Evaluation of a Machine Learning Model Based on Laboratory Parameters for the Prediction of Influenza A and B in Chongqing, China: Multicenter Model Development and Validation Study.

BACKGROUND

Influenza viruses are major pathogens responsible for acute respiratory infections in humans, which present with symptoms such as fever, cough, sore throat, muscle pain, and fatigue. While molecular diagnostics remain the gold standard, their limited accessibility in resource-poor settings underscores the need for rapid, cost-effective alternatives. Routine blood parameters offer promising predictive value but lack integration into intelligent diagnostic systems for influenza subtyping.

OBJECTIVE

This study aimed to develop a machine learning model using 24 routine blood parameters to predict influenza A and B infections and validate a deployable diagnostic tool for low-resource clinical settings.

METHODS

In this multicenter retrospective study, 6628 adult patients (internal cohort: n=2951; external validation: n=3677) diagnosed with influenza A virus infection (A+ group), influenza B virus infection (B+ group), or those presenting with influenza-like symptoms but testing negative for both viruses (A-/B- group) were enrolled from 3 hospitals between January 2023 and May 2024. The CatBoost (CATB) algorithm was optimized via 5-fold cross-validation and random grid search using 24 routine blood parameters. Model performance was evaluated using metrics such as the area under the curve (AUC), accuracy, specificity, sensitivity, positive predictive value, negative predictive value, and F1-score across internal testing and external validation cohorts, with Shapley additive explanations analysis identifying key biomarkers. The Artificial Intelligence Prediction of Influenza A and B (AI-Lab) tool was subsequently developed on the basis of the best-performing model.

RESULTS

In the internal testing cohort, 7 models (K-nearest neighbors, naïve Bayes, decision tree, random forest, extreme gradient boosting, gradient-boosting decision tree, and CatBoost) were evaluated. The AUC values for diagnosing influenza A ranged from 0.788 to 0.923, and those for influenza B from 0.672 to 0.863. The CATB-based AI-Lab model achieved superior performance in influenza A detection (AUC 0.923, 95% CI 0.897-0.947) and influenza B (AUC 0.863, 95% CI 0.814-0.911), significantly outperforming conventional models (K-nearest neighbors, RF, and XGBoost; all P<.001). During external validation, AI-Lab demonstrated high performance, achieving an accuracy of 0.913 for differentiating the A+ group from the A-/B- group and 0.939 for distinguishing the B+ group from the A-/B- group.

CONCLUSIONS

The CATB-based AI-Lab tool demonstrated high diagnostic accuracy for influenza A and B subtyping using routine laboratory data, achieving performance comparable to rapid antigen testing. By enabling timely subtype differentiation without specialized equipment, this system addresses critical gaps in managing influenza outbreaks, particularly in resource-constrained regions.