Machine Learning Algorithms for Sepsis Detection

SUMMARY:

  • Early identification and treatment of sepsis is central for identification and successful treatment of sepsis.

  • Widespread systematic sepsis screening (manual or automated) is the recommended approach to detecting early sepsis.

  • To date, employing systematic screening with traditional sepsis scoring tools has been suboptimal.

  • Machine learning algorithms for sepsis screening offer a promising improvement in early sepsis detection and identification.

REVIEW:

  • Arecent systematic review and network meta-analysis compared machine learning techniques to standard sepsis screening tools.

    • 73 articles with 457,932 patients were reviewed, published from 2017-2023

    • Area Under the Curve of the Receiver Operating Characteristic (AUROC) was used as the performance metric, with 95% confidence interval.

  • Standard screening tools assessed were:

    • Sequential Organ Failure Assessment (SOFA)

    • Quick Sequential Organ Failure Assessment (qSOFA)

    • National Early Warning Score/National Early Warning Score 2 (NEWS/NEWS2)

    • Modified early Warning Score (MEWS)

    • Simplified Acute Physiology Score (SAPS II)

    • Systemic Inflammatory Response Syndrome (SIRS)

  • Machine Learning Models assessed were:

    • Neural Network Models – 6 different models

    • Decision Trees – 4 different models

    • Regression Models – 3 models

    • Support Vector Machine

    • K-Nearest Neighbors

    • Generalized Linear Model

    • Naïve Bayes

    • All Machine Learning algorithms were consistently statistically (P<0.0001) superior to traditional screening tools.
      • Neural Network and Decision Tree models had the highest AUROC metrics.
  • Other factors including sepsis prevalence, laboratory indicators and number of predictors did not influence the models.

CONCLUSIONS:

  • An ongoing concern is the “Black Box Syndrome” associated with machine learning algorithms, limiting the clinicians ability to completely view and understand the logic behind the outputs.

  • Although machine learning algorithms can be incorporated into clinical practice, , it is not known how these predictions will impact:

    • Timing between notification and clinical recognition of sepsis

    • Timing, initiation of treatments

    • Clinical outcomes

To receive articles like these in your Inbox, you can subscribe to Sepsis Program Optimization Insights.

Subscribe to sepsis program optimization insights

Erkan Hassan is the Co-Founder & Chief Clinical Officer of Sepsis Program Optimization where he designs & oversees the implementation of solutions to optimize sepsis programs.

To discuss your organization’s Barriers of Effective Sepsis Care, contact Erkan by phone (844) 4SEPSIS (844-473-7747), email (erkan@spo.icu), or video chat.