OBJECTIVE  To compare the differences in cytokines in blood and cerebrospinal fluid between bacterial meningitis (BM) and viral encephalitis (VE) in infants, and to construct a machine learning model for early differentiation and diagnosis.

METHODS  A total of 120 infants admitted to the neurology department of the Women and Children's Hospital of Ningbo University with suspected central nervous system infections from Jan. 2022 to Nov. 2024 were selected as the study subjects. Venous blood and cerebrospinal fluid samples were collected from all patients immediately at the admisson to the hospital. The patients were ultimately divided into three groups based on the gold standard: 40 in the BM group, 40 in the VE group and 40 in the non-central nervous system infection (NC) group. Differences in blood and cerebrospinal fluid cytokines among the three groups were compared. Eight machine learning training models were used, and the optimal model was selected based on indicators such as receiver operating characteristic (ROC) curves, calibration curves and decision curves. The SHAP method was employed to interpret the model.

RESULTS  The serum Interleukin-6 (IL-6) level in the BM group was 105.85 (25.19, 192.46) pg/ml, which was higher than that in the VE group 40.85 (12.68, 69.81) pg/ml and the NC group (P＜0.05). The VE group had a higher serum IL-6 level than the NC group (P＜0.05). The serum Interferon γ (IFN-γ) level in the VE group was 7.29 (1.66, 37.37) pg/ml, which was higher than that in the BM group 2.21 (1.12, 5.65) pg/ml and the NC group (P＜0.05). The serum IL-17A level in the BM group was 1.96 (0.92, 4.88) pg/ml, which was higher than that in the NC group (P＜0.05). The cerebrospinal fluid IL-6 and IL-10 levels in the BM group were 102.13 (31.38, 569.60) pg/ml and 15.88 (5.56, 43.79) pg/ml, respectively, which were both higher than those in the VE group 21.70 (9.64, 40.00) pg/ml and 8.50 (4.10, 15.05) pg/ml and the NC group (P＜0.05). The cerebrospinal fluid IL-6 and IL-10 levels in the VE group were both higher than those in the NC group (P＜0.05), and the cerebrospinal fluid IFN-γ level in the VE group was 1.61 (1.33, 2.74) pg/ml, which was higher than that in the BM group 1.13 (0.30, 1.75) pg/ml and the NC group (P＜0.05). Among the eight machine learning models constructed based on the above indicators, the Gradient Boosting Machine (GBM) model performed the best (area under the curve=0.938, accuracy=0.917, sensitivity=0.917, specificity=0.917, precision=0.917, F1 score=0.917). The SHAP analysis revealed that the features contributing to the model in descending order of importance were: cerebrospinal fluid IL-6, serum IFN-γ, serum IL-6, cerebrospinal fluid IL-10 and cerebrospinal fluid IFN-γ.

CONCLUSIONS  There is a correlation between cerebrospinal fluid IL-6, serum IL-6, cerebrospinal fluid IL-10 and infant BM, as well as between serum IFN-γ, cerebrospinal fluid IFN-γ and infant VE. These indicators are expected to serve as auxiliary markers for early differentiation between BM and VE.