Sepsis Prediction for the General Ward Setting.

Area of Science:

• Medical Informatics
• Clinical Prediction Models
• Machine Learning in Healthcare

Background:

• Sepsis prediction in general hospital wards remains a challenge.
• Early detection is crucial for improving patient outcomes.
• Existing prediction tools may have limitations in real-world clinical settings.

Purpose of the Study:

• To develop and evaluate a machine learning model for sepsis prediction in general wards.
• To assess the model's performance using a pseudo-prospective trial design.
• To provide a realistic estimation of the model's clinical implementation performance.

Main Methods:

• Retrospective analysis of electronic health records (EHR) from adult, non-surgical inpatients.
• Development of a machine learning model to predict sepsis onset 6 hours in advance.
• Evaluation using Area Under the Receiver Operating Characteristic curve (AUROC) and Area Under the Precision-Recall Curve (AUPRC).
• A pseudo-prospective trial was conducted to simulate real-world performance.

Main Results:

• The developed machine learning model achieved an AUROC of 0.862 ± 0.011 and AUPRC of 0.294 ± 0.021.
• The model outperformed Logistic Regression and NEWS 2 scoring systems.
• In the pseudo-prospective trial, 69.7% of septic patients were alerted with 81.4% specificity.
• Within 24 hours of alert, 20.9% experienced a sepsis-related event.

Conclusions:

• A machine learning model effectively predicts sepsis in general wards using EHR data.
• The pseudo-prospective trial demonstrated a Positive Predictive Value (PPV) of 29.1% for sepsis-related outcomes within 48 hours.
• The model offers a promising tool for early sepsis detection and intervention in clinical practice.