Comparing two machine learning approaches in predicting lupus hospitalization using longitudinal data.

Area of Science:

• * Computational medicine and health informatics.
• * Application of artificial intelligence in rheumatology.

Background:

• * Systemic lupus erythematosus (SLE) is a complex autoimmune disease with unpredictable flares.
• * Hospitalizations represent a significant cost driver in SLE management.
• * Predicting SLE hospitalizations is crucial for proactive patient care and resource allocation.

Purpose of the Study:

• * To evaluate two machine learning approaches for predicting SLE hospitalizations using electronic health record (EHR) data.
• * To compare the performance and suitability of a Differential approach versus a Long Short-Term Memory (LSTM) model.
• * To determine optimal patient monitoring periods for different prediction horizons.

Main Methods:

• * Utilized longitudinal EHR data from 925 patients in a multicenter lupus cohort.
• * Implemented a Differential approach incorporating lagged variables to capture time dependencies.
• * Evaluated a Long Short-Term Memory (LSTM) deep learning model for time-series prediction.

Main Results:

• * Both Differential and LSTM models demonstrated effectiveness in predicting SLE hospitalizations.
• * The Differential approach is advantageous for short observation periods and offers flexibility in handling class imbalance.
• * LSTM excels with long observation intervals, capturing long-term dependencies, but requires more balanced training data.

Conclusions:

• * Machine learning models offer viable strategies for predicting SLE hospitalizations.
• * The choice between Differential and LSTM depends on data characteristics (observation length) and availability of balanced datasets.
• * Findings inform optimal patient monitoring strategies for improved SLE management.