Digital Phenotyping in Clinical Neurology

Eye Tracking during Passage Reading Supports Precise Oculomotor Assessment in Ataxias

Abnormal eye movements occur early in the course of disease in many ataxias. However, clinical assessments of oculomotor function lack precision, limiting sensitivity for measuring progression and the ability to detect subtle early signs. Quantitative assessment of eye movements during everyday behaviors such as reading has potential to overcome these limitations and produce functionally relevant measures. In this study, we analyze eye movements in individuals with ataxia during passage reading. Binocular gaze sampled at 1000 Hz was collected from 102 individuals with ataxia diagnoses (including 36 spinocerebellar ataxias, 12 Friedreich's ataxia, and 5 multiple system atrophy among other conditions) and 70 healthy controls participating in the Neurobooth study. Longitudinal data were available for 26 participants with ataxia. Saccades were categorized as progressive (rightward) saccades, regressive saccades, or sweeps (large displacement saccades primarily generated when scanning to the beginning of the next line) based on their direction and displacement. Saccade and fixation kinematics were summarized using 28 statistical features. A linear model was trained to estimate clinician-performed ataxia rating scale scores. Model scores were reliable (ICC=0.96, p<0.001) and demonstrated convergent validity with Brief Ataxia Rating Scale total (r=0.82, p<0.001), oculomotor (r=0.52, p<0.001), and speech (r=0.73, p<0.001) scores, as well as patient surveys. The scores were also sensitive to disease progression (d=0.36, p=0.03), demonstrated strong separability between healthy controls and participants with ataxias (AUC=0.89, p<0.001), and showed evidence of the ability to detect subclinical oculomotor patterns (AUC=0.69, p=0.02). Several kinematic saccade and fixation features demonstrated strong differences across disease severity groups. Notable features included the mean angular displacement of fixations ( η 2 =0.44, p<0.001), the number ( η 2 =0.27, p<0.001) and frequency of saccades ( η 2 =0.25, p<0.001), and the proportion of regressive saccades ( η 2 =0.11, p<0.001). Quantitative assessment of eye movements during passage reading were highly informative of ataxia severity, were sensitive to disease progression, and enabled detection of subclinical signs. These properties support the inclusion of video-oculography-based measures of reading in natural history studies and clinical trials. Furthermore, this study demonstrates the feasibility of integration of oculomotor assessments in clinical workflows.

Abstract Figure

The Newborn Screening Programme Revisited: An Expert Opinion on the Challenges of Rett Syndrome

Genomic sequencing has the potential to revolutionise newborn screening (NBS) programmes. In 2024, Genomics England began to recruit for the Generation Study (GS), which uses whole genome sequencing (WGS) to detect genetic changes in 500 genes in more than 200 rare conditions. Ultimately, its purpose is to facilitate the earlier identification of rare conditions and thereby improve health-related outcomes for individuals. The adoption of rare conditions into the GS was guided by four criteria: (1) the gene causing the condition can be reliably detected; (2) if undiagnosed, the rare condition would have a serious impact; (3) early or presymptomatic testing would substantially improve outcomes; and (4) interventions for conditions screened are accessible to all. Rett syndrome (RTT, OMIM 312750), a paediatric neurodevelopment disorder, was not included in the list of rare conditions in the GS. In this opinion article, we revisit the GS and discuss RTT from the perspective of these four criteria. We begin with an introduction to the GS and then summarise key points about the four principles, presenting challenges and opportunities for individuals with RTT. We provide insight into how data could be collected during the presymptomatic phase, which could facilitate early diagnosis and improve our understanding of the prodromal stage of RTT. Although many features of RTT present a departure from criteria adopted by the GS, advances in RTT research, combined with advocacy from parent-based organisations, could facilitate its entry into future newborn screening programmes.

Sensitive Quantification of Cerebellar Speech Abnormalities Using Deep Learning Models

Objective, sensitive, and meaningful disease assessments are critical to support clinical trials and clinical care. Speech changes are one of the earliest and most evident manifestations of cerebellar ataxias. This work aims to develop models that can accurately identify and quantify clinical signs of ataxic speech. We use convolutional neural networks to capture the motor speech phenotype of cerebellar ataxia based on time and frequency partial derivatives of log-mel spectrogram representations of speech. We train classification models to distinguish patients with ataxia from healthy controls as well as regression models to estimate disease severity. Classification models were able to accurately distinguish healthy controls from individuals with ataxia, including ataxia participants who clinicians rated as having no detectable clinical deficits in speech. Regression models produced accurate estimates of disease severity, were able to measure subclinical signs of ataxia, and captured disease progression over time. Convolutional networks trained on time and frequency partial derivatives of the speech signal can detect sub-clinical speech changes in ataxias and sensitively measure disease change over time. Learned speech analysis models have the potential to aid early detection of disease signs in ataxias and provide sensitive, low-burden assessment tools in support of clinical trials and neurological care.

Remote monitoring of amyotrophic lateral sclerosis using wearable sensors detects differences in disease progression and survival: a prospective cohort study

BACKGROUND

There is an urgent need for objective and sensitive measures to quantify clinical disease progression and gauge the response to treatment in clinical trials for amyotrophic lateral sclerosis (ALS). Here, we evaluate the ability of an accelerometer-derived outcome to detect differential clinical disease progression and assess its longitudinal associations with overall survival in patients with ALS.

METHODS

Patients with ALS wore an accelerometer on the hip for 3-7 days, every 2-3 months during a multi-year observation period. An accelerometer-derived outcome, the Vertical Movement Index (VMI), was calculated, together with predicted disease progression rates, and jointly analysed with overall survival. The clinical utility of VMI was evaluated using comparisons to patient-reported functionality, while the impact of various monitoring schemes on empirical power was explored through simulations.

FINDINGS

In total, 97 patients (70.1% male) wore the accelerometer for 1995 days, for a total of 27,701 h. The VMI was highly discriminatory for predicted disease progression rates, revealing faster rates of decline in patients with a worse predicted prognosis compared to those with a better predicted prognosis (p < 0.0001). The VMI was strongly associated with the hazard for death (HR 0.20, 95% CI: 0.09-0.44, p < 0.0001), where a decrease of 0.19-0.41 unit was associated with reduced ambulatory status. Recommendations for future studies using accelerometery are provided.

INTERPRETATION

The results serve as motivation to incorporate accelerometer-derived outcomes in clinical trials, which is essential for further validation of these markers to meaningful endpoints.

FUNDING

Stichting ALS Nederland (TRICALS-Reactive-II).

Clinical Trial-Ready Patient Cohorts for Multiple System Atrophy: Coupling Biospecimen and iPSC Banking to Longitudinal Deep-Phenotyping

Multiple system atrophy (MSA) is a fatal neurodegenerative disease of unknown etiology characterized by widespread aggregation of the protein alpha-synuclein in neurons and glia. Its orphan status, biological relationship to Parkinson's disease (PD), and rapid progression have sparked interest in drug development. One significant obstacle to therapeutics is disease heterogeneity. Here, we share our process of developing a clinical trial-ready cohort of MSA patients (69 patients in 2 years) within an outpatient clinical setting, and recruiting 20 of these patients into a longitudinal "n-of-few" clinical trial paradigm. First, we deeply phenotype our patients with clinical scales (UMSARS, BARS, MoCA, NMSS, and UPSIT) and tests designed to establish early differential diagnosis (including volumetric MRI, FDG-PET, MIBG scan, polysomnography, genetic testing, autonomic function tests, skin biopsy) or disease activity (PBR06-TSPO). Second, we longitudinally collect biospecimens (blood, CSF, stool) and clinical, biometric, and imaging data to generate antecedent disease-progression scores. Third, in our Mass General Brigham SCiN study (stem cells in neurodegeneration), we generate induced pluripotent stem cell (iPSC) models from our patients, matched to biospecimens, including postmortem brain. We present 38 iPSC lines derived from MSA patients and relevant disease controls (spinocerebellar ataxia and PD, including alpha-synuclein triplication cases), 22 matched to whole-genome sequenced postmortem brain. iPSC models may facilitate matching patients to appropriate therapies, particularly in heterogeneous diseases for which patient-specific biology may elude animal models. We anticipate that deeply phenotyped and genotyped patient cohorts matched to cellular models will increase the likelihood of success in clinical trials for MSA.

At-home wearables and machine learning sensitively capture disease progression in amyotrophic lateral sclerosis

Amyotrophic lateral sclerosis causes degeneration of motor neurons, resulting in progressive muscle weakness and impairment in motor function. Promising drug development efforts have accelerated in amyotrophic lateral sclerosis, but are constrained by a lack of objective, sensitive, and accessible outcome measures. Here we investigate the use of wearable sensors, worn on four limbs at home during natural behavior, to quantify motor function and disease progression in 376 individuals with amyotrophic lateral sclerosis. We use an analysis approach that automatically detects and characterizes submovements from passively collected accelerometer data and produces a machine-learned severity score for each limb that is independent of clinical ratings. We show that this approach produces scores that progress faster than the gold standard Amyotrophic Lateral Sclerosis Functional Rating Scale-Revised (-0.86 ± 0.70 SD/year versus -0.73 ± 0.74 SD/year), resulting in smaller clinical trial sample size estimates (N = 76 versus N = 121). This method offers an ecologically valid and scalable measure for potential use in amyotrophic lateral sclerosis trials and clinical care.

Sensitive quantification of cerebellar speech abnormalities using deep learning models

Objective

Objective, sensitive, and meaningful disease assessments are critical to support clinical trials and clinical care. Speech changes are one of the earliest and most evident manifestations of cerebellar ataxias. The purpose of this work is to develop models that can accurately identify and quantify these abnormalities.

Methods

We use deep learning models such as ResNet 18 , that take the time and frequency partial derivatives of the log-mel spectrogram representations of speech as input, to learn representations that capture the motor speech phenotype of cerebellar ataxia. We train classification models to separate patients with ataxia from healthy controls as well as regression models to estimate disease severity.

Results

Our model was able to accurately distinguish healthy controls from individuals with ataxia, including ataxia participants with no detectable clinical deficits in speech. Furthermore the regression models produced accurate estimates of disease severity, were able to measure subclinical signs of ataxia, and captured disease progression over time in individuals with ataxia.

Conclusion

Deep learning models, trained on time and frequency partial derivatives of the speech signal, can detect sub-clinical speech changes in ataxias and sensitively measure disease change over time.

Significance

Such models have the potential to assist with early detection of ataxia and to provide sensitive and low-burden assessment tools in support of clinical trials and neurological care.

Real-life ankle submovements and computer mouse use reflect patient-reported function in adult ataxias

Novel disease-modifying therapies are being evaluated in spinocerebellar ataxias and multiple system atrophy. Clinician-performed disease rating scales are relatively insensitive for measuring disease change over time, resulting in large and long clinical trials. We tested the hypothesis that sensors worn continuously at home during natural behaviour and a web-based computer mouse task performed at home could produce interpretable, meaningful and reliable motor measures for potential use in clinical trials. Thirty-four individuals with degenerative ataxias (spinocerebellar ataxia types 1, 2, 3 and 6 and multiple system atrophy of the cerebellar type) and eight age-matched controls completed the cross-sectional study. Participants wore an ankle and wrist sensor continuously at home for 1 week and completed the Hevelius computer mouse task eight times over 4 weeks. We examined properties of motor primitives called 'submovements' derived from the continuous wearable sensors and properties of computer mouse clicks and trajectories in relationship to patient-reported measures of function (Patient-Reported Outcome Measure of Ataxia) and ataxia rating scales (Scale for the Assessment and Rating of Ataxia and the Brief Ataxia Rating Scale). The test-retest reliability of digital measures and differences between ataxia and control participants were evaluated. Individuals with ataxia had smaller, slower and less powerful ankle submovements during natural behaviour at home. A composite measure based on ankle submovements strongly correlated with ataxia rating scale scores (Pearson's r = 0.82-0.88), strongly correlated with self-reported function (r = 0.81), had high test-retest reliability (intraclass correlation coefficient = 0.95) and distinguished ataxia and control participants, including preataxic individuals (n = 4) from controls. A composite measure based on computer mouse movements and clicks strongly correlated with ataxia rating scale total (r = 0.86-0.88) and arm scores (r = 0.65-0.75), correlated well with self-reported function (r = 0.72-0.73) and had high test-retest reliability (intraclass correlation coefficient = 0.99). These data indicate that interpretable, meaningful and highly reliable motor measures can be obtained from continuous measurement of natural movement, particularly at the ankle location, and from computer mouse movements during a simple point-and-click task performed at home. This study supports the use of these two inexpensive and easy-to-use technologies in longitudinal natural history studies in spinocerebellar ataxias and multiple system atrophy of the cerebellar type and shows promise as potential motor outcome measures in interventional trials.

The visualization of Orphadata neurology phenotypes

Disease phenotypes are characterized by signs (what a physician observes during the examination of a patient) and symptoms (the complaints of a patient to a physician). Large repositories of disease phenotypes are accessible through the Online Mendelian Inheritance of Man, Human Phenotype Ontology, and Orphadata initiatives. Many of the diseases in these datasets are neurologic. For each repository, the phenotype of neurologic disease is represented as a list of concepts of variable length where the concepts are selected from a restricted ontology. Visualizations of these concept lists are not provided. We address this limitation by using subsumption to reduce the number of descriptive features from 2,946 classes into thirty superclasses. Phenotype feature lists of variable lengths were converted into fixed-length vectors. Phenotype vectors were aggregated into matrices and visualized as heat maps that allowed side-by-side disease comparisons. Individual diseases (representing a row in the matrix) were visualized as word clouds. We illustrate the utility of this approach by visualizing the neuro-phenotypes of 32 dystonic diseases from Orphadata. Subsumption can collapse phenotype features into superclasses, phenotype lists can be vectorized, and phenotypes vectors can be visualized as heat maps and word clouds.

Accuracy and Reliability of At-home Quantification of Motor Impairments Using a Computer-based Pointing Task with Children with Ataxia-Telangiectasia

Methods for obtaining accurate quantitative assessments of motor impairments are essential in accessibility research, design of adaptive ability-based assistive technologies, as well as in clinical care and medical research. Currently, such assessments are typically performed in controlled laboratory or clinical settings under professional supervision. Emerging approaches for collecting data in unsupervised settings have been shown to produce valid data when aggregated over large populations, but it is not yet established if in unsupervised settings measures of research or clinical significance can be collected accurately and reliably for individuals. We conducted a study with 13 children with ataxia-telangiectasia and 9 healthy children to analyze the validity, test-retest reliability, and acceptability of at-home use of a recent active digital phenotyping system, called Hevelius. Hevelius produces 32 measures derived from the movement trajectories of the mouse cursor, and it produces a quantitative estimate of motor impairment in the dominant arm using the dominant arm component of the Brief Ataxia Rating Scale (BARS). The severity score estimates generated by Hevelius from single at-home sessions deviated from clinician-assigned BARS scores more than the severity score estimates generated from single sessions conducted under researcher supervision. However, taking a median of as few as 2 consecutive sessions produced severity score estimates that were as accurate or better than the estimates produced from single supervised sessions. Further, aggregating as few as 2 consecutive sessions resulted in good test-retest reliability (ICC = 0.81 for A-T participants). This work demonstrated the feasibility of performing accurate and reliable quantitative assessments of individual motor impairments in the dominant arm through tasks performed at home without supervision by the researchers. Further work is needed, however, to assess how broadly these results generalize.

Assessing Cerebellar Disorders with Wearable Inertial Sensor Data Using Time-Frequency and Autoregressive Hidden Markov Model Approaches

Wearable sensor data is relatively easily collected and provides direct measurements of movement that can be used to develop useful behavioral biomarkers. Sensitive and specific behavioral biomarkers for neurodegenerative diseases are critical to supporting early detection, drug development efforts, and targeted treatments. In this paper, we use autoregressive hidden Markov models and a time-frequency approach to create meaningful quantitative descriptions of behavioral characteristics of cerebellar ataxias from wearable inertial sensor data gathered during movement. We create a flexible and descriptive set of features derived from accelerometer and gyroscope data collected from wearable sensors worn while participants perform clinical assessment tasks, and use these data to estimate disease status and severity. A short period of data collection (<5 min) yields enough information to effectively separate patients with ataxia from healthy controls with very high accuracy, to separate ataxia from other neurodegenerative diseases such as Parkinson’s disease, and to provide estimates of disease severity.

Automatic Classification and Severity Estimation of Ataxia From Finger Tapping Videos

Digital assessments enable objective measurements of ataxia severity and provide informative features that expand upon the information obtained during a clinical examination. In this study, we demonstrate the feasibility of using finger tapping videos to distinguish participants with Ataxia (N = 169) from participants with parkinsonism (N = 78) and from controls (N = 58), and predict their upper extremity and overall disease severity. Features were extracted from the time series representing the distance between the index and thumb and its derivatives. Classification models in ataxia archived areas under the receiver-operating curve of around 0.91, and regression models estimating disease severity obtained correlation coefficients around r = 0.64. Classification and prediction model coefficients were examined and they not only were in accordance, but were in line with clinical observations of ataxia phenotypes where rate and rhythm are altered during upper extremity motor movement.