Enhancing Antibiotic Stewardship: �A Machine Learning Approach to Predicting Antibiotic Resistance in Inpatient Care

​

Infectious Diseases - Going Viral

S19

​

Fateme Nateghi, Manoj Maddali

Stanford University

Connect at fnateghi@stanford.edu, mmaddali@stanford.edu

​

​

​

​

SPEAKER DISCLOSURE

​

Funding: This research was supported in part by NIAID R01 funding.

​

Data Infrastructure: We would also like to acknowledge the support from CTSA resources for data infrastructure.

​

Consulting: I also disclose that I provided consulting services to ISHI Health.

​

​

​

​

​

​

​

​

​

​

​

​

​

​

LEARNING OUTCOMES

​

After this session, attendees will be able to:

​

Antibiotic resistance is a growing threat to global human health

• “Antibiotic resistance is an urgent global public health threat” – CDC
• Associated with 35,000 deaths/year in the US alone
• ~1.3 million yearly deaths worldwide due to antibiotic resistance, projected to increase over the next 10 years.

Global burden of bacterial antimicrobial resistance in 2019. Lancet. 2022.

Stewardship is challenging

Heterogeneous practice patterns geographically and by institution type

Clinicians must make decisions days before culture results return

Diagnosis of infection can be "subjective" and based on clinical impression

Clinicians can be thus biased toward overly broad antibiotic prescription

Motivation: Develop a parsimonious classifier model to predict antibiotic sensitivity and facilitate prospective evaluation

​

​

​

​

​

Focused on 5 common IV antibiotics

Antibiotics saved by NPV for resistance

* Number of orders placed for that antibiotic in the next 24 hours after culture order.

Conclusion

Key Takeaways:

• Personalized predictions reduce broad-spectrum antibiotic use.
• High NPV targets improve safety and coverage.

​

Clinical Impact:

• Enhances antibiotic stewardship.
• Supports real-time decision-making in clinical settings.

​

Future Directions:

• Validate the model with multi-site datasets.
• Incorporate real-time feedback for continuous improvement.

​

Acknowledgements