Application of machine learning techniques in screening for depression in older adults in Primary Health Care

Depression in older adults is a significant public health issue, often underdiagnosed in Primary Health Care (PHC), particularly due to the atypical presentation of symptoms and structural limitations of services. Machine learning techniques have emerged as innovative tools for the early screening of mental health conditions such as depression, given their ability to integrate multiple predictive factors and generate interpretable analyses. This study aimed to apply a decision tree model to detect suspected depression in older adults assisted in PHC, evaluating its diagnostic accuracy and identifying the main associated factors. This was an observational study based on secondary data, analyzing sociodemographic and behavioral variables. The results showed that marital status, advanced age, lack of social participation, and race/color were the main predictors. The model demonstrated high sensitivity but low specificity, indicating potential for screening purposes when complemented by clinical evaluation. It is concluded that the responsible use of machine learning can strengthen depression screening in later life, supporting early and qualified interventions in primary care settings.