Prediction of Antimicrobial Resistance in People Living With Cystic Fibrosis Using Machine Learning

Area of Science:

• Medical Informatics
• Infectious Diseases
• Pulmonology

Background:

• Antimicrobial resistance (AMR) poses a significant threat to individuals with cystic fibrosis (pwCF), often necessitating prompt antibiotic treatment.
• Standard antibiotic susceptibility testing delays treatment initiation, as results typically take several days.

Purpose of the Study:

• To evaluate the potential of electronic healthcare records (EHR) in predicting antibiotic resistance in sputum cultures from adult pwCF.
• To inform earlier treatment decisions for lung infections in pwCF.

Main Methods:

• Retrospective analysis of 12,618 sputum cultures from 209 pwCF (2012-2022).
• Linking EHR data including antibiotic exposure, spirometry, demographics, and historical resistance patterns.
• Utilizing machine learning models, including extreme gradient boosting, for resistance prediction.

Main Results:

• Extreme gradient boosting models achieved consistent performance with AUCs ranging from 0.75-0.80.
• Model accuracy was comparable for cultures with and without Pseudomonas aeruginosa, with exceptions for ceftazidime and meropenem.
• Historical resistance data proved more influential than recent results in predicting future resistance.

Conclusions:

• EHR-derived data can feasibly estimate antimicrobial resistance prior to definitive culture results.
• Further validation is required, including broader antibiotic exposure data and temporal shift assessment, before clinical implementation.