Interpretable machine learning for identifying high-risk depression among men who have sex with men: an online cross-sectional study based on the Sexual Minority Stress Model

Men who have sex with men (MSM) experience a substantially elevated risk of depressive symptoms due to unique psychosocial stressors. However, most existing depression screening tools are developed for the general population and do not adequately capture MSM-specific stressors, resulting in limited accuracy in risk identification. In the Chinese context, the mental health burden among MSM remains underestimated, and there is a notable lack of precise assessment approaches that integrate multidimensional psychosocial factors. Between November and December 2022, this study recruited 1,369 MSM using a non-probability sampling approach. Based on the Minority Stress Model, candidate risk factors were collected, encompassing sociodemographic characteristics, MSM-related stigma, mental resilience, social support, and MSM sexual identity, with depressive symptoms as the primary outcome. 3 feature selection methods combined with ten machine learning algorithms were employed to construct predictive models, and model performance was evaluated to identify the optimal model. SHAP analysis was then applied to interpret the contributions of individual risk factors. To enhance clinical applicability, the risk probabilities of the optimal model were defined as risk scores (RS), which were subsequently used for risk stratification. Furthermore, the prediction results of the optimal model were integrated into web-based applications to support probability estimation and interactive decision-making. Among all compared algorithms, the LightGBM model demonstrated the best performance, achieving an AUC of 0.945 and exhibiting superior overall performance across multiple evaluation metrics. Key risk factors included mental resilience, identity superiority, experiences of stigma, and internalized homophobia. Risk stratification analysis revealed that individuals in the high-risk group had a significantly higher prevalence of depressive symptoms compared with those in the low-risk group ( P < 0.001). Visualization tools were employed to intuitively demonstrate the model’s classification threshold and its potential practical utility. Machine learning models can effectively identify MSM individuals at high risk for depressive symptoms by integrating multidimensional features. Risk stratification informed by model-derived probabilities and web-based applications scores could assist with earlier recognition and more tailored support strategies, which may contribute to improving the personalization of mental health screening for this population.