Objective To explore the predictive value of changes in monocytes (M), lymphocytes (L), white blood cell-to-platelet ratio (WPR), neutrophil-to-lymphocyte ratio (NLR) in peripheral blood and changes in temperature (T-37) for bacteremia and 90-day mortality in patients with severe burns during the first perioperative period.Methods Data from 169 patients treated at the First Affiliated Hospital of Sun Yat-sen University from 2011 to 2020 were retrospectively analyzed. The patients were divided into a bacteremia group and a control group based upon blood culture. Data on M, L, NLR, WPR and T-37 were collected at the day before the first surgery (0), the first day after the first surgery (1) and the third day after the first surgery (3). Independent risk factors for bacteremia and 90-day mortality were identified. The risk prediction models were established with multivariate logistic regression analysis and externally validated in another cohort of 191 patients with severe burns.Results The 90-day mortality rate was 21.3% (36/169), which was significantly higher in the Bac group than in the con group (42.8% vs. 10.6%, P<0.01). WPR0, M0, WPR1, and T3-37 were significantly higher in the Bac group than in the con group (P<0.01), whereas M1 and M3 were significantly lower in the Bac group than in the con group (P<0.01). Multivariate regression analysis showed that SOFA0, WPR0, M3, and T3-37 were independently associated with bacteremia. The prediction model for bacteremia Xbac = 0.1809 × SOFA0 + 6.532 × WPR0-1.171 × M3 + 0.6987 × ⎮T3-37⎮- 2.297. TBSAB, SOFA0, and ∆M (M0-M3) were independently correlated with 90-day mortality. The risk prediction model X90d-m= 0.055 × TBSAB + 0.301×SOFA0+1.508 × ∆M) - 7.196. External validation of the prediction models suggested that the specificity, sensitivity, and AUC were 90.7%, 62.5% and 0.797, respectively, for Xbac and 69.2%, 90.0% and 0.873, respectively, for X90d-m.Conclusion ∆M and WPR0 were independent risk factors for bacteremia and 90-day mortality in patients with severe burns. The predictive models could inform clinical antimicrobial judgment and prognostication.