Innovating Drug Development through Advanced Statistical and AI Methods for Digital Wearable Sensor Data

Narcolepsy type 1 (NT1) is a neurological disorder caused by orexin deficiency and is characterized by excessive daytime sleepiness, cataplexy, hallucinations, sleep paralysis and disrupted nocturnal sleep. In-lab polysomnogaphy (PSG) and multiple sleep latency tests (MSLT) are recommended for assessments of pathological sleep for NT1 diagnosis, requiring substantial patient time and trained personnel costs. The Dreem 3S (D3S) headband is an FDA-cleared device for ambulatory EEG-based sleep staging and is operable by patients at home. The ongoing D3KAH study (NCT06531876) evaluates D3S utility in subjects with central disorders of hypersomnolence, including NT1. An interim analysis (IA) of 15 NT1 completers (as of 30 January 2025) assesed data from at-home D3S monitoring over 6 nights and a 24-hour period, and 2 nights of concurrent in-lab PSG and D3S monitoring. We present statistical methods to assess wear adherence, sleep staging performance, and explore the potential utility of repeated sleep assessment at home. All primary endpoints met pre-defined success criteria at IA. Primary, secondary, and exploratory endpoints at full readout will be presented. 

As the global population ages, the need for effective treatment strategies for chronic diseases, particularly cardiovascular conditions, becomes increasingly critical. This presentation explores the innovative integration of wearable technology and biosensor data into clinical trials, focusing on their potential to enhance patient monitoring and outcome prediction. We discuss a series of analytical approaches applied to accelerometry data collected from wearable devices alongside the assessment of clinical outcomes. These methodologies enable the identification of unique clinical phenotypes within patient groups and improve predictive accuracy for clinical outcomes in heart failure. By leveraging high-resolution data from wearables, researchers can gain deeper insights into patients' physical activity patterns, sedentary behavior, and sleep, facilitating a more nuanced understanding of their relationship with clinical outcomes. Ultimately, the findings from this work underscore the importance of embracing novel technologies and advanced statistical methods to identify heterogeneous patient subgroups, thereby informing and guiding clinical development programs for improved patient care. 

Postural Orthostatic Tachycardia Syndrome (POTS) is a disorder of orthostatic intolerance (OI) characterized by an excessive heart rate increase when transitioning from supine to upright, accompanied by symptoms such as lightheadedness. Orthostatic characteristics measured in-clinic and patient-reported data on symptoms and negative impacts on quality of life have been published, but minimal objective data collected in free-living settings are available. Wearable devices that continuously collect ambulatory electrocardiography (ECG) and accelerometry may improve characterization of POTS by capturing physical activity and heart rate characteristics. We further hypothesize that joint modeling of ECG and accelerometry data can capture free-living orthostatic characteristics. We will discuss how accelerometry data were used to estimate objective characteristics of free-living physical activity in POTS patients and healthy controls using arctools, including total amount per day and intensity. In addition, we will cover a novel algorithm that uses ECG modelled jointly with accelerometry to dynamically characterize heart rate during postural transitions. We hope that this novel approach will allow for better characterization of this disorder enabling more efficient clinical trial designs and improved patient care. These results complement published patient survey data on the significant negative impact of POTS on quality of life.  

There has been much interest but limited movement towards incorporating circadian timing into healthcare. An enduring reason cited is the challenge of tracking true circadian time at scale without the need for invasive, expensive tests. In this presentation, I will discuss gold standard biomarkers of circadian rhythms and the extent to which they can be estimated from consumer wearable devices or phones, as well as the natural patterns of circadian variation in society. I will show simulations that highlight how this variation, when coupled with the known time-of-day effects of approved drugs, can blunt the observed effect size of an intervention. I will also demonstrate the ways in which estimates of both circadian phase and amplitude can be used to clarify trial results, as well as examples from the literature where circadian phase and amplitude are hypothesized to have played a role.  

Excessive daytime sleepiness (EDS) is a hallmark symptom of narcolepsy type 1 (NT1). The Maintenance of Wakefulness Test (MWT) is used in clinical trials to assess treatment effects on EDS by measuring the ability to stay awake until persistent sleep occurs. Sleep onset latency, the primary MWT endpoint, does not capture the quality of this wakefulness. We explored alternative EDS endpoints using manually scored microsleeps (3-5 sec sleep episodes) and hypnodensity-based sleepiness scores from MWT data in a phase 2 randomized, placebo-controlled trial of oveporexton (TAK-861), an oral orexin receptor 2 agonist (NCT05687903). Participants with NT1 (n=112) were randomized to oveporexton (4 doses) or placebo. Four 40-minute MWTs were performed at baseline, Day 28, and Day 56. Generalized mixed-effects models analyzed microsleep rates and average sleepiness scores. Among participants with NT1, oveporexton significantly reduced microsleep rates by Day 28 and 56 (from ~6 to <2 microsleeps/10 min) and reduced sleepiness scores (p<0.005 vs baseline). Placebo-treated participants showed minimal changes. These findings demonstrate oveporexton's potential for treating EDS in people with NT1.