Predictive Modelling of nephrotoxicity from vancomycin and piperacillin-tazobactam (VPT) combination and vancomycin as monotherapy: an artificial intelligence (AI) Approach
Introduction: The concomitant use of vancomycin with β-lactam antibiotics, such as piperacillin-tazobactam, presents a clinical challenge due to the heightened risk of acute kidney injury (AKI). This study aims to develop a predictive model using an artificial intelligence (AI) approach to elucidate...
Saved in:
| Main Authors: | , , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2025-03-01
|
| Series: | International Journal of Infectious Diseases |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S1201971224007719 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Introduction: The concomitant use of vancomycin with β-lactam antibiotics, such as piperacillin-tazobactam, presents a clinical challenge due to the heightened risk of acute kidney injury (AKI). This study aims to develop a predictive model using an artificial intelligence (AI) approach to elucidate the relationship between vancomycin dosage, vancomycin and piperacillin-tazobactam (VPT) combination co-administration and AKI incidence. Methods: De-identified, unique ward prescriptions and comorbidities data (n=30,242) between 01.01.2020 and 31.12.2023 from Imperial College Healthcare NHS Trust (ICHT), hosted by the Imperial BRC funded Imperial Clinical Analytics, Research and Evaluation (iCARE) secure infrastructure, included real-time inpatient electronic health records from 5 sites of ICHT: St Mary's Hospital, Hammersmith Hospital, Charing Cross Hospital, Queen Charlotte and Chelsea Hospital, and the Western Eye Hospital. A binomial regression and random forest model were constructed to produce preliminary data which assessed the impact of vancomycin dosage on AKI likelihood. Neural network and decision tree models will be built to assess accuracy of prediction. The same methods will be utilised to assess probability of developing AKI with increased dosage of piperacillin-tazobactam and vancomycin co-administration. Results: Our preliminary data from the binomial regression and random forest models indicate that the best model was random forest, with a mean AUC score of 0.886, which showed an increase in AKI with increased vancomycin exposure. This aligns with our hypothesis that as vancomycin dosage escalates, the odds of AKI occurrence increase. Neural network and decision tree models will also be built to assess accuracy of prediction across these machine learning algorithms. From utilising the same methods, the probability of developing AKI from increased dosage of VPT will also be assessed. We expect that the AKI likelihood will be further increased upon patients receiving VPT in comparison to vancomycin monotherapy. Discussion: The empirical combination of vancomycin with piperacillin-tazobactam has become standard practice in hospital settings. However, concerns regarding nephrotoxicity have been reported within literature, where three to four-fold increase in AKI rates have been observed clinically among patients receiving vancomycin and piperacillin-tazobactam concurrently compared to vancomycin monotherapy. Despite challenges in conducting randomised trials due to ethical considerations, retrospective analyses emphasise the heightened nephrotoxic potential of this antibiotic combination. Our study attempts to eliminate this gap in understanding through a data-driven approach. Conclusion: Our results may provide clinicians with actionable insights into optimising empirical antimicrobial regimens while minimising nephrotoxicity risks through the use of AI, which can support better clinical decision-making. |
|---|---|
| ISSN: | 1201-9712 |