Prediction and analysis of toxic and side effects of tigecycline based on deep learning

Background: In recent years, with the increase of antibiotic resistance, tigecycline has attracted much attention as a new broad-spectrum glycylcycline antibiotic. It is widely used in the treatment of complex skin and soft tissue infections, complex abdominal infections and hospital-acquired pneumonia by inhibiting bacterial protein synthesis. Tigecycline can exhibit significant time-dependent bactericidal activity, and its efficacy is closely related to pharmacokinetics. It can be evaluated by the ratio of AUC0-24 to the minimum inhibitory concentration (MIC) of pathogens. However, tigecycline may cause nausea, vomiting, diarrhea and a few patients have elevated serum aminotransferase, especially in critically ill patients. The safety of patients still needs further study.

Methods: In this study, the clinical data of 263 patients with pulmonary infection in Shengjing Hospital of China Medical University and the Second Affiliated Hospital of Dalian Medical University were collected retrospectively, and the hepatotoxicity prediction model was established. The potential correlation between the toxic and side effects of tigecycline and the number of hospitalization days was preliminarily discussed, and the correlation analysis between the number of hospitalization days and continuous variables was established. Finally, the deep learning model was used to predict the hospitalization days of patients through simulated blood drug concentration and clinical laboratory indicators.

Results: The degree of abnormal liver function was significantly correlated with AST, GGT, MCHC and hospitalization days. Secondly, the correlation between hospitalization time and clinical test indexes and simulated drug concentration was analyzed. It was found that multiple clinical laboratory parameters of patients (such as EO #, HCT, HGB, MCHC, PCT, PLT, WBC, AST, ALT, Urea), first dose (Dose), age and APACHE II score were significantly correlated with hospitalization days. The simulated blood drug concentration was correlated with the length of hospital stay from 12 h after administration, and reached the strongest between 24 and 48 h. The AUC of the liver function prediction model can reach 0.90. Further analysis showed that there was a potential correlation between hepatotoxicity and hospitalization days. The median hospitalization days of patients in the non-hepatotoxicity group, liver function injury group and hepatotoxicity group were 20, 23, and 30 days, respectively. Based on these results, the length of hospital stay was predicted by the deep learning prediction model with an error within 1 day.

Conclusion: In this study, the hospitalization days of infected patients were predicted by deep learning model with low error. It was found that it was related to clinical test parameters, hepatotoxicity and dosage after administration. The results provided an important reference for the clinical application of tigecycline, and emphasized the need to pay attention to its toxic and side effects in use.