Digital Biomarkers最新文献

Introduction: Huntington's disease (HD) is a progressive neurodegenerative disorder characterized by motor, cognitive, and psychiatric decline. The Unified Huntington's Disease Rating Scale Total Motor Score (UHDRS-TMS) is standard for staging manifest disease, but is relatively insensitive to subtle premanifest changes. Speech abnormalities are emerging as candidate digital biomarkers; however, reliably separating premanifest HD (preHD) from healthy controls remains challenging. Here, we assess the feasibility of a speech-only approach by training and comparing multiple classifiers across diverse feature sets and structured tasks to determine whether speech alone can discriminate preHD from controls.

Methods: Speech samples were collected from 94 individuals with HD (38 premanifest, 56 manifest) and 36 controls using a standardized six-task protocol administered via tablet. From these recordings, 188 lexical and prosodic features were automatically extracted. We trained 4 machine learning classifiers: random forest, support vector machine, XGBoost, and deep neural networks (DNNs), within 10-fold cross-validation using three feature configurations: (1) all tasks (188 features), (2) the top 30 ANOVA-ranked features, and (3) 22 features from the Caterpillar passage alone.

Results: Traditional classifiers showed limited accuracy. A DNN using only the Caterpillar task achieved 81% unweighted accuracy for classifying preHD versus controls. Accuracy increased to 83% for prodromal HD and 87% when all HD participants were compared to controls. Adding features from additional tasks did not improve performance.

Conclusion: A brief, structured speech task combined with deep learning enabled accurate classification of preHD. These findings support speech analysis as a scalable, objective tool for early disease detection and monitoring.

Introduction: The identification of biomarkers for treatment response in major depression is critical to the further development of personalized treatment. There is a recognized relationship between facial expression and depression of mood, and previous literature also indicates that facial expression is associated with treatment outcomes in depression. This suggests that facial expression may have use as a biomarker for treatment response. There is no previous synthesis of related research to drive the development of new digital approaches.

Methods: We conducted a systematic review using three databases (MEDLINE, Scopus, and PsycINFO), identifying English-language publications (journal articles or books) that assessed either facial muscle activity or expression as predictors of treatment response or correlates of treatment outcome in depression. Risk of bias was assessed using a National Institutes of Health quality assessment tool.

Results: We identified 12 studies, involving a total of 389 participants, which used a variety of different assessment methods and thus assessment outcomes, including electromyography, observer-related assessments (including Facial Action Coding System), and automated tools of facial expression assessment. Depression treatment response correlated with an increase in facial expressivity. Greater activity in the corrugator and zygomatic muscles, and lower levels of lip tightening and downward lip movement, may predict treatment response.

Conclusions: Included studies were limited by heterogeneity in facial expression assessment tools and outcomes, along with demographic homogeneity. The findings of the review suggest that facial expression analysis may offer an avenue for biomarkers of depression status and treatment response prediction.

Introduction: Actigraphy-quantified physical activity (PA) allows for continuous measurements of PA that are reflective of real-world day-to-day functioning and morbidity in persons living with cardiomyopathy. This analysis reports the results of actigraphy monitoring and relates these to other clinical outcome assessments in the phase 3, multinational REALM-dilated cardiomyopathy (DCM) (NCT03439514) clinical trial in LMNA-related DCM.

Methods: Between 2020 and 2022, REALM-DCM randomized 37 patients with actigraphy worn on the nondominant wrist continuously to monitor daily PA. Of those, 35 participants had analyzable data for this analysis.

Results: The median duration of actigraphy monitoring for all participants was 293 days across 120 patient visits. Over 85% of the visits met a predefined threshold of wear-time compliance of 10 h of awake wear time for at least 4 days within the 2-week monitoring period prior to and after clinic visits. Kansas City Cardiomyopathy Questionnaire (KCCQ) physical limitation scores were positively associated with several actigraphy-quantified PA metrics, including moderate-to-vigorous physical activity (MVPA), moderate activity, non-sedentary behavior, total step counts, total activity counts (all 3 axes and their vector magnitude). Six-minute walk time distance was positively associated with time spent in MVPA and moderate activity, and total step counts. Patient Global Impression (PGI) Symptom Heart Failure Severity was negatively associated with non-sedentary behavior, total activity counts (vector magnitude, X- and Y-axes), and light activity. Actigraphy endpoints also distinguished between NYHA class II and class III patients. Actigraphy endpoints did not correlate with the KCCQ total score.

Conclusion: This is the largest and most longitudinal dataset of LMNA-DCM patients collected and reported to date using wearable sensors to gain understanding of PA patterns in these patients. These data help understand the potential use of actigraphy monitoring and wearable technologies in genetic cardiomyopathy and heart failure clinical trials.

Introduction: Eye movements are key biomarkers for diagnosing and monitoring neuro-otological, neuro-ophthalmological and neurodegenerative disorders. Video-oculography (VOG) systems enable detection of small, rapid eye movements and subtle oculomotor pathologies that may be missed during clinical exams. However, they rely on high-quality input, struggle with torsional movements, and are often limited by high costs in clinical and research settings.

Methods: To overcome these limitations, we developed 3DeepVOG, a deep learning-based framework for three-dimensional monocular gaze tracking (horizontal, vertical, and torsional rotation) that operates robustly across varied imaging conditions, including low-light and noisy environments. The method combines automated pupil and iris segmentation with geometrically interpretable estimation using a two-sphere anatomical eyeball model with corneal refraction correction. Torsion is tracked in real time using a novel mini-patch template matching approach. The system was trained on over 24,000 annotated samples obtained across multiple devices and clinical scenarios. Application was tested against a gold-standard VOG system in healthy controls.

Results: 3DeepVOG operates in real time (>300 fps) and achieves gaze errors of ∼0.1° in all three dimensions. Oculomotor measures - saccadic peak velocity, smooth pursuit gain, and optokinetic nystagmus slow-phase velocity - show good-to-excellent agreement with a clinical gold-standard system. As proof of concept, we present a case of acute unilateral vestibular failure where 3DeepVOG reliably captures 3D nystagmus.

Conclusions: 3DeepVOG enables accurate, quantitative eye movement tracking across three dimensions under diverse conditions. As an open-source framework, it provides an accessible and scalable tool for advancing research and clinical assessment in neurological oculomotor disorders.

Introduction: Sleep disturbances associated with menopause (SDM) are common and bothersome, but there are currently no specifically licensed treatments, and studies thus far have used different methodologies to measure sleep quality. Among those, digital health technologies (DHTs) present an innovative approach that supports patient-centric drug development by providing insights into how a patient responds to treatment in real-world settings. DHTs therefore may offer a solution to provide unobtrusive objective measurement of SDM. Here we describe the joint development of a novel DHT-derived endpoint for assessing sleep quality in menopausal women through a collaborative approach from evidence generation to analytical, clinical, and usability validation based on regulatory guidance.

Methods: To demonstrate the fit-for-purpose of the novel DHT-derived endpoint, Bayer (drug developer), Sleepiz AG (DHT provider), and DEEP Measures (collaboration platform provider) partnered and applied established frameworks to leverage prior work while compiling comprehensive data, conducting a gap analysis, and curating evidence in the DEEP Measures collaboration platform based on and in preparation for discussions with health authorities. Initial regulatory feedback from health authorities provided useful input and supported the study design on the incorporation of the DHT-derived endpoint into the clinical development program of elinzanetant. Through collaborative efforts between the drug developer and the DHT provider, the novel DHT-derived endpoint (Sleepiz One+ for continuous, home-based measurement of wake after sleep onset in SDM and other sleep parameters) was implemented as an exploratory endpoint in a phase 2 pilot study where data to demonstrate fit-for-purpose were generated and validated against polysomnography, the gold-standard objective measure for sleep. The study outcomes alongside the results of the gap analyses and leveraging prior work were then structured systematically in the DEEP Measures platform. Data were organized according to the DEEP Stack model (which included information on the measurement definition, target solution profile, and instrumentation), and these facilitated the integration of our outputs directly into the regulatory package used for following health authority interactions to drive the acceptance of the novel endpoint.

Conclusion: We outline how various stakeholders collaborated to leverage prior evidence, interacted with regulatory authority, and incorporated a novel DHT-derived endpoint into clinical development programs. Evidence and data generated in the present project have the potential to build the basis for further endpoint and DHT development and validation.

Introduction: Older men on androgen suppression for prostate cancer experience substantial symptom burden that is often missed between clinic visits. In prior work from our group, frequency-domain features ranked highly for predicting geriatric impairment, motivating a focus on interpretable spectral measures from open-source wrist accelerometry. Our overall objective was to identify accelerometry features from a pre-specified library that track weekly symptom burden in older men on ADT, and to characterize the temporal scale of the top candidates; spectral features were of particular interest.

Methods: A retrospective secondary analysis of an open-source pilot was performed. Ten men ≥65 years with metastatic prostate cancer completed weekly symptom burden and self-rated health over ∼100 days. Symptom-triggered (and random) 48-h, 10-Hz wrist-accelerometry sessions were aggregated to 60-s counts-per-minute (CPM) and vector-magnitude change. From these, 98 pre-specified statistical and spectral features were extracted. Associations with a weekly Symptom Burden + Self-Rated Health Index composite (SBSI) were assessed using linear mixed-effects models (days + random intercept), Spearman correlations across five 30-day bins, penalized mixed-effects regression least absolute shrinkage and selection operator (λ = 0.5, 1), and a 500-tree random forest.

Results: Nine participants provided 44 monitoring windows (14-48 h). In mixed-effects models, two CPM features were nominally associated with SBSI but did not survive false-discovery-rate adjustment. Across 30-day bins, a minute-scale restlessness pattern (CPM_top_15_freq3) rose with higher SBSI (ρ = +0.95; p = 0.012), while an overall rhythm balance measure (CPM_median_freq) tended to shift lower (ρ = -0.88; p = 0.049). Penalized models (λ = 1) retained both features, and random-forest importance ranked them highest. Within-participant plots showed restlessness increased during higher symptom weeks, while slower rhythm balance showed individual variability.

Conclusion: Two interpretable CPM spectral features - restlessness (CPM_top_15_freq3) and global rhythm balance (CPM_median_freq) - were consistently associated with weekly symptom burden in this cohort. Findings are preliminary and warrant prospective validation for remote symptom monitoring.

Introduction: Myasthenia gravis (MG) is a chronic autoimmune neuromuscular disease. Patients with MG are typically evaluated by neuromuscular experts through in-person neurologic examinations. These assessments are time-consuming, require significant disease expertise, and capture only a snapshot of disease.

Methods: Given this need, we developed a multimodal digital health technology (DHT) called BioDigit MG, for monitoring MG symptoms and objectively measuring disease severity. BioDigit MG includes tablet-guided speech and video-based assessments, electronic patient-reported outcomes relevant to MG, and a wearable sensor to measure physical activity and posture during activities of daily living.

Results: We assessed the feasibility and acceptability of BioDigit MG by conducting a clinical study with 20 participants with MG who used the DHT. During the study, a total of 219 speech tasks and 119 videos were collected by the DHT, achieving 100% reliability in data collection and transfer. To evaluate technology acceptance and usability, we conducted face-to-face interviews with the 20 MG patients and 5 expert clinicians. Participants found the DHT highly effective, easy to use, and well-suited to their needs. Efficient and reliable data transfer capabilities of BioDigit MG ensured that patient-generated data were promptly and securely delivered to healthcare providers.

Conclusion: These feasibility findings demonstrate that BioDigit MG is capable of reliable multimodal data collection and is acceptable to both patients and clinicians, supporting its potential for use in future larger scale validation studies.

Introduction: Incorporating outcome measures that assess the most impactful symptoms is a priority for clinical trials. We qualitatively examined whether caregivers of individuals with Rett syndrome deemed breathing dysfunction as a meaningful and measurable aspect of health.

Methods: We conducted semi-structured interviews (N = 13) with caregivers of individuals with Rett syndrome followed by thematic analysis grounded in theory to examine themes.

Results: Themes and subthemes for experiences with breathing dysfunction emerged: (1) meaningfulness; (2) impact; and (3) connecting with other symptoms. Two themes for preferences for digitally measuring breathing dysfunction emerged: (1) conditional willingness and (2) benefits of digital measurement.

Conclusion: Caregivers reported that breathing dysfunction was meaningful and measurable and had significant impacts on their child's lives as well as theirs and their families. This study lays the groundwork for guiding the development of novel measures and outcomes within future clinical trials managing breathing dysfunction in Rett syndrome.

Introduction: Cognitive performance declines with age and predicts important life outcomes, making it a promising - yet underutilized - biomarker of aging. In this study, we aimed to establish the feasibility and value of game-based digital biomarkers of cognitive aging using data from a home-based cognitive assessment game.

Methods: Participants (N = 871; age 18-75) completed Tunnel Runner, a 20-25 min cognitive game measuring reaction speed, response inhibition, interference control, response-rule switching, and decision-making. To assess the game's out-of-sample predictive accuracy, we trained machine learning models to predict participants' chronological age based on 17 game-based cognitive metrics and evaluated their performance using nested cross-validation. Cognitive aging scores were calculated as out-of-sample prediction errors from the best-performing model, and then adjusted for age-dependence using generalized additive models. These aging scores were then considered alongside three other variables: depression, ADHD, and gamer identity.

Results: The best-performing model, stacked ensemble from the automated machine learning framework AutoGluon, predicted out-of-sample chronological age with a mean absolute error of 6.97 years, a correlation of 0.626, and concordance of 0.698. No evidence of bias in predictive accuracy was found for gender or gaming identity. Prediction patterns and cognitive aging values met several expectations based on previous research: reduced cognitive aging in participants with self-reported ADHD, negative association between cognitive aging and gamer identity, and limited predictive differentiation under age 30. Findings regarding self-reported depression were inconclusive, though consistent with prior work.

Conclusion: Game-based assessment can produce accessible digital biomarkers of cognitive aging that reflect meaningful individual differences. This approach enables scalable and low-burden cognitive aging assessment, with potential applications for early detection of cognitive decline, longitudinal tracking, and intervention evaluation.

Introduction: A primary goal of physical medicine and rehabilitation is restoring community mobility after injury or illness. However, there is no clinically accepted real-world method to measure community mobility, which fundamentally limits our ability to evaluate treatment effectiveness. This study aimed to develop and validate a digital framework using GPS-enabled smartphones and inertial sensors to monitor community mobility and estimate clinical function in individuals with chronic stroke or lower limb amputation (LLA).

Methods: Ninety individuals with chronic stroke or LLA underwent remote monitoring for 3-9 months. Participants completed standard clinical assessments, and daily mobility data were extracted from GPS and step count features. We conducted four analyses: (1) characterization of group- and individual-level community mobility, (2) evaluation of mobility changes following a mobility-targeted intervention in a single case participant, (3) development of machine-learned models to predict clinical gait outcomes using community data, and (4) estimation of the minimum number of days needed to reliably predict functional outcomes.

Results: Community mobility measures revealed substantial variability both across and within individuals, reflecting diverse functional profiles. In a case study, a participant with LLA demonstrated increased activity and movement diversity following a personalized intervention. Machine-learned models estimated 6-Minute Walk Test and 10-Meter Walk Test scores with clinically acceptable error margins (7-10%) using as few as 14 days of community data. Reliable predictions were achievable with just 3-6 days of monitoring.

Conclusions: GPS- and smartphone-based monitoring offer a feasible and scalable approach to assess real-world mobility. This approach could close a critical gap in the care continuum and enable us to fully evaluate the real-world impact of treatment interventions while also reducing reliance on frequent in-person evaluations.