ARMD-Predict: multi-site machine learning to predict antibiotic resistance before culture results
Abstract
Clinicians select empiric antibiotics before culture susceptibility results are available. Electronic health record (EHR) models could inform that choice, but most are developed at a single hospital and transport poorly to others. We developed ARMD-Predict using 7.5 million culture-antibiotic observations from four academic medical centers. One set of models estimates susceptibility to individual antibiotics; a second, hierarchical set predicts the infecting organism and then organism-specific resistance phenotypes. Across 23 pooled susceptibility models evaluated in same-period, patient-grouped test partitions, median area under the receiver operating characteristic curve (AUROC) was 0.77 for inpatient urine and 0.70 for inpatient blood cultures; chronological retraining reduced AUROC by approximately 0.04. At an exploratory threshold with 95% positive predictive value, derived and evaluated on the same test observations, the model identified 55% of ceftriaxone-susceptible inpatient urine observations as suitable for narrower therapy and flagged 60% of ceftriaxone-resistant observations for broader coverage. Organism classifiers achieved top-3 accuracy of 51.3-83.3%; phenotype AUROCs were conditional on the observed organism. Research applications are available at https://armd-predict.stanford.edu. Independent calibration, threshold and end-to-end prospective validation are required before clinical use.
Related
- Response to comment on “Using genomic data and machine learning to predict antibiotic resistance: A tutorial paper”
- Fitness costs of antibiotic resistance mutations in rich media do not reliably predict those in infection-like environments
- Serum proteins predict incident inflammatory bowel disease 14 years before diagnosis: a machine-learning and Cox modelling in the UK Biobank
- A Machine Learning Approach to Predict Listeria monocytogenes Growth by Integrating FSSP and PMP Simulations: A Focus on t₁₀₀ as a Possible Endpoint Parameters for Predictive Models.
- Can Intra-Operative Stimulation Data Predict Post-Operative Deep Brain Stimulation Outcomes? — A Machine Learning Approach Based on Probabilistic Maps
BibTeX
@misc{haredasht2026armdpredict,
title = {{ARMD-Predict: multi-site machine learning to predict antibiotic resistance before culture results}},
author = {Fateme Nateghi Haredasht and Nicholas P. Marshall and Wenyuan Chen and Lauren N. Cooper and Ziming Wei and Harrison Konsker and Jacob S. Chen and Michael Klompas and Cara McKenna and Richard J. Medford and Sanjat Kanjilal and Austin J. Schoeffler and Stephen P. Ma and Niaz Banaei and Steven M. Asch and Amy Chang and Stanley C. Deresinski and Mary K. Goldstein and Robert Tibshirani and Jonathan H. Chen},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2610.32998},
eprint = {2610.32998},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.32998},
}
RIS
TY - UNPB
TI - ARMD-Predict: multi-site machine learning to predict antibiotic resistance before culture results
AU - Fateme Nateghi Haredasht
AU - Nicholas P. Marshall
AU - Wenyuan Chen
AU - Lauren N. Cooper
AU - Ziming Wei
AU - Harrison Konsker
AU - Jacob S. Chen
AU - Michael Klompas
AU - Cara McKenna
AU - Richard J. Medford
AU - Sanjat Kanjilal
AU - Austin J. Schoeffler
AU - Stephen P. Ma
AU - Niaz Banaei
AU - Steven M. Asch
AU - Amy Chang
AU - Stanley C. Deresinski
AU - Mary K. Goldstein
AU - Robert Tibshirani
AU - Jonathan H. Chen
PY - 2026
DA - 2026/10/06
PB - arXiv
DO - 10.48550/arxiv.2610.32998
UR - https://arxiv.org/abs/2610.32998
AB - Clinicians select empiric antibiotics before culture susceptibility results are available. Electronic health record (EHR) models could inform that choice, but most are developed at a single hospital and transport poorly to others. We developed ARMD-Predict using 7.5 million culture-antibiotic observations from four academic medical centers. One set of models estimates susceptibility to individual antibiotics; a second, hierarchical set predicts the infecting organism and then organism-specific resistance phenotypes. Across 23 pooled susceptibility models evaluated in same-period, patient-grouped test partitions, median area under the receiver operating characteristic curve (AUROC) was 0.77 for inpatient urine and 0.70 for inpatient blood cultures; chronological retraining reduced AUROC by approximately 0.04. At an exploratory threshold with 95% positive predictive value, derived and evaluated on the same test observations, the model identified 55% of ceftriaxone-susceptible inpatient urine observations as suitable for narrower therapy and flagged 60% of ceftriaxone-resistant observations for broader coverage. Organism classifiers achieved top-3 accuracy of 51.3-83.3%; phenotype AUROCs were conditional on the observed organism. Research applications are available at https://armd-predict.stanford.edu. Independent calibration, threshold and end-to-end prospective validation are required before clinical use.
ER -