Development of a predictive model for FIC index of meropenem-colistin combination against carbapenem-resistant Acinetobacter baumannii

Background: Acinetobacter baumannii represents a critical antimicrobial resistance challenge in Vietnamese healthcare settings, with carbapenem resistance rates reaching 55-90%. Traditional checkerboard methodology for evaluating antibiotic combinations requires substantial time and resources, limiting routine application in research settings. Objective: To develop a predictive model for the FIC index of meropenem-colistin combination based on MIC values. Methods: A cross-sectional analytical study on 61 carbapenem-resistant A. baumannii strains (6/2022 - 12/2022). MICs were determined by broth microdilution, and synergy was evaluated through FIC index. Eight regression models were compared using 80/20 train-test split with 5-fold cross-validation for stability assessment. Results: All strains were resistant to meropenem but intermediate to colistin. The combination showed synergy in 77% and additivity in 23%. A logarithmic model FIC = 0.649 - 0.155 × ln(Mero) - 0.227 × ln(Col) achieved high accuracy (R² = 0.554). The model demonstrated 95% sensitivity and 75% specificity for synergy prediction with an overall classification accuracy of 88.5%. Conclusion: This is the first study applying machine learning for quantitative FIC prediction. While offering potential research applications, clinical implementation requires extensive validation, given recent evidence against therapeutic efficacy of this combination.