Prediction model for prolonged fever in patients with Mycoplasma pneumoniae pneumonia: a retrospective study of 716 pediatric patients

OBJECTIVE: To identify patients with Mycoplasma pneumoniae pneumonia (MPP) with a risk of prolonged fever while on macrolides. METHODS: A retrospective study was performed with 716 children admitted for MPP. Refractory MPP (RMPP-3) was defined as fever persisting for > 72 h without improvement in clinical and radiologic findings after macrolide antibiotics (RMPP-3) or when fever persisted for > 120 h (RMPP-5) without improvement in clinical and radiologic findings. Radiological data, laboratory data, and fever profiles were compared between the RMPP and non-RMPP groups. Fever profiles included the highest temperature, lowest temperature, and frequency of fever. Prediction models for RMPP were created using the logistic regression method and deep neural network. Their predictive values were compared using receiver operating characteristic curves. RESULTS: Overall, 716 patients were randomly divided into two groups: training and test cohorts for both RMPP-3 and RMPP-5. For the prediction of RMPP-3, a conventional logistic model with radiologic grouping showed increased sensitivity (63.3%) than the model using laboratory values. Adding laboratory values in the prediction model using radiologic grouping did not contribute to a meaningful increase in sensitivity (64.6%). For the prediction of RMPP-5, laboratory values or radiologic grouping showed lower sensitivities ranging from 12.9 to 16.1%. However, prediction models using predefined fever profiles showed significantly increased sensitivity for predicting RMPP-5, and neural network models using 12 sequential fever data showed a greatly increased sensitivity (64.5%). CONCLUSION: RMPP-5 could not be effectively predicted using initial laboratory and radiologic data, which were previously reported to be predictive. Further studies using advanced mathematical models, based on large-sized easily accessible clinical data, are anticipated for predicting RMPP.