Researchers at MIT and Harvard, using records from two Boston hospitals, built a machine learning model trained on patient history, prior antibiotic exposure and local resistance patterns to recommend the narrowest antibiotic likely to work for a given patient.
Tested retrospectively on 3,629 patients with urinary infections, the model's recommendations would have reduced use of broad-spectrum antibiotics while cutting prescriptions of second-line drugs by about 67 percent compared with what clinicians actually prescribed, while lowering the rate of ineffective treatment by 18 percent.
The analysis compared the algorithm's hypothetical choices against doctors' real decisions after the fact, rather than testing it prospectively on real patients in real time, so a live clinical trial is the next step. Overprescribing broad-spectrum antibiotics is a major driver of resistance, which is why even a retrospective signal this size drew attention.
Read the original study
Kanjilal et al., Science Translational Medicine, 2020 · doi.org/10.1126/scitranslmed.aay5067
A model picked narrower drugs without missing infections.
Source
- Kanjilal, S., Oberst, M., Boominathan, S., Zhou, H., Hooper, D. C., & Sontag, D. (2020). A decision algorithm to promote outpatient antimicrobial stewardship for uncomplicated urinary tract infection. Science Translational Medicine, 12(568). https://doi.org/10.1126/scitranslmed.aay5067