Deep learning model for the prediction of all-cause mortality among long term care people in China: a prospective cohort study

Abstract

Abstract This study aimed to develop a deep learning model to predict the risk stratification of all-cause death for older people with disability, providing guidance for long-term care plans. Based on the government-led long-term care insurance program in a pilot city of China from 2017 and followed up to 2021, the study included 42,353 disabled adults aged over 65, with 25,071 assigned to the training set and 17,282 to the validation set. The administrative data (including baseline characteristics, underlying medical conditions, and all-cause mortality) were collected to develop a deep learning model by least absolute shrinkage and selection operator. After a median follow-up time of 14 months, 17,565 (41.5%) deaths were recorded. Thirty predictors were identified and included in the final models for disability-related deaths. Physical disability (mobility, incontinence, feeding), adverse events (pressure ulcers and falls from bed), and cancer were related to poor prognosis. A total of 10,127, 25,140 and 7086 individuals were classified into low-, medium-, and high-risk groups, with actual risk probabilities of death of 9.5%, 45.8%, and 85.5%, respectively. This deep learning model could facilitate the prevention of risk factors and provide guidance for long-term care model planning based on risk stratification.

Keywords

• Disability
• Long-term care
• Predictors
• Risk stratification