Characterising symptom clusters: examining profiles of somatic symptoms and their psychosocial predictors among chinese youths using longitudinal data with a machine learning approach

Objective: Somatic symptoms (SS) pose public health burdens, but it remains unclear whether SS should be defined by overall severity or specific symptoms. Therefore, we aimed to identify SS profiles and explore how psychosocial predictors shape these profiles based on the biopsychosocial model.
Methods and measures: From October to December 2020, Chinese college students (n = 598) completed three waves of survey on SS and psychosocial predictors. We used cluster analysis to identify SS profiles, classification and regression tree (CART) to examine psychosocial predictors, and the area under the receiver operator curve (AUC) to assess CART’s predictive performance.
Results: Three profiles were identified: high severity and featured gastrointestinal symptoms (high, 23.6%), moderate severity and featured tiredness (moderate, 32.9%), and low in all symptoms (low, 43.5%). These profiles differed in rumination, anxiety symptoms, perceived stress, and peer attachment. Specifically, more rumination and higher anxiety symptoms predicted the membership of the high group. The CART models effectively distinguished the high group (AUC ≥ 0.84) and moderate group (AUC ≥ 0.70) from the low group.
Conclusion: The three symptom patterns enhance our understanding of similarities and differences in SS profiles. Key predictors will inform targeted prevention and intervention for those at higher risk.
© 2026 Informa UK limited, trading as taylor & Francis group.