Childhood trauma and recent stressors in predicting subclinical psychotic symptoms among Chinese university students in southwest China: a machine learning analysis within a gender-specific framework

Abstract

Background: subclinical psychotic symptoms (SPS) are common among college students and can lead to future mental health issues. However, it is still not clear which specific childhood trauma, stressors and health factors lead to SPS, partly due to confounding factors and multicollinearity.

Objective: to use machine learning to find the main predictors of SPS among university students, with special attention to gender differences.

Methods: a total of 21,208 university students were surveyed regarding SPS and a wide range of stress-related factors, including academic pressure, interpersonal difficulties, and abuse. Nine machine learning models were used to predict SPS. We examined the relationship between SPS and individual stressors using chi-square tests, multicollinearity analysis, and Pearson heatmaps. Feature engineering, t-SNE, and SHAP values helped identify the most important predictors. We also assessed calibration with calibration curves and Brier scores, and evaluated clinical usefulness with decision curve analysis (DCA) to provide a thorough assessment of the models. In addition, we validated this model using independent external data.

Findings: the XGBoost model had the best prediction results, with an AUC of 0.89, and validated with external data. It also showed good calibration, and decision curve analysis indicated clear clinical benefit. Interpersonal difficulties, academic pressure, and emotional abuse emerged as the strongest predictors of SPS. Gender-stratified analyses revealed that academic pressure and emotional abuse affected males more, while health issues like chest pain and menstrual pain were stronger predictors for females.

Conclusions: machine learning models effectively identified key stressors associated with SPS in university students. These findings highlight the importance of gender-sensitive approaches for the early detection and prevention of psychotic symptoms.

Clinical implications: SPS in college students can be predicted by interpersonal difficulties, academic stress, and childhood emotional abuse. This information can help mental health professionals develop better ways to prevent and address SPS.

What is already known on this topic: the close associations between childhood trauma, stressful life events, health factors, and SPS have already been established. However, it remains unclear which specific types of trauma and stressful events have the greatest impact. Comprehensive predictive models based on these factors have not yet been developed, especially for undergraduate populations.

What this study adds: by applying nine machine learning models, this study identified the optimal predictive model and found that interpersonal difficulties, academic stress, and childhood emotional abuse are the three most influential factors for SPS. Additionally, gender differences were observed: males face higher risk mainly due to greater academic stress, while pain symptoms have a stronger impact on females. The model's performance was also validated using external data.

How this study might affect research, practice, or policy: this study provides a foundation for further optimization and refinement of predictive models for assessing SPS risk in college students, as well as the development of targeted intervention strategies. It also offers clinical guidance for prevention and early intervention.

More information

Identifiers

ORCID for Miguel Garcia-Argibay: orcid.org/0000-0002-4811-2330

ORCID for Anoop Kadan: orcid.org/0000-0002-4335-5544

ORCID for Samuele Cortese: orcid.org/0000-0001-5877-8075

Catalogue record

Export record

Altmetrics