Antimicrobial resistance is a growing threat to public health worldwide. In hospitals, if antibiotics do not treat severe infections effectively this can lead to more deaths and longer hospital stays. However, prompt adjustments to antibiotic treatments to switch to an antibiotic that works may potentially improve outcomes for patients.
A recent study has shown that using past test results from other patients can be used to guide treatment for patients not responding to first-line antibiotics. We will extend the relatively simple population-level approach in the previous study to make personalized recommendations to guide escalation of antibiotic treatment in patients who do not respond well to initial treatment. We will develop statistical and machine learning approaches to predict resistance to commonly used antibiotics in patients who have infections resistant to first-line antibiotics. We will use multiple types of patient data (e.g., patient demographics, past medical conditions, vital signs, lab test results) to make predictions.
See publications:
Machine learning and clinician predictions of antibiotic resistance in Enterobacterales bloodstream infections
Transformers and large language models are efficient feature extractors for electronic health record studies