Using machine learning algorithms based on patient admission laboratory parameters to predict adverse outcomes in...

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Clinical Prognostics

Background:

• The COVID-19 pandemic highlighted the need for rapid patient prognosis.
• Accurate survival assessment is crucial for timely clinical decision-making.
• Existing prognostic tools may lack speed or rely on complex data.

Purpose of the Study:

• To develop a machine learning model for predicting patient survival using readily available clinical laboratory data.
• To create a fast and accurate prognostic assessment tool for clinicians.
• To identify key laboratory parameters predictive of patient outcomes.

Main Methods:

• Utilized routine clinical laboratory test data for model development.
• Integrated feature selection and binary classification algorithms.
• Employed a combined Lasso and Support Vector Machine (SVM) methodology.
• Optimized algorithm selection through parameter control.

Main Results:

• Developed a predictive model using 8 clinical laboratory parameters.
• Achieved an area under the ROC curve (AUC) of 0.9277.
• Demonstrated the efficacy of using basic laboratory data for prognostication.

Conclusions:

• Machine learning models can effectively predict patient survival using simple laboratory tests.
• The developed model provides clinicians with an expeditious and precise prognostic tool.
• This approach simplifies data processing and enhances clinical decision support.