Diagnostic performance of machine learning models using cell population data for the detection of sepsis: a...

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Sepsis Diagnostics

Background:

• Sepsis diagnosis relies on timely evaluation of patient data.
• Machine learning (ML) and artificial intelligence (AI) offer advanced analytical capabilities.
• Routine blood tests are crucial for initial patient assessment.

Purpose of the Study:

• To compare the efficacy of AI/ML algorithms against traditional logistic regression for sepsis evaluation.
• To assess the performance of various ML models using routinely collected blood test data.
• To identify the most effective AI/ML approach for early sepsis detection.

Main Methods:

• Development and evaluation of ML/AI models using patient data from Emergency Department admissions.
• Utilized complete blood counts (CBC) and Cell Population Data (CPD).
• Performance metrics included Area Under the Receiver Operating Curve (AUC), calibration plots, and Decision Curve Analysis (DCA).

Main Results:

• All tested methods achieved an AUC greater than 0.90.
• Logistic regression (LR) showed comparable performance to some ML/AI models.
• The multi-layer perceptron (MLP) model excelled in discrimination, calibration, and clinical utility.
• Naïve Bayes and K-nearest neighbor (KNN) models lacked good calibration.

Conclusions:

• AI and ML models, particularly MLP, offer superior performance for early sepsis detection.
• These advanced models can significantly aid in evaluating patients with suspected sepsis.
• Further external validation is recommended to refine and update these diagnostic models.