Background: Inflammatory bowel disease (IBD) flares occur frequently and are typically assessed by biomarkers in the blood or stool or by imaging or endoscopy after a patient presents with symptoms. Wearable devices may permit continuous monitoring for flares by longitudinally assessing physiologic metrics. We sought to develop wearable device signatures to identify and predict IBD flares.

Methods: The IBD Forecast Study enrolled individuals in the US ≥18 years of age with IBD who were willing to use an Apple Watch, Fitbit or Oura Ring. Subjects downloaded our ehive app, by which patients also provided consent, verified inclusion criteria, and had symptoms assessed over 18 months. Symptoms were assessed with daily Patient Reported Outcome-2 surveys (asymptomatic; PRO-2 CD <8; PRO-2 UC ≤ 1 with rectal bleeding score=0, stool frequency score ≤ 1). A clinical flare was defined as ≥2 symptomatic days in a 7-day period. Inflammatory flares were defined as a C-reactive protein >5mg/L or a fecal calprotectin >150μg/g. Heart rate variability (HRV) from Apple Watches were represented in the time domain SDNN (standard deviation of N-N intervals) and analyzed using a mixed effect cosinor model bootstrapped 500 times. Daily steps, mean daily heart rate (HR) and resting heart rate (RHR) were compared using linear mixed effects models.

Results: 348 participants were recruited across 38 states in the US (Table 1). Mean HR and RHR were higher, and steps lower, during inflammatory flares (HR 79.98±0.91; RHR 74.47±0.85; 2,716±96.71 steps) compared to periods with normal inflammatory markers (HR 76.79±0.90, p<0.0001; RHR 61.92±0.84, p<0.0001; 3,063±74.19 steps, p<0.0001). The MESOR, or midline of the circadian pattern of the SDNN, was lower (38.73; 95%CI 35.85-41.40) during an inflammatory flare compared to periods with normal inflammatory markers (41.02; 95%CI 38.38-43.62; p<0.0001). Mean HR and RHR were higher, and steps lower, during clinical flares (HR 77.99±0.58; RHR 64.87±0.54; 2,050±253.93 steps) compared to periods in clinical remission (HR 77.72±0.57, p<0.0001; RHR 64.70±0.53, p<0.01; 2,083±253.83 steps, p<0.001). The amplitude, or height, of the circadian pattern of SDNN was higher (4.41; 95%CI 3.68-5.16) during clinical flares compared to periods in clinical remission (3.79; 95%CI 3.02-4.47; p=0.004). The MESOR of SDNN was lower (36.41; 95%CI 33.53-39.77) during clinical flares with inflammation compared to clinical flares with no inflammation (40.61; 95%CI 37.90-43.75; p<0.0001). The MESOR was lower during periods with no symptoms but with inflammation (38.87; 95%CI 36.00-41.86) compared to periods with no symptoms and no inflammation (40.37; 95%CI 37.63-43.46; P=0.03) (Figure 1).

Conclusions: Wearable device metrics identify clinical and inflammatory flares and identify the presence or absence of inflammation in symptomatic and asymptomatic individuals.

Table 1. Baseline demographic information.

Figure 1. Relationship between HRV circadian rhythm and IBD activity. (A) Daily HRV rhythm during inflammatory flares compared to periods with no inflammation. (B) Daily HRV circadian rhythm during periods of symptoms and inflammation compared to periods of symptoms without underlying inflammation. (C) Daily HRV rhythm during periods in symptomatic remission when inflammation is present compared to periods of symptomatic remission but when inflammation is absent.

Girish Nadkarni

Appearances