Abstract
BACKGROUND: Brucellosis is endemic in north-western China, and Brucella spp. bacteraemia is a clinically important manifestation that may indicate more severe disease. However, robust and clinically interpretable predictors to prioritise blood cultures in resource-limited county-level hospitals are lacking. We aimed to compare admission clinical and routine laboratory features between bacteraemic and non-bacteraemic brucellosis and to develop a simple, clinically interpretable model to predict bacteraemia. METHODS: We retrospectively analysed 518 adults with brucellosis admitted to two county-level hospitals in Ningxia, China, from January 2014 to June 2024. Brucellosis was confirmed by blood culture and/or concordant admission serology (Rose Bengal test positivity with a standard tube agglutination test titre ≥ 1:160). Brucella spp. bacteraemia was defined as ≥ 1 positive blood culture obtained at admission; non-bacteraemic infection was defined as all admission blood cultures negative. We compared demographic, clinical, and routine laboratory variables and fitted multivariable logistic regression models with bootstrap internal validation and calibration assessment. RESULTS: Of 518 patients, 198 were bacteraemic and 320 were non-bacteraemic. Bacteraemic patients were older, more often had acute-onset disease, and more frequently had objectively documented fever and chills than non-bacteraemic patients. They also had higher inflammatory and liver-injury markers and lower albumin concentrations. For single predictors, the area under the receiver operating characteristic curve (AUC, 95% confidence interval) was 0.827 (0.790-0.863) for LDH, 0.749 (0.700-0.799) for CRP, 0.717 (0.670-0.765) for albumin (lower concentrations), and 0.688 (0.638-0.731) for ALT. The combined clinical-laboratory model achieved an AUC of 0.906 (0.878-0.929). CONCLUSIONS: A combined clinical-laboratory model based on readily available admission data showed excellent discrimination for early risk stratification of bacteraemia among adults with brucellosis, with acceptable internal calibration, and warrants external validation in other endemic, resource-limited settings.