A-083 Rapid screening for acute chikungunya virus infection using an interpretable machine learning model based on routine complete blood count parameters.

Abstract

Background The 2025 Chikungunya virus (CHIKV) outbreak in Foshan, China, highlighted an urgent need for rapid screening tools. Molecular diagnostics are constrained by time, cost, and availability. This study aimed to develop and validate an interpretable machine learning model using routine complete blood count (CBC) parameters to identify patients at high risk of CHIKV infection. Methods A retrospective cohort study was conducted on 1,537 suspected CHIKV patients from two medical centers in Foshan. CBC and cell population data (CPD) were obtained using Mindray BC-7500 analyzers. CHIKV infection was confirmed by reverse transcription polymerase chain reaction (RT-PCR). After rigorous data preprocessing and feature selection using LASSO regression, four models Logistic Regression (LR), Support Vector Machine (SVM), Random Forest (RF), and LightGBM (LGBM) were developed on a training set (n=1,132 from Center A) and evaluated on an internal test set (n=192) and an independent external test set (n=213 from Center B). Model performance was assessed by the area under the receiver operating characteristic curve (AUC), sensitivity, specificity, and net benefit. Results In the external validation set, the LR model demonstrated superior and robust performance (AUC = 0.807, sensitivity = 88.0%, specificity = 65.0%, NPV = 93.0%), significantly outperforming the tree-based models. Decision curve analysis confirmed its clinical utility across a wide range of risk thresholds. SHAP analysis revealed that lower white blood cell count (WBC_N), lower lymphocyte percentage (Lym%), and increased neutrophil morphological heterogeneity (D_Neu_SSC_CV) were the strongest predictors, aligning with the established pathophysiology of CHIKV infection. Conclusion We developed an interpretable LR model based on CBC parameters that can effectively screen for acute CHIKV infection with high sensitivity and NPV. This model serves as a reliable, rapid, and cost-effective tool for patient triage during outbreaks, especially in resource-limited settings awaiting confirmatory PCR testing.