Machine learning models for identifying factors associated with mild cognitive impairment in older adults in El Salvador

Abstract

Introduction. Mild cognitive impairment in older adults represents an emerging public health challenge in Latin America due to its prevalence, progression to dementia, and functional, emotional, and social repercussions. In El Salvador, local evidence is needed to guide interventions. Objective. To analyze the factors associated with mild cognitive impairment in older adults in El Salvador. Methodology. An analytical cross-sectional study was conducted using data from the 2022 National Mental Health Survey. A total of 1897 older adults with complete records were analyzed. Descriptive analyses, multivariate logistic regression as a supervised prediction model, oversampling of the minority group with mild cognitive impairment using ROSE to reduce imbalance, cross-validation, Monte Carlo simulations, and K-means clustering to characterize territorial vulnerability profiles were applied in RStudio 4.5.0. Results. The prevalence of mild cognitive impairment in older adults was 17.7 %. The probability of the event increased with age (OR 1.05), female sex (OR 1.51), symptoms of anxiety and depression (OR 1.39 and OR 1.04, respectively), and age discrimination (OR 1.79). In contrast, literacy (OR 0.26), a higher socioeconomic status (OR 0.90), living in urban areas (OR 0.75), and labor inactivity (OR 0.55) showed a protective association (p < 0.05). The model demonstrated moderate discriminatory capacity (AUC 0.75). Clustering identified a higher concentration of highly vulnerable profiles in Morazán, Cabañas, and La Unión. Conclusion. Mild cognitive impairment in older Salvadoran adults is multifactorial. These findings can guide timely screening, comprehensive care, and territorial public policies.