Introduction: The High Stakes of Sepsis and Resistance

Gram-negative bloodstream infections (GN-BSI) remain among the most severe and lethal infectious diseases, particularly in intensive care units and regions where multidrug resistance (MDR) is endemic. Mortality rates can exceed 50% in MDR cases, and delays in initiating appropriate antimicrobial therapy are closely linked to poor outcomes. As carbapenem-resistant strains and other MDR organisms proliferate globally, clinicians face growing uncertainty when selecting empiric treatment—often resorting to broad-spectrum antibiotics that risk both undertreatment and overtreatment.1,2

This clinical conundrum has spurred interest in whether artificial intelligence (AI) can help bridge the diagnostic gap. Specifically, can AI models predict antibiotic resistance profiles in GN-BSI before formal susceptibility results become available? Early evidence suggests that AI-powered tools—integrating patient demographics, microbiology history, and clinical context—can support more precise empiric prescribing. Such innovations hold potential to improve survival and reduce unnecessary antibiotic use, a critical advance in the fight against antimicrobial resistance.3

The Problem: Slow Diagnostics, Empiric Guesswork

Traditional blood cultures and antimicrobial susceptibility testing (AST) remain the cornerstone of GN-BSI diagnosis but typically require 24 to 72 hours for actionable results. During this time, clinicians must initiate empiric therapy—often broad-spectrum regimens that are either too weak to control resistant pathogens or unnecessarily potent for susceptible ones. This “guesswork window” is clinically dangerous: inappropriate empiric therapy significantly increases mortality, while indiscriminate use of broad-spectrum antibiotics accelerates resistance.

Although rapid AST technologies exist and can shorten time-to-result by up to 40 hours, they are not universally implemented. Moreover, diagnostic limitations and workflow constraints hinder their widespread clinical impact. In this setting, predictive analytics powered by AI may offer an interim solution—providing early risk assessments that can inform treatment decisions while awaiting laboratory confirmation.4,5,6

The AI Approach: Predicting Resistance Before the Lab Does

AI-based models are designed to synthesize complex, multidimensional clinical data to estimate the likelihood that a patient’s bloodstream infection is resistant to specific antibiotics—before lab results are available. These models are trained on large retrospective datasets that include variables such as patient age, comorbidities, prior infection history, antibiotic exposures, hospital stay details, and geographic or institutional resistance trends.7,8

Different machine learning algorithms are applied depending on the clinical context and data complexity. Logistic regression is valued for its transparency and ease of interpretation; random forests excel at managing non-linear interactions; and deep learning models like neural networks can detect subtle patterns across high-dimensional datasets. These models output a probability score—indicating, for example, the likelihood of resistance to ceftriaxone or a carbapenem—which clinicians can use to guide empiric therapy decisions more accurately than with standard protocols alone.9,10,11

Importantly, these tools are not replacements for conventional microbiological diagnostics but rather complementary aids. Their value lies in enabling timely, informed interventions in the early phase of infection, when decisions are most consequential. Integrated into electronic health records (EHRs), AI models can deliver real-time resistance predictions at the bedside, supporting antimicrobial stewardship while improving patient outcomes.12

Reference:

  1. Tsachouridou O, Pilalas D, Nanoudis S, Antoniou A, Bakaimi I, Chrysanthidis T, Markakis K, Kassomenaki A, Mantzana P, Protonotariou E, Skoura L, Metallidis S. Mortality due to Multidrug-Resistant Gram-Negative Bacteremia in an Endemic Region: No Better than a Toss of a Coin. Microorganisms. 2023 Jun 30;11(7):1711. doi: 10.3390/microorganisms11071711. PMID: 37512883; PMCID: PMC10383448.

  2. Global and regional burden of bloodstream infections caused by carbapenem-resistant Gram-negative bacteria in 2019: A systematic analysis from the MICROBE database. https://doi.org/10.1016/j.ijid.2024.107769

  3. Assessing the burden of disease of gram-negative bloodstream infections in a Brazilian hospital: A retrospective cohort study from 2015 to 2019. https://doi.org/10.1016/j.ijregi.2024.100401

  4. Messiaen AS, Vandendriessche S, De Muynck E, Strubbe G, De Bus L, Schelstraete P, Decommer K, De Smet S, Soetens A, Naesens L, Timmermans K, De Waele JJ, Veld DHI', Boelens J. Impact of reporting rapid susceptibility results in Gram negative bloodstream infections: a real world prospective study. Eur J Clin Microbiol Infect Dis. 2025 Apr;44(4):847-853. doi: 10.1007/s10096-025-05046-3. Epub 2025 Jan 25. PMID: 39862301.

  5. Alviset S, Rodari S, Poupet H, Mizrahi A, Gastli N, Philippart F, Charpentier J, Pilmis B, Kernéis S. Potential impact of real-time processing and rapid susceptibility testing of blood samples in Gram-negative bloodstream infections in intensive care patients. Infect Dis Now. 2023 Oct;53(8):104773. doi: 10.1016/j.idnow.2023.104773. Epub 2023 Aug 22. PMID: 37619962.

  6. Dilek Ince, Suiyini Fiawoo, Rebecca Choudhury, Sara E Cosgrove, David Dobrzynski, Howard Gold, Jae Hyoung Lee, Kelly M Percival, Stephanie Shulder, Deepthi Sony, Emily S Spivak, Pranita D Tamma, Priya Nori, Epidemiology of Gram-Negative Bloodstream Infections in the United States: Results From a Cohort of 24 Hospitals, Open Forum Infectious Diseases, Volume 10, Issue 7, July 2023, ofad265, https://doi.org/10.1093/ofid/ofad265

  7. Bonazzetti C, Rocchi E, Toschi A, Derus NR, Sala C, Pascale R, Rinaldi M, Campoli C, Pasquini ZAI, Tazza B, Amicucci A, Gatti M, Ambretti S, Viale P, Castellani G, Giannella M. Artificial Intelligence model to predict resistances in Gram-negative bloodstream infections. NPJ Digit Med. 2025 May 29;8(1):319. doi: 10.1038/s41746-025-01696-x. PMID: 40442363; PMCID: PMC12122886.

  8. Machine learning to predict antibiotic susceptibility in Enterobacterales bloodstream infections compared to clinician prescribing. Kevin Yuan, Augustine Luk, Jia Wei, A Sarah Walker, Tingting Zhu, David W Eyre. medRxiv 2024.10.02.24314776; doi:https://doi.org/10.1101/2024.10.02.24314776

  9. Liang Q, Ding S, Chen J, Chen X, Xu Y, Xu Z, Huang M. Prediction of carbapenem-resistant gram-negative bacterial bloodstream infection in intensive care unit based on machine learning. BMC Med Inform Decis Mak. 2024 May 14;24(1):123. doi: 10.1186/s12911-024-02504-4. PMID: 38745177; PMCID: PMC11095031.

  10. Liao H, Xie L, Zhang N, Lu J, Zhang J. Advancements in AI-driven drug sensitivity testing research. Front Cell Infect Microbiol. 2025 May 2;15:1560569. doi: 10.3389/fcimb.2025.1560569. PMID: 40384974; PMCID: PMC12081381.

  11. Ferrari D, Arina P, Edgeworth J, Curcin V, Guidetti V, Mandreoli F, Wang Y. Using interpretable machine learning to predict bloodstream infection and antimicrobial resistance in patients admitted to ICU: Early alert predictors based on EHR data to guide antimicrobial stewardship. PLOS Digit Health. 2024 Oct 16;3(10):e0000641. doi: 10.1371/journal.pdig.0000641. PMID: 39413052; PMCID: PMC11482717.

  12. Yuan K, Luk A, Wei J, Walker AS, Zhu T, Eyre DW. Machine learning and clinician predictions of antibiotic resistance in Enterobacterales bloodstream infections. J Infect. 2025 Feb;90(2):106388. doi: 10.1016/j.jinf.2024.106388. Epub 2024 Dec 30. PMID: 39742978; PMCID: PMC11893473.