selected as the original data for data preprocessing, and the ensemble learning algorithm was used to establish a model, and the selected ensemble algorithms were: , RF (random forest), XGBoost (limit gradient boosting tree), GBDT (gradient boosting tree), Adaboost, In this study, three different datasets were used for modeling evaluation, namely the original dataset, which contained the sample information of 1186 respondents, the male investigator dataset, the male dataset with the gender in the original dataset, and the female investigator dataset, the dataset with the gender in the original dataset. Through the comprehensive evaluation and analysis of different models, 18 features with high correlation with frailty were screened out from the case dataset.

Results:

By modeling 18 risk factors screened out by the machine learning algorithm and evaluating the accuracy of the model, we found that the proportion of female investigators was more frail than that of male investigators, and the model performance of the male investigator dataset was better than that of the female investigator dataset, among which the accuracy of random forest and XGBoost was close to 100%, and the model accuracy was very high.

Conclusions:

According to the comprehensive evaluation results of multiple models, it is shown that gender, age, MMSE score, presence or absence of domestic insurance, hypertension, coronary heart disease, sleep, grip strength, current exercise, disability, cognitive level, number of diseases, BMI, surgical history, income, occupation, and stroke will have obvious effects on frailty, all of which are influencing factors of frailty in the elderly in Inner Mongolia (p<0.05).