A prediction model based on routine blood indicators for the early diagnosis of serious bacterial infection in infants: a case-control study (Preprint)

BACKGROUND Routine blood test is an easily accessible, more economical, quick and routine method to examine infectious-related diseases. The prediction effect of routine blood parameters has rarely reported in the diagnosis of serious bacterial infections (SBI). OBJECTIVE This study aims to develop a model for early diagnosis of SBI using routine blood parameters. METHODS This case-control study was conducted based on the data collected from the Medical Information Mart for Intensive Care III (MIMIC-III) included children under one year. SBI was defined as urinary tract infections, meningitis, and sepsis. Lasso regression was used to screen the potential determiners. The model was developed with random forest analysis based on routine blood parameters, and performance was assessed through calculating the area under the curve (AUC). Propensity score-matching (PSM) method was used to eliminate the impact of ethnicity-related bias. RESULTS After PSM, a total of 1,160 participants were included, with 232 in SBI group and 928 in non-SBI group. Red blood cell distribution width (RDW), hemoglobin (HGB), neutrophil, platelets (PLT) counts, monocyte, white blood cell (WBC) counts, mean corpuscular volume (MCV), neutrophil-to-lymphocyte ratio (NLR), monocyte-to-lymphocyte ratio (MLR), and systemic immune-inflammation index (SII) were screened out to develop the model, with AUC of 0.824 [95% confidence interval (CI): 0.789-0.859]. The variable importance of random forest demonstrated that WBC, MCV, SII and other top-ranked variables were shown as significant factors for the diagnosis of SBI. CONCLUSIONS Our model based on routine blood parameters showed a good performance, indicating the availability of this model in the clinic.