Explainable machine learning for discontinuation of therapeutic antibiotics in intensive care patients
Authors' conclusion
Although prediction of reinitiation remained challenging, the most important and consistent predictor for reinitiation of antibiotics was a shorter duration of the last started antibiotic.
Study context
- Population
- Adult ICU patients treated with antibiotics in two tertiary academic hospitals in the Netherlands, October 2015 to January 20223,645 antibiotic courses
- Participants
- 2,486 patients
- Exposure
- Logistic regression, lightGBM and AutoPrognosis models to predict antibiotic reinitiation
- Outcome
- Restarting antibiotics within 72 h
- Follow-up
- 72 h
Methodological detailstwo tertiary academic hospitals, explainable machine learning prediction model
- Two tertiary academic hospitals
- Explainable machine learning prediction model
Main finding
As reported in the abstract
Primary outcome occurrence
708
19.4 %
Same subgroup restarted
468
66.10 %
Exact same combination restarted
278
39.27 %
Best performing model: logistic regression, AUROC
0.675
Observational data: associations, not effects.
Further findings
Cultures collected / positive
%
- Cultures collected
- 80.4
- Cultures positive
- 43.3
90-day mortality, reinitiation vs no reinitiation
%
- Reinitiation
- 39.8
- No reinitiation
- 25.0
Original publication
Explainable machine learning for discontinuation of therapeutic antibiotics in intensive care patients. J Crit Care. Published 27 August 2025. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →