A team from Shantou University and Peking University applied five machine learning algorithms, including logistic regression and SHAP explanations, to CHARLS health data, building four-year fall risk models for middle-aged and older adults with and without pain.
Key points
Logistic regression model achieved highest AUC-ROC (0.732 for pain, 0.692 for non-pain) among five ML algorithms on CHARLS data.
SHAP analysis revealed shared predictors (fall history, height) and exclusive features like WBC, platelets, functional limitations for pain cohort versus cognitive function and environment for non-pain.
LASSO feature selection identified 24 variables in the pain model and 27 in non-pain, enabling interpretable and targeted fall risk profiling.
Why it matters:
This interpretable ML approach pinpoints unique fall risk factors, improving precision prevention and personalized care for older adults with and without pain.
Read full article
Academy
Short Physical Performance Battery (SPPB)
Overview: The Short Physical Performance Battery, or SPPB, is a validated assessment of lower-body physical function widely used in geriatric and longevity research. It measures an individual’s ability to perform tasks related to mobility and balance, helping predict fall risk and overall vitality in older adults.
Components of the SPPB:
- Gait Speed Test: Participants walk a short fixed distance (usually 8 feet) at their usual pace. The time taken is converted into a score from 0 (unable) to 4 (fastest quartile).
- Chair Stand Test: Participants rise from a chair and sit back down five times as quickly as possible without using their arms. Completion time is scored from 0 to 4.
- Balance Tests: Participants attempt three standing positions: side-by-side, semi-tandem (heel beside big toe), and tandem (heel directly in front of toe) for 10 seconds each. Each position earns 0–4 points based on successful completion.
Scoring: Each of the three tasks is scored 0–4, yielding a total SPPB score between 0 and 12. Higher scores indicate better physical performance. Researchers often categorize scores as poor (0–6), fair (7–9), and good (10–12).
Importance in Longevity Science: Lower extremity strength, balance, and mobility are critical predictors of an older adult’s ability to live independently and maintain quality of life. Declines in these areas often precede falls, hospitalization, loss of autonomy, and mortality. The SPPB serves as an early warning tool to identify individuals who may benefit from targeted interventions such as strength training, balance exercises, and environmental modifications.
Interpretable Machine Learning in Fall Risk Prediction
Role of Machine Learning: Machine learning algorithms, such as logistic regression, random forest, and XGBoost, can process high-dimensional health and lifestyle data—like that collected in CHARLS—to uncover complex patterns associated with fall risk. By training models on historical data, researchers predict which individuals are most likely to fall in the next years.
Interpretability with SHAP: While many ML models are “black boxes,” SHAP (Shapley Additive Explanations) provides a unified framework to explain individual predictions. It attributes each feature—such as pain severity, SPPB score, or white blood cell count—a quantitative impact on the predicted fall risk, making the insights actionable for clinicians.
Application: Combining SPPB scores with other biomarkers and demographic data, interpretable models help healthcare providers:
- Identify high-risk older adults early.
- Design personalized fall prevention programs.
- Monitor intervention effectiveness over time.
In longevity science, integrating functional tests like the SPPB with interpretable AI models enhances preventive care, supports healthy aging, and may delay disability onset by targeting the most vulnerable individuals.
