Development and internal validation of an interpretable machine learning model using routine laboratory data for...

Area of Science:

• Biomedical Informatics
• Machine Learning in Healthcare
• Pulmonary Hypertension Research

Background:

• Systemic sclerosis-associated pulmonary arterial hypertension (SSc-PAH) requires early diagnosis for effective management.
• Optimizing diagnostic resource utilization in SSc-PAH is crucial for patient outcomes and healthcare efficiency.

Purpose of the Study:

• To develop and validate an interpretable machine learning (ML) model for predicting SSc-PAH risk using routine laboratory data.
• To support clinical decision-making and personalize care for SSc-PAH patients.

Main Methods:

• Extracted 39 laboratory indicators and demographic variables from EHRs of SSc patients.
• Developed and compared seven ML algorithms, selecting the top five features using LASSO.
• Utilized AdaBoost classifier and SHAP values for model interpretability and feature importance analysis.

Main Results:

• The interpretable AdaBoost model achieved high discrimination in training (AUC=0.892) and test sets (AUC=0.968).
• Decision Curve Analysis demonstrated superior net clinical benefit compared to standard strategies.
• Key predictors identified included NT-proBNP, rT3, LDH, HIV Ag Ab, Ca, and CK-MB.

Conclusions:

• The developed AdaBoost model is a practical tool for early SSc-PAH screening and diagnosis.
• This approach can reduce patient disease burden, enhance quality of life, and optimize medical resource allocation.