Early prediction model for antibiotic treatment failure in bacteraemia: Japan bacteremia inpatient cohort association

Early antibiotic treatment failure (EATF) is a clinically important concept but remains underrecognized among frontline physicians. Moreover, no validated prediction models currently exist that specifically target EATF in patients with bacteraemia. The present study aims to develop and internally validate a clinical prediction model for EATF in hospitalised adults with bacteraemia.

This multi-centre retrospective cohort study included adult inpatients with bacteremia from the Japan Bacteremia Inpatient Cohort Association, conducted across eight community hospitals in Japan from 2018 to 2022. Based on review results, we defined EATF as an outcome that included changes in antibiotics and advanced treatment after 72 h of admission. Using 28 candidate variables from previous studies, models and clinical expertise, we developed our model through backward elimination based on the Akaike Information Criterion. After constructing the score model, discrimination was assessed using the area under the curve (AUC), while calibration was evaluated using a plot. Internal validation used the bootstrap method to assess AUC and overfitting.

Among the 1084 eligible patients, 388 experienced EATF. The model included abnormal white blood cell count, low serum albumin, occlusive lesions, transfer admission and oxygen therapy. Internal validation demonstrated an AUC of 0.66, 95% confidence interval (0.62, 0.69) and a calibration slope of 1.01.

Although the model's discrimination was modest, this study highlights the importance of EATF as a clinical outcome and demonstrates the inherent difficulty of developing predictive models for bacteraemia. External validation is warranted.