Digital Biomarkers

Wearable-based activity recognition algorithms

Background

Crohn’s is an immune-mediated condition that causes significant life changes. Symptoms fluctuate in a remission-relapse cycle and include frequent and irregular toilet use, urgency, pain, and shorter meals. In addition, Crohn’s patients adapt to their new standards by limiting social interactions and their work/school schedules.  

Approach

Circadic built algorithms that can run on consumer smartwatches, and can identify individual Crohn’s symptoms.

Under the hood

Circadic ran pilot studies in which participants were provided with smartwatches and an iPhone-based activity annotation app. The data was used to train activity recognition algorithms that subsequently were able to detect meals, toilet use, and pain/urgency with an AUC>0.85. In addition, the detected symptoms were associated with the Physician Global Assessment (PGA) score of each patient using a second ML algorithm that could detect flare ups with AUC>0.9, by identifying anomalies in the daily behaviorgram (a time series index of disease-related activities).