Systemic lupus erythematosus with high disease activity identification based on machine learning.

Area of Science:

• Medical informatics
• Machine learning in healthcare
• Rheumatology

Background:

• Clinical evaluation of systemic lupus erythematosus (SLE) disease activity is challenging and inconsistent.
• High disease activity in SLE significantly impacts patient outcomes.

Purpose of the Study:

• To develop a machine learning model for identifying SLE patients with high disease activity.
• To improve the accuracy and consistency of SLE disease activity assessment.

Main Methods:

• Utilized data from 1014 low-activity and 453 high-activity SLE patients.
• Collected 94 clinical, laboratory, and 17 meteorological indicators.
• Employed mutual information and multisurf for feature selection, followed by machine learning modeling (LGB, Naive Bayes).

Main Results:

• Integrated feature selection identified key indicators: hematuria, proteinuria, pyuria, low complement, precipitation, and sunlight.
• The LGB model demonstrated superior performance with an ROC AUC of 0.930 and PRC AUC of 0.911.
• Feature selection consistency was confirmed in the composite feature importance bar plot.

Conclusions:

• A straightforward machine learning pipeline effectively identifies SLE patients with high disease activity.
• The LGB model, leveraging selected features like proteinuria and hematuria, shows significant promise for clinical application.