An Efficient Machine Learning Pipeline for Distinguishing Cancer and Non-Cancer Patients in Systemic Lupus...

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Clinical Decision Support Systems

Background:

• Systemic lupus erythematosus (SLE) patients have a complex disease profile.
• Distinguishing cancer in SLE patients is clinically significant.
• Developing accurate predictive models for cancer in SLE is a critical unmet need.

Purpose of the Study:

• To develop and validate a machine learning (ML) model for differentiating cancer from non-cancer patients within a cohort of SLE patients.
• To identify key clinical and laboratory features predictive of cancer in SLE.
• To establish a practical tool for aiding in cancer diagnosis among SLE individuals.

Main Methods:

• Utilized a dataset of 2811 SLE patients, including 208 with concurrent cancers.
• Performed data preprocessing, feature transformation, and integrated feature selection.
• Developed and optimized various ML models, including logistic regression, for predictive analysis.

Main Results:

• Integrated feature selection identified Age, C4, WBC (urine), BASO_R, ALP, AST, and C3 as significant predictors.
• Logistic regression emerged as the optimal ML model after parameter tuning.
• The logistic regression model demonstrated superior predictive performance, validated by ROC curves, PRC plots, accuracy, F1 score, and recall.

Conclusions:

• A straightforward and effective machine learning model was successfully developed for cancer detection in SLE patients.
• The model provides a convenient and useful method for distinguishing cancer from non-cancer status within the SLE population.
• This tool has the potential to improve early cancer detection and management in individuals with SLE.