Machine Learning Approaches to Classify Self-Reported Rheumatoid Arthritis Health Scores Using Activity Tracker Data:...

Area of Science:

• Digital health
• Machine learning in healthcare
• Rheumatoid arthritis patient monitoring

Background:

• Activity trackers passively collect physical data in mobile health studies.
• This reduces the burden of actively contributing patient-reported outcome (PRO) information.

Purpose of the Study:

• Develop machine learning models to classify and predict PRO scores.
• Utilize Fitbit data from rheumatoid arthritis patients.

Main Methods:

• Developed two models: a random forest classifier and a hidden Markov model.
• Compared model performance on binary (normal vs. severe PRO) and multiclass PRO state classification tasks.

Main Results:

• The hidden Markov model significantly outperformed the random forest model for both binary and multiclass tasks (P<.05).
• Highest performance metrics included an area under the curve of 0.750, Pearson correlation coefficient of 0.479, and Cohen κ coefficient of 0.471.

Conclusions:

• Physical activity tracker data can classify rheumatoid arthritis patient health status over time.
• This enables timely preventive clinical interventions and holds potential for improving care in other chronic conditions.