Abstract
Background
Vancomycin is the standard therapy for methicillin-resistant Staphylococcus aureus (MRSA) infection; however, nephrotoxicity happened with a high incidence of 15%~40%. Weighting the risk before receiving vancomycin treatment facilitates timely prevention of nephrotoxicity, but no standardized strategy exists for this purpose.
Methods
A retrospective cohort study was performed. A total of 524 hospitalized patients treated with vancomycin were included in this study. They were divided into derivation cohort (n=341) and externally validation cohort (n=183) according to their admission time. Using univariate and multivariable logistic regression, we identified potential predictors of vancomycin-associated acute kidney injury (AKI) and developed a risk score by plotting nomogram. The predictive performance of this novel risk score was assessed and validated by discrimination and calibration. Besides, the risk score was also compared with existing prediction models according to integrated discrimination index (IDI) and net reclassification index (NRI).
Results
The incidence of AKI was 16.1% (55/341) in the derivation cohort and 16.4% (30/183) in the validation cohort. Three factors (vancomycin serum trough concentration, piperacillin/tazobactam and furosemide) were determined as predictors for vancomycin-associated AKI. The established three-item risk score showed a comparable discrimination in both derivation cohort (AUC=0.793, 95% CI: 0.732–0.855) and validation cohort (AUC=0.788, 95% CI: 0.698–0.877). The risk score also demonstrated a good calibration in the derivation cohort (χ2=6.079, P=0.638>0.05) and validation cohort (χ2=5.665, P=0.686>0.05). Compared with prediction by Cmin alone, this risk score significantly improved reclassification accuracy (IDI=0.050, 95% CI: 0.024–0.076, P<0.001, NRI=0.166, 95% CI: 0.044–0.289, P=0.007).
Conclusion
The established model in this study is a simplified three-item risk score, which provides a robust tool for the prediction of AKI after receiving vancomycin treatment.
Acknowledgments
We would like to thank our co-worker Youlei Wang for the instructive suggestions about the method part. We also acknowledged to some students, like Tingting Pan and Chencheng Xu for the help in data collection.
Disclosure
All authors declared no potential conflicts of interest.