Development and validation of machine learning-based prediction model for severe pneumonia: A multicenter cohort...

Area of Science:

• Medical research
• Computational biology
• Respiratory medicine

Background:

• Severe pneumonia (SP) presents high mortality and limited early prediction.
• Existing scoring systems are time-consuming and lack early predictive power.

Purpose of the Study:

• To develop a machine learning model for early SP prediction.
• Utilize peripheral blood inflammatory markers for risk assessment.

Main Methods:

• Collected clinical and laboratory data from 204 pneumonia patients.
• Developed and evaluated multiple machine learning models (XGBoost, RF, etc.).
• Selected key predictors (age, WBC, CRP, etc.) using LASSO regression and clinical insight.

Main Results:

• The XGBoost model achieved an AUC of 0.901 in the test cohort.
• Key predictors included elevated White Blood Cell (WBC) count, older age, and elevated C-reactive protein (CRP).
• The model demonstrated high accuracy (0.803) and sensitivity (0.844) for SP prediction.

Conclusions:

• Machine learning models using inflammatory biomarkers can rapidly assess SP risk.
• This approach facilitates timely preventive interventions for severe pneumonia.
• The developed model shows significant potential for clinical application in early SP detection.