Authors :
Presenting Author: Mark Abboud, BA – Case Western Reserve University
Terence Shaw II, BS – Baylor College of Medicine
Saifina Karedia, BS – Baylor College of Medicine
Nandani Adhyapak, BS – Baylor College of Medicine
Vaishnav Krishnan, MD, PhD – Baylor College of Medicine
Rationale:
“Dizziness” can be variably applied to report lightheadedness, feeling faint, imbalanced or frank vertigo. “Sleepiness” (or somnolence/sedation) is a self-recognition of an increased propensity to fall asleep, particularly during sedentary situations. Together, dizziness and sleepiness/somnolence are amongst the most commonly reported adverse effects associated with antiseizure medication (ASM) intake. In this study, we explored whether subjectively rated dizziness and/or sleepiness symptoms may be encoded in patterns of rest and activity recorded by continuously worn wrist accelerometers. Methods:
All research protocols were approved by the Baylor College of Medicine IRB. For this pilot study, we recruited 7 patients with drug-refractory epilepsy (4/7 female, aged 22 to 65, 1/7 with genetic generalized epilepsy). Actigraphy data was collected remotely via Centerpoint Insight watches (Actigraph). To obtain serial and longitudinal measurements of subjective dizziness or sleepiness, subjects were asked to respond to a daily 7-point Likert style survey administered through a smartphone-based app (ecological momentary assessment, EMA). Subjects were instructed to wear the watch and answer survey questions for a total of 1 year and were not penalized for skipped survey responses. From actigraphy counts derived every minute of the day, we calculated a range of metrics in serially overlapping 14-day windows, including those pertaining to RAR stability and fragmentation (interdaily stability, intradaily variability), timing (acrophase), active/rest ratios (“alpha”), steepness/squareness (“beta”), midpoint activity levels (mesor) and total daily sleep time. For each subject, we applied multiple linear regression to define explainable models to connect RAR metrics and normalized subjective dizziness and sleepiness scores. Wear compliance, measured as inferred non-wear times, were also factored into our predictive models.
Results:
7 patients completed the study. Watch non-wearing rates varied between 0.04 to 4.43h/d, while survey compliance rates varied from ~15 to 92% (58 to 339 responses over the year). Model performance among the 14 models generated (dizziness and sleepiness x 7 subjects) was modest, with root mean square errors (RMSEs) between 0.59 to 1.3. As expected, RMSEs were inversely related to the range of their responses (between 1-7), training set size and average nonwear times. Importantly, RAR metrics that predicted dizziness and/or sleepiness were markedly patient-specific. Conclusions:
From our data, we conclude that wrist-actigraphy derived RAR metrics may encode subjective variations in dizziness and sleepiness in some patients. However, a single generalizable model for all patients is likely to be considerably less accurate than patient-specific models. Therefore, predictive models to infer subjective dizziness and sleepiness from continuously worn wrist accelerometers may have the greatest value when manually and systematically calibrated first by motivated individual subjects. Funding: VK thanks the Mike Hogg Fund and acknowledges NIH support (R01NS131399).