Prediction of chronic damage in systemic lupus erythematosus by using machine-learning models

Area of Science:

• Rheumatology
• Artificial Intelligence
• Medical Informatics

Background:

• Systemic Lupus Erythematosus (SLE) survival has increased, leading to a higher incidence of chronic damage in up to 50% of patients.
• Preventing chronic damage is a critical objective in managing SLE.
• Predictive models can aid in early intervention and improved patient outcomes.

Purpose of the Study:

• To develop and validate a predictive model for chronic damage in SLE patients using neural networks.
• To identify patients at high risk of developing chronic damage for targeted prevention strategies.
• To leverage longitudinal clinical data for accurate damage prediction.

Main Methods:

• Recurrent Neural Networks (RNNs) were employed as a machine-learning model.
• Longitudinal clinical data sequences from 413 SLE patients were utilized.
• The SLICC/ACR Damage Index (SDI) was used to assess chronic damage.

Main Results:

• The RNN model achieved an Area Under the Curve (AUC) of 0.77 for predicting damage development.
• A threshold of 0.35 demonstrated a sensitivity of 0.74 and specificity of 0.76 in identifying at-risk patients.
• 35.8% of patients presented with an SDI of 1 or higher at their initial visit.

Conclusions:

• Recurrent Neural Networks (RNNs) can effectively predict the risk of chronic damage in SLE patients.
• Utilizing longitudinal data from the Sapienza Lupus Cohort enabled the construction of a robust predictive model.
• This approach has the potential to identify patients requiring closer monitoring and early intervention to prevent damage progression.