Machine-learning models flag depression risk in older Chinese adults with chronic disease by cognitive status
In survey data from 5,798 older Chinese adults with chronic illness, three machine-learning models predicted depression risk with accuracy of 0.755 to 0.767; the top predictor differed by cognitive status. The abstract reports no external validation yet. [1]
Automatic summary · not editorially reviewed
The key findings
The study used data from the Chinese Longitudinal Healthy Longevity Survey with 5,798 participants. [1]
Accuracy of the three models ranged from 0.755 to 0.767, and all Brier scores were 0.160 or lower. [1]
'Feeling energetic' was the top predictor in the unimpaired and mildly impaired groups; 'sleep duration per day' in the severely impaired group. [1]
The source at a glance
- Study type / format
- Machine-learning modelling study using survey data (abstract)
- Population / context
- Older adults with chronic diseases in the Chinese Longitudinal Healthy Longevity Survey, grouped by level of cognitive impairment.
- Summary based on
- Study abstract
- Original publication
- September 11, 2026
Survey of 5,798 older adults with chronic illness
The study drew on data from the Chinese Longitudinal Healthy Longevity Survey, with 5,798 participants. It aimed to build machine-learning models that identify depression risk among older adults with chronic diseases at different levels of cognitive impairment. The abstract notes that depression is more common in this group than in the general population and that cognitive impairment together with chronic disease strongly affects patients' lives.
Supporting source [1]
Accuracy 0.755–0.767; top predictors differ by cognition
Models were trained with the XGBoost algorithm, and their interpretability was supported with SHapley Additive exPlanations; each model was then simplified based on feature importance. Accuracy of the three models ranged from 0.755 to 0.767, and all Brier scores were 0.160 or lower, which the authors describe as good predictive performance. 'Feeling energetic' was the most important predictor in the cognitively unimpaired and mildly impaired groups, while 'sleep duration per day' was the top predictor in the severely impaired group.
Supporting source [1]
Promising but not yet externally validated
The authors conclude that the proposed models showed promising performance in identifying depression risk across cognitive levels and suggest that key predictors may differ by cognitive status, supporting cognition-stratified screening strategies. They state that further external validation is needed before use in routine clinical practice. The abstract does not report how depression was measured, the follow-up period or the split between training and test data.
Supporting source [1]
How to interpret this
Abstract only; the full text was not reviewed. The abstract does not report how depression was measured, the follow-up period or the training/test split, and the authors state that external validation is still needed.
Based on the study abstract; the full paper, tables and supplementary data were not reviewed.
Source & supporting passages
Qiu Y, Li R. · europepmc.org · September 11, 2026