Prediction of 1-Year Activity in Systemic Lupus Erythematosus: Hierarchical Machine Learning Approach.

Area of Science:

• Artificial Intelligence in Medicine
• Machine Learning for Healthcare
• Predictive Analytics in Rheumatology

Background:

• Systemic lupus erythematosus (SLE) is a complex chronic disease with unpredictable activity affecting multiple organs.
• Predicting SLE activity events is challenging due to individual variability and temporal disease patterns.
• Current management relies on monitoring, highlighting the need for predictive tools.

Purpose of the Study:

• To develop and validate a hierarchical machine learning model for predicting 12-month SLE activity.
• To define SLE activity as hospitalization, new organ involvement, or specific manifestations.
• To identify key predictive features for enhanced clinical decision-making.

Main Methods:

• A hierarchical model combining a random forest and decision tree was developed.
• The model utilized longitudinal data from 262 SLE patients (2012-2020).
• Features included demographics, clinical history, lab results, and treatments over different timeframes.

Main Results:

• The hierarchical model achieved an AUC of 0.743, outperforming the initial model (AUC 0.696).
• Explainable AI identified 15 key features influencing predictions, such as age and therapy response.
• The model demonstrated improved performance for specific patient subgroups.

Conclusions:

• An explainable and reliable AI tool for 1-year SLE activity prediction is introduced.
• The model serves as a decision-support system to enhance patient management and personalize treatment.
• The methodology is adaptable for predicting outcomes in other chronic autoimmune diseases.