Background

Memory difficulty is common in later life and is clinically relevant because it may signal neurocognitive change, psychiatric distress, or functional vulnerability.

Aims

To develop and interpret machine-learning models that classify memory difficulty among adults aged ≥65 years using a nationally representative survey dataset, and to identify the dominant predictors of classification.

Method

We conducted a cross-sectional analysis of the 2023 National Survey on Drug Use and Health (NSDUH) public use file (U.S. civilian non-institutionalised population). The outcome was any memory difficulty derived from the NSDUH item assessing difficulty remembering or concentrating (dichotomised as none versus any). Candidate predictors included demographic characteristics, self-rated health, functional limitations (hearing, vision, mobility, communication, self-care), psychological distress (Kessler-6), major depressive episode with severe role impairment, substance-use indicators, and cardiometabolic conditions. We compared logistic regression, decision tree, random forest, and gradient-boosted trees with cross-validation and survey-weighted training/evaluation. Test-set performance was assessed using weighted area under the curve (AUC) and average precision with bootstrap 95% confidence intervals (CIs). Explainability used grouped SHAP values and permutation importance.

Results

The analytic sample included 4,941 respondents aged ≥65 years. The weighted prevalence of any memory difficulty was 39.0%. The best-performing model was a random forest, with test weighted AUC 0.796 (95% CI 0.749–0.836) and weighted average precision 0.713 (95% CI 0.631–0.781). Explainability analyses consistently identified psychological distress (Kessler-6) as the dominant predictor, followed by functional limitations, particularly hearing and mobility difficulty.

Conclusions

Interpretable machine learning classified memory difficulty with good discrimination in a nationally representative older adult sample. Psychological distress and functional limitations dominated predictions, underscoring clinically actionable domains relevant to assessment pathways for cognitive symptoms in later life.

Keywords

Memory difficulty; Older adults; Psychological distress; Functional limitations; Machine learning; Population-based study