Diagnostic and comparative performance for the prediction of tuberculous pleural effusion using machine learning...

Area of Science:

• Medical diagnostics
• Machine learning in healthcare
• Pulmonology

Background:

• Early diagnosis of tuberculous pleural effusion (TPE) is crucial but challenging.
• Accurate differential diagnosis impacts patient prognosis and treatment efficacy.

Purpose of the Study:

• To develop and compare nine machine learning (ML) algorithms for early TPE diagnosis.
• To identify the optimal ML model for TPE discrimination.

Main Methods:

• Retrospective analysis of 1435 patients with pleural effusions (PEs).
• Feature selection and model development using nine ML algorithms.
• Validation of the optimal model using external data and SHAP analysis.

Main Results:

• Support Vector Machine (SVM) achieved the highest performance (AUC 0.914).
• External validation confirmed SVM's diagnostic accuracy (AUC 0.898).
• Key diagnostic features included PE adenosine deaminase (ADA), PE carcinoembryonic antigen (CEA), and serum CYFRA21-1.

Conclusions:

• A validated SVM model shows promise for early TPE diagnosis.
• This ML approach can support clinical decision-making for TPE management.