Speaker
Description
Introduction:
With global increases in life expectancy due to advances in healthcare and technology, the ageing population is growing rapidly, accompanied by health challenges such as frailty. This study aimed to develop a diagnostic model for identifying frailty among older Ghanaian adults using machine learning approaches.
Methods:
Data from Waves 1 and 2 of the WHO Study on Global Ageing and Adult Health (SAGE) were used to develop and validate predictive models. Five machine learning algorithms were applied following a structured pipeline involving data preprocessing, feature selection, data partitioning, and hyperparameter tuning. Model performance was assessed using calibration plots and metrics such as area under the curve (AUC), sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and Cohen’s kappa.
Results:
The analysis included 4,841 participants aged 60 years and older, with an observed frailty prevalence of 18.9%. Sixteen predictors were identified, with physical inactivity, gait speed, handgrip strength, and fatigue emerging as key features. Among the models tested, Random Forest (AUC = 0.985) and Extreme Gradient Boosting (AUC = 0.982) outperformed others. The Random Forest model showed high sensitivity (93.6%) and specificity (95.0%), while XGBoost offered superior calibration. Shapley Additive Explanation analysis underscored physical and functional status as critical predictors of frailty.
Conclusion:
Tree-based machine learning models offer accurate and interpretable tools for detecting frailty in older adults. Their implementation in primary care could enhance early risk identification, support healthy ageing initiatives, and inform health resource planning in Ghana and similar settings.
strong text