Development of a clinical decision support system for predicting antibiotic resistance

Objective: This project aims to develop a clinical decision support system prototype to predict antibiotic resistance in patients with urinary tract infections. Method: In the present study, decision trees, KNN, logistic regression, XGBoost, and random forest methods were used to predict the resistance of urinary tract infection pathogens to cephalosporins, TZP, carbapenems, TMP-SMX, and fluoroquinolones. By analyzing the internal workings of the models using SHAP analysis, the significant variables in predicting antibiotic resistance were identified, and a clinical decision support system prototype was developed. Results: The random forest method demonstrated superior performance compared to other methods, with AUC values of 0.777, 0.864, 0.877, 0.881, and 0.884 for predicting resistance to cephalosporins, TZP, carbapenems, TMP-SMX, and fluoroquinolones, respectively, in the training set; and AUC values of 0.638, 0.630, 0.665, 0.670, and 0.721, respectively, in the test set. According to SHAP analysis, the number of previous admissions, the first culture time, chronic lower respiratory tract diseases, pre-infection drug use, and duration of use were important variables for predicting resistance to all antibiotics. Conclusion: This study provides a foundation for the development of a machine learning-based clinical decision support system prototype for predicting antibiotic resistance in patients with urinary tract infections. With future improvements, the system has the potential to support clinical decision-making processes and serve as an effective tool in combating antibiotic resistance.