• HOME
  • NEWS
  • EXPLORE
    • CAREER
      • Companies
      • Jobs
    • EVENTS
    • iGEM
      • News
      • Team
    • PHOTOS
    • VIDEO
    • WIKI
  • BLOG
  • COMMUNITY
    • FACEBOOK
    • INSTAGRAM
    • TWITTER
Friday, November 20, 2026
BIOENGINEER.ORG
No Result
View All Result
  • Login
  • HOME
  • NEWS
  • EXPLORE
    • CAREER
      • Companies
      • Jobs
        • Lecturer
        • PhD Studentship
        • Postdoc
        • Research Assistant
    • EVENTS
    • iGEM
      • News
      • Team
    • PHOTOS
    • VIDEO
    • WIKI
  • BLOG
  • COMMUNITY
    • FACEBOOK
    • INSTAGRAM
    • TWITTER
  • HOME
  • NEWS
  • EXPLORE
    • CAREER
      • Companies
      • Jobs
        • Lecturer
        • PhD Studentship
        • Postdoc
        • Research Assistant
    • EVENTS
    • iGEM
      • News
      • Team
    • PHOTOS
    • VIDEO
    • WIKI
  • BLOG
  • COMMUNITY
    • FACEBOOK
    • INSTAGRAM
    • TWITTER
No Result
View All Result
Bioengineer.org
No Result
View All Result
Home NEWS Science News Health

Hospital Algorithm Predicts Resistant Gram-Negative Infections at Initial Culture Ordering

by
November 20, 2026
in Health
Reading Time: 5 mins read
0
Hospital Algorithm Predicts Resistant Gram-Negative Infections at Initial Culture Ordering

Hospital Algorithm Predicts Resistant Gram-Negative Infections at Initial Culture Ordering

Share on FacebookShare on TwitterShare on LinkedinShare on RedditShare on Telegram

Medical researchers have developed and externally validated a predictive computational framework named ORDER-CARB that forecasts the probability of carbapenem resistance in Gram-negative bacterial infections at the exact moment a clinical culture is ordered. In hospital medicine, physicians treating suspected severe bacterial infections routinely must select empirical antibiotic therapy during a critical 36-to-48-hour window before laboratory pathogen identification and formal antimicrobial susceptibility testing are finalized. Because carbapenem antibiotics represent a vital defense reserved for severe infections and pathogens with limited alternative therapies, inappropriate empirical treatment can lead to therapeutic failure, whereas broad overuse accelerates resistance. ORDER-CARB was engineered to bridge this diagnostic gap by using structured electronic health record data recorded prior to culture collection to provide early risk stratification, assisting clinicians and stewardship teams before microbiology results return.

The study, published in Infectious Diseases and Therapy by an investigative team including Emmanuel Pio Pastore, Daniele Roberto Giacobbe, Antonio Vena, and Matteo Bassetti, evaluated model performance across internal chronological development cohorts and two geographically distinct hospital systems. The primary analysis examined microbiologically labeled urine, respiratory, and blood culture orders that subsequently yielded at least one Gram-negative organism with interpretable susceptibility to imipenem, meropenem, doripenem, or ertapenem. In internal temporal testing, the frozen model identified subsequent carbapenem-resistant isolates with an area under the receiver operating characteristic curve of 0.858. When transported to external cohorts and recalibrated using older local hospital records, the algorithm maintained high discriminative ranking, achieving an area under the receiver operating characteristic curve of 0.797 at East Carolina University Health and 0.876 at the University of Texas Southwestern Medical Center.

To prevent artificial inflation of predictive accuracy, the investigators established strict chronological boundaries and leakage controls designed around the real-world clinical decision point. Many retrospective machine learning studies inadvertently introduce temporal leakage by incorporating variables influenced by downstream diagnostic workflows, such as post-collection lab values or preliminary pathogen classifications. In contrast, ORDER-CARB was strictly restricted to information documented in the electronic record prior to the index culture order. Eligible predictors included patient demographics, care location, prior procedure records, nursing home exposure, and historical laboratory summaries, alongside longitudinal microbiology records and prior antimicrobial exposures. Any microbiological evidence generated by the index event itself, including organism identity, culture descriptions, and minimum inhibitory concentration values, was rigorously excluded from the feature pipeline.

The developmental foundation of ORDER-CARB relied on the Antibiotic Resistance Microbiology Dataset, an electronic health record repository comprising 72,691 eligible culture orders. Following data cleaning and pre-index eligibility filtering, the internal cohort was partitioned chronologically into a development subset of 36,270 culture orders and an internal temporal test set of 15,545 culture orders. The development cohort was subsequently split into a model-fitting split of 29,016 orders and a temporally later calibration set of 7,254 orders. The core algorithm was a class-weighted random forest model selected via grid search, which yielded a cross-validated area under the receiver operating characteristic curve of 0.934 during fitting before undergoing sigmoid probability calibration on the internal calibration split. Across these internal splits, the baseline prevalence of carbapenem resistance shifted markedly from 32.4 percent in the model-fitting split to 11.3 percent in the internal temporal testing split.

External transportability was assessed using independent datasets from East Carolina University Health and the University of Texas Southwestern Medical Center, both structured within the same repository schema. The external datasets were divided chronologically into historical baseline subsets and untouched future evaluation hold-outs. The older historical portions were split into an initial 30 percent block dedicated to site-specific logistic recalibration on the logit scale and a subsequent 70 percent block utilized for selecting operational decision thresholds. The final future hold-outs comprised 29,788 culture orders from East Carolina University Health and 17,744 culture orders from the University of Texas Southwestern Medical Center. In both external environments, observed resistant-event prevalence was substantially lower than in the developmental cohort, ranging from 3.0 percent to 3.6 percent across subsets.

Because clinical adoption requires translating continuous probability scores into practical workflows, the researchers tested rule-out threshold strategies tailored to hospital antibiotic stewardship teams. Using older local data, thresholds were calibrated to identify approximately 9 out of 10 resistant isolates by enforcing a minimum historical sensitivity floor of 0.90, aiming to prompt early review of empirical coverage while susceptibility assays remained pending. When these historical cutoffs were applied to future hold-out data, the algorithm achieved a future sensitivity of 91.6 percent at East Carolina University Health and 85.7 percent at the University of Texas Southwestern Medical Center. However, due to low baseline resistance prevalence, these thresholds prompted review in 55.4 percent and 27.7 percent of all eligible cultures, yielding positive predictive values of 5.7 percent and 9.4 percent, respectively.

Operational workload analyses detailed the institutional burden associated with maintaining these sensitivity floors across divergent hospital environments. At East Carolina University Health, maintaining the 0.90 sensitivity threshold resulted in approximately 554 flagged cultures per 1,000 orders, leaving 2.9 missed resistant cultures per 1,000, which translated into 16.6 false-positive reviews and 17.6 total culture reviews per true-positive identification. At the University of Texas Southwestern Medical Center, the same historical target produced 277 flagged cultures per 1,000 orders, with 4.3 missed resistant cultures per 1,000, corresponding to 9.6 false-positive reviews and 10.6 total reviews per true positive. A secondary analysis exploring negative-predictive-value-constrained thresholds demonstrated that high negative predictive values could paradoxically coincide with unacceptable clinical capture; for example, a negative predictive value of 0.966 yielded an operational sensitivity of only 2.3 percent.

To evaluate decision-making across all patients undergoing evaluation rather than only those with confirmed Gram-negative isolates, the investigators conducted a secondary analysis focusing on bloodstream infection episodes. Blood culture orders collected within 48-hour windows were aggregated into discrete episodes encompassing culture-negative presentations, Gram-positive bacteria, fungi, and Gram-negative organisms. This clinical analysis implemented a two-stage modeling approach combining the transported ORDER-CARB resistance probability with a locally fitted model predicting Gram-negative bacteremia. In future hold-out testing involving 55,226 episodes at East Carolina University Health and 37,372 episodes at the University of Texas Southwestern Medical Center, the two-stage model achieved bloodstream infection discrimination areas under the curve of 0.928 and 0.967, respectively, although the total number of carbapenem-resistant events remained low at 35 and 18 cases.

Exploratory model interpretation revealed that historical microbiological findings served as the principal anchor of the algorithm’s predictive architecture. Prior isolation of Pseudomonas species dominated global feature importance and predictor ablation rankings across multiple analyses. Conversely, documented history of Escherichia species isolation and previous nitrofurantoin exposure correlated with lower resistance risk, reflecting recurrent, lower-risk urinary phenotypes. Additional predictive weight was derived from prior Staphylococcus species isolation, cumulative historical organism counts, antibiotic administration timing, intensive care admission, nursing home residence, and specific culture-ordering modes. These features functioned as modifiers capturing accumulated therapeutic pressure and clinical exposure, demonstrating that the model leveraged microbiological ecology and care context rather than non-specific demographic markers.

The study authors emphasized critical methodological limitations, noting that ORDER-CARB is designed as an informational risk stratification tool rather than an autonomous prescribing engine. The primary model is conditional on cultures subsequently yielding Gram-negative organisms, meaning its operational predictive values cannot be generalized directly to unselected bedside orders without two-stage adjustment. Furthermore, retrospective data lacked clinical adjudication between true invasive infection, colonization, or specimen contamination, and quantitative colony counts were unavailable. Because local resistance prevalence strongly dictates positive predictive values and review workloads, the authors concluded that implementation requires site-specific recalibration and tailored operational thresholds, alongside prospective silent validation to verify real-time feature stability before clinical deployment.

Subject of Research: Clinical evaluation and transportability of an electronic health record model predicting carbapenem resistance at culture ordering

Article Title: Early Prediction of Carbapenem Resistance in Gram-Negative Organisms at Culture Ordering from Routinely Collected Hospital Data: Temporal and External Validation of ORDER-CARB

Article References: Pastore, E. P., Giacobbe, D. R., Vena, A., & Bassetti, M. (2026). Early Prediction of Carbapenem Resistance in Gram-Negative Organisms at Culture Ordering from Routinely Collected Hospital Data: Temporal and External Validation of ORDER-CARB. Infectious Diseases and Therapy, 15(11), 3151-3167. https://doi.org/10.1007/s40121-026-01441-1

Image Credits: AI Generated

DOI: 10.1007/s40121-026-01441-1

Keywords: antimicrobial resistance, carbapenem resistance, Gram-negative bacteria, ORDER-CARB, clinical decision support, antimicrobial stewardship, electronic health records, predictive modeling, bloodstream infections, external validation, model transportability, logistic recalibration, random forest, diagnostic stewardship, bacterial culture, empirical antibiotic therapy

News Source: Ophelia Keating. (November 20, 2026). Hospital Algorithm Predicts Resistant Gram-Negative Infections at Initial Culture Ordering. Scienmag.

Tags: Antimicrobial ResistanceAntimicrobial Stewardshipbacterial culturebloodstream infectionsCarbapenem resistanceClinical decision supportdiagnostic stewardshipelectronic health recordsempirical antibiotic therapyexternal validationGram-negative bacterialogistic recalibrationmodel transportabilityORDER-CARBpredictive modelingRandom Forest
Share12Tweet7Share2ShareShareShare1

Related Posts

Intensive Care Microlearning Program Associated With Higher CRRT Knowledge Scores

Intensive Care Microlearning Program Associated With Higher CRRT Knowledge Scores

November 20, 2026
Language Models Improve Accuracy of Medical Labels in Large Chest Scan Dataset

Language Models Improve Accuracy of Medical Labels in Large Chest Scan Dataset

November 20, 2026

Euroanaesthesia Sessions Shifted Toward Technology as Female Speaker Share Rose

November 20, 2026

Laparoscopic Colonoscopic Technique Removes Complex Benign Polyps Without Major Resection

November 20, 2026

POPULAR NEWS

  • Ethiopian Study Documents 70 Wild Edible Plant Species Across Aseko District

    Ethiopian Study Documents 70 Wild Edible Plant Species Across Aseko District

    29 shares
    Share 12 Tweet 7
  • Meta-Analysis Finds Half of Surveyed Pakistani Populations Report Insomnia Symptoms

    29 shares
    Share 12 Tweet 7
  • Genotype D1 Hepatitis B Virus Protein Mutations Follow Phase-Dependent Structural Shifts

    29 shares
    Share 12 Tweet 7
  • Frequency Ratio Model Outperforms Competing Methods for Himalayan Landslide Susceptibility Mapping

    29 shares
    Share 12 Tweet 7

About

We bring you the latest biotechnology news from best research centers and universities around the world. Check our website.

Follow us

Recent News

Korea10K Maps Whole Genomes and Uncovers Epigenetic Array Biases

Intensive Care Microlearning Program Associated With Higher CRRT Knowledge Scores

Generative AI in High-Dimensional Biology Remains Insufficiently Validated in Small-Sample Studies

Subscribe to Blog via Email

Enter your email address to subscribe to this blog and receive notifications of new posts by email.

Join 89 other subscribers
  • Contact Us

Bioengineer.org © Copyright 2023 All Rights Reserved.

Welcome Back!

Login to your account below

Forgotten Password?

Retrieve your password

Please enter your username or email address to reset your password.

Log In
No Result
View All Result
  • Homepages
    • Home Page 1
    • Home Page 2
  • News
  • National
  • Business
  • Health
  • Lifestyle
  • Science

Bioengineer.org © Copyright 2023 All Rights Reserved.