Development and validation of a prediction model for falls among older people using community-based data.

Abstract

This is the first study to employ multilevel modeling analysis to develop a predictive tool for falls in individuals who have participated in community group exercise over a year. The tool may benefit healthcare workers in screening community-dwelling older adults with various levels of risks for falls.

Purpose: The aim of this study was to develop a calculation tool to predict the risk of falls 1 year in the future and to find the cutoff value for detecting a high risk based on a database of individuals who participated in a community-based group exercise.

Methods: We retrospectively reviewed a total of 7726 physical test and Kihon Checklist data from 2381 participants who participated in community-based physical exercise groups. We performed multilevel logistic regression analysis to estimate the odds ratio of falls for each risk factor and used the variance inflation factor to assess collinearity. We determined a cutoff value that effectively distinguishes individuals who are likely to fall within a year based on both sensitivity and specificity.

Results: The final model included variables such as age, sex, weight, balance, standing up from a chair without any aid, history of a fall in the previous year, choking, cognitive status, subjective health, and long-term participation. The sensitivity, specificity, and best cutoff value of our tool were 68.4%, 53.8%, and 22%, respectively.

Conclusion: Using our tool, an individual's risk of falls over the course of a year could be predicted with acceptable sensitivity and specificity. We recommend a cutoff value of 22% for use in identifying high-risk populations. The tool may benefit healthcare workers in screening community-dwelling older adults with various levels of risk for falls and support physicians in planning preventative and follow-up care.