Neglected brucellosis in pediatric populations from non-endemic regions: Clinical manifestations and prediction of severe disease in Yunnan Province, China.

BACKGROUND

Although Yunnan Province is not an endemic region for brucellosis, the disease remains a diagnostic and therapeutic challenge in children due to its atypical clinical manifestations and potential for severe complications.

OBJECTIVE

This study aims to explore the clinical features of pediatric brucellosis in the region and establish a prediction model for severe complications.

METHODS

This study included 62 children diagnosed with brucellosis at the Kunming Children's Hospital between 2015 and 2024. The patients were divided into two groups based on the presence of severe complications: the severe complications group (n = 15) and the general group (n = 47). Clinical features were extracted from electronic medical records, and the Boruta algorithm was used to select core predictive factors. Six machine learning models, including Random Forest and XGBoost, were constructed. The performance of the models was assessed using receiver operating characteristic curve (ROC) curves and decision curve analysis (DCA), and a web-based prediction tool was developed.

RESULTS

The study revealed that the most common clinical symptoms were fever (95.2%), joint pain (51.6%). Meningoencephalitis was observed in 13 cases (21%), and sacroiliitis was present in 2 cases (3%). Laboratory findings indicated that the erythrocyte sedimentation rate (ESR) and IgM levels were significantly higher in the severe complications group compared to the general group. Culture results showed that the positive rate of bone marrow cultures was 95% (19/20), blood cultures had a positive rate of 84% (52/62), synovial fluid cultures had a positive rate of 67% (2/3), and cerebrospinal fluid cultures had a low positive rate of 2% (1/43). Machine learning models demonstrated that the Random Forest model performed best in predicting severe complications (AUC = 0.970), and DCA indicated that it had the best clinical utility. Key predictive factors were disease duration, fever duration, IgM, and ESR. A Shiny-based web tool was developed for real-time clinical risk assessment.

CONCLUSION

This study indicated that pediatric brucellosis should not be neglected in non-endemic areas like Yunnan Province, China. Combining inflammatory markers with Random Forest models can effectively predict the risk of severe complications in pediatric brucellosis.