• 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 Biology

Machine learning model predicts kidney injury in children with sepsis

by
November 6, 2026
in Biology, Technology
Reading Time: 4 mins read
0
Machine learning model predicts kidney injury in children with sepsis

Machine learning model predicts kidney injury in children with sepsis

Share on FacebookShare on TwitterShare on LinkedinShare on RedditShare on Telegram

Children with sepsis can develop sudden kidney failure within hours, and clinicians often lack reliable tools to spot which patients are at highest risk before the damage is done. A new multicenter study from China reports that an interpretable machine learning model, built from eight routine blood and clinical parameters, can predict sepsis-associated acute kidney injury in pediatric intensive care patients with strong accuracy, outperforming traditional scoring systems in most of the cohorts where it was tested.

The research, published in BioData Mining, analyzed data from 3,437 children with sepsis admitted to pediatric intensive care units at five Chinese centers. Overall, 25.7 percent of these patients developed sepsis-associated acute kidney injury, a complication the study notes significantly increases mortality in pediatric sepsis. Because early warning tools for children have remained scarce, the research team set out to build and validate a prediction model that could flag high-risk patients within the first 24 hours of admission.

The study design was deliberately conservative. Patients were divided into a training set, an internal temporal validation set, and four separate external validation cohorts, a structure intended to test whether the model would hold up on data it had never seen, including data collected at different times and different hospitals. Feature selection began with 41 clinical parameters that are routinely available in intensive care, and a rigorous screening process narrowed these down to the most predictive variables.

Ten machine learning algorithms were trained and optimized on the data, and the algorithm achieving the highest area under the curve, a standard measure of diagnostic discrimination, was selected as the final model. That winner was XGBoost, a gradient-boosted decision tree method that has become popular in clinical prediction research because it handles nonlinear relationships between variables well. The final model uses eight routinely available parameters, meaning it could in principle be deployed without any specialized tests beyond what pediatric intensive care units already measure.

The performance numbers were consistent across settings. In the training set, the model achieved an area under the curve of 0.899, with a 95 percent confidence interval of 0.88 to 0.91. In the internal temporal validation set, the area under the curve was 0.854, with a confidence interval of 0.81 to 0.89. Across the four external validation sets, the values ranged from 0.794 to 0.844, indicating that the model retained useful discrimination even when applied to patient populations from different hospitals.

Head to head comparisons with established clinical scores showed a clear advantage in most settings. The model significantly outperformed pSOFA, the pediatric Sequential Organ Failure Assessment score, and Phoenix, a newer sepsis score, in most cohorts, with all comparisons reaching statistical significance at p less than 0.05. The one exception was the PIC validation cohort, where the model’s discrimination was comparable to that of pSOFA. Calibration, meaning how well predicted probabilities match observed outcomes, was acceptable in most cohorts but poorer in the PIC cohort, where the model under-predicted kidney injury risk. This is an important caveat, because a model that systematically underestimates risk in a particular setting could lead clinicians to underestimate danger for those patients.

Beyond raw discrimination, the researchers examined whether the model would actually help clinical decisions. Decision curve analysis, a method that quantifies net benefit across a range of risk thresholds, confirmed net clinical benefit within the clinically plausible threshold range of 5 percent to 50 percent. The model also outperformed a treat-all strategy at thresholds above approximately 10 percent, suggesting it could help clinicians decide which children warrant intensified kidney-protective monitoring without unnecessarily flagging everyone.

A key feature of the study is its use of SHAP, or SHapley Additive exPlanations, a technique borrowed from game theory that assigns each input variable a contribution to each individual prediction. This transparency matters in medicine, where black-box predictions are difficult for clinicians to trust or act on. The SHAP analysis revealed that standard base excess, a measure of the body’s acid-base balance, aspartate aminotransferase, a liver enzyme, and potassium were the top three predictors of sepsis-associated acute kidney injury in this pediatric population. Dependence plots generated by the analysis elucidated the nonlinear relationships between these variables and kidney injury risk, showing how risk changes across different values of each parameter rather than assuming a simple linear pattern.

The clinical logic behind these predictors is plausible. Metabolic acidosis reflected in base excess, elevated liver enzymes suggesting multi-organ stress, and disturbed potassium handling are all consistent with deteriorating kidney function, though the study’s SHAP findings describe statistical associations within the modeled data rather than proven causal mechanisms. What the interpretability analysis does provide is a window into why the model flags certain children as high risk, which could help clinicians sanity-check its outputs against their own judgment.

The authors are careful about the limits of their work. The study was retrospective, meaning it analyzed existing records rather than prospectively monitoring patients, and all validation took place in Chinese centers, raising questions about how the model would transfer to populations with different demographics, care patterns, or sepsis etiologies. The conclusion states that prospective implementation and impact studies are warranted before clinical deployment, and that the tool holds promise for guiding early kidney-protective strategies rather than demonstrating that it already does so. For now, the study establishes that an explainable model built from routine pediatric intensive care data can discriminate kidney injury risk across multiple centers more accurately than traditional scores in most tested settings, addressing what the researchers describe as a critical gap in pediatric sepsis-associated acute kidney injury prediction.

Subject of Research: Development and validation of an interpretable machine learning model for early prediction of sepsis-associated acute kidney injury in pediatric sepsis patients

Article Title: Development and validation of a machine learning prediction model for pediatric sepsis-associated acute kidney injury: a multicenter cohort study

Article References: Huo, J., Wu, Y., Zhu, L., Wang, Y., Zhou, L., Tao, Y., Shi, C., Xiao, L., Xu, X., Li, J., Tan, L., & Li, S. (2026). Development and validation of a machine learning prediction model for pediatric sepsis-associated acute kidney injury: a multicenter cohort study. BioData Mining. https://doi.org/10.1186/s13040-026-00617-2

Image Credits: AI Generated

DOI: 10.1186/s13040-026-00617-2

Keywords: pediatric sepsis, acute kidney injury, machine learning, XGBoost, prediction model, SHAP interpretability, multicenter cohort study, pSOFA, Phoenix score, pediatric intensive care, clinical decision support, retrospective study

News Source: Teresa Odom. (November 6, 2026). Machine learning model predicts kidney injury in children with sepsis. Scienmag.

Tags: acute kidney injuryClinical decision supportMachine LearningMulticenter Cohort Studypediatric intensive carepediatric sepsisPhoenix scoreprediction modelpSOFARetrospective studySHAP interpretabilityXGBoost
Share12Tweet7Share2ShareShareShare1

Related Posts

Charging Infrastructure Drives Electric Vehicle Adoption in China’s Post-Subsidy Era

Charging Infrastructure Drives Electric Vehicle Adoption in China’s Post-Subsidy Era

November 20, 2026
Microbial Cell-Free DNA Sequencing Detects Overlooked Pathogens in Critically Ill Children

Microbial Cell-Free DNA Sequencing Detects Overlooked Pathogens in Critically Ill Children

November 20, 2026

Fire Mosaics Differentially Shape Genetic Connectivity in Australian Woodland Bats

November 20, 2026

Letter Identifies Methodological Limitations in Study Evaluating MX3 Hydration Testing System

November 20, 2026

POPULAR NEWS

  • Narrative Review Weighs Targeted Nerve Blocks for Geriatric Hip Fractures

    Narrative Review Weighs Targeted Nerve Blocks for Geriatric Hip Fractures

    29 shares
    Share 12 Tweet 7
  • Machine Learning Model Predicts In-Hospital Mortality for Cirrhosis Patients with Sepsis

    29 shares
    Share 12 Tweet 7
  • Semi-folded arenas reveal spontaneous termite trail emergence in whole colonies

    29 shares
    Share 12 Tweet 7
  • French Survey Reveals Mixed Perspectives on Operating Room Black Box Technology

    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

Charging Infrastructure Drives Electric Vehicle Adoption in China’s Post-Subsidy Era

Microbial Cell-Free DNA Sequencing Detects Overlooked Pathogens in Critically Ill Children

Meeting Postpartum 24-Hour Movement Guidelines Linked to Lower Depression and Stress

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.