An interpretable machine learning model for biomarker identification and diagnostic nomogram development in...

Area of Science:

• Hematology
• Machine Learning in Medicine
• Biomarker Discovery

Background:

• Thrombocytopenia is a common hematological disorder with varied causes.
• Distinguishing hypo-productive from hyper-destructive thrombocytopenia is crucial for timely treatment.
• Current diagnostic methods may lack efficiency for early-stage hypo-productive thrombocytopenia.

Purpose of the Study:

• To identify reliable biomarkers for early recognition of hypo-productive thrombocytopenia.
• To develop an interpretable diagnostic nomogram for hypo-productive thrombocytopenia.
• To leverage machine learning for improved diagnostic accuracy in hematological disorders.

Main Methods:

• Retrospective analysis of 185 thrombocytopenia patients.
• Feature selection using LASSO, SVM-RFE, and Boruta algorithms.
• Development and comparison of six machine learning models, with interpretation via SHAP analysis.

Main Results:

• Age, platelet-to-lymphocyte ratio (PLR), neutrophil-to-lymphocyte ratio (NLR), and mean platelet volume (MPV) identified as key predictors.
• A generalized linear model (GLM)-based nomogram showed high discriminative ability (AUC = 0.945).
• The nomogram demonstrated excellent calibration and superior net benefit compared to individual predictors via Decision Curve Analysis (DCA).

Conclusions:

• An interpretable machine learning-derived nomogram can aid in the early identification of hypo-productive thrombocytopenia.
• The nomogram serves as a practical, non-invasive tool for clinical decision support.
• Biomarkers like PLR, NLR, and MPV are valuable in diagnosing thrombocytopenia subtypes.