Classifying Self-Reported Rheumatoid Arthritis Flares Using Daily Patient-Generated Data From a Smartphone App:...

Area of Science:

• Rheumatology
• Artificial Intelligence
• Digital Health

Background:

• Predicting rheumatoid arthritis (RA) flares between clinic visits is crucial for timely interventions.
• Real-time, patient-generated data can potentially prevent disease worsening.

Purpose of the Study:

• Investigate the feasibility of using machine learning (ML) to classify self-reported RA flares.
• Utilize daily symptom data collected via a smartphone app for flare prediction.

Main Methods:

• Collected daily symptom and weekly flare data from 20 RA patients over 3 months using the REMORA app.
• Engineered summary features of daily symptoms (pain, fatigue) from the preceding week.
• Trained and evaluated three binary classifiers: logistic regression (with/without elastic net), random forest, and naive Bayes, using AUC.

Main Results:

• Logistic regression with elastic net regularization achieved the highest AUC of 0.82.
• At 80% specificity, sensitivity was 60%, with a positive predictive value (PPV) of 53% and negative predictive value (NPV) of 85%.
• Model performance varied: prioritizing NPV (0.92) decreased PPV (0.51), while prioritizing PPV (0.65) resulted in lower NPV.

Conclusions:

• Machine learning prediction of RA flares from daily symptom data is feasible.
• Further validation with larger datasets and external cohorts is necessary.
• Future applications may enable just-in-time adaptive interventions for RA management.