Digital Biomarkers in Neuromuscular Disorders: Recent Advances 

Neuromuscular disease (NMD) can affect any level of the neuroaxis from cortex to myocyte. Neuromuscular function controls one’s ability to communicate, sense, and navigate the environment; therefore, NMDs can profoundly affect quality of life. NMDs are heterogeneous with respect to onset, course, severity, localization, mechanism, pathophysiology, immunoresponsiveness, and treatment. The morbidity and heterogeneity of NMDs complicate their clinical management and research aimed at finding solutions and cures. As such, new approaches are needed to better understand and treat NMDs. Digital health technologies (DHTs) have the potential to aid in the study and management of these challenging conditions. 

Terminology 

With the rapid expansion of DHT, appropriate vocabulary is necessary to facilitate communication between developers and regulators.¹ DHTs are systems that use computing platforms, connectivity, software, and sensors for health care and related uses. The US Food and Drug Administration (FDA) and the National Institutes of Health Biomarker Working Group define a biomarker as “a defined characteristic that is measured as an indicator of normal biological processes, [pathologic] processes, or biological responses to an exposure or intervention, including therapeutic interventions.”² Digital biomarkers (DBs) are biomarkers obtained from DHTs. Despite a lack of consensus regarding the definition of DBs,³ they are minimally defined by their context of use (eg, diagnosis), type of data measured (eg, electrocardiogram), and collection method (eg, smartwatch sensor). Digital clinical outcome assessments are also derived from DHTs and involve a clinician, observer, or patient describing or reflecting how an individual feels, functions, or survives.² In the following, we focus on recent DB advances for NMDs.

Current Use

The technology enabling the rapid development of DHTs includes computers, smartphones, wearables, and implantable, ingestible, or endoscopic devices that integrate sensors measuring forces, accelerations, sound, video, electrophysiology, temperature, and/or biochemical changes. The context of the disease under study dictates how these sensors are applied to report clinically relevant features, such as heart rate variability, EMG, limb range of motion, velocity, orientation, gait, falls, transfers, facial and ocular kinematics, and speech. A 2021 systematic review of DBs for NMDs examined 9 human studies in amyotrophic lateral sclerosis (ALS), spinal muscular atrophy, and Duchenne muscular dystrophy (DMD).⁴ Neuromuscular junction (NMJ) disease (ie, myasthenia gravis) has also been studied.⁵ In the following sections, we highlight some of the notable advances in DHTs that are tailored for use in NMDs. 

Stride Velocity Using Wearables in DMD

The pathway for regulatory approval for DHT end points in clinical trials is at an early stage and frameworks for verification have only recently been developed. It should be noted that standalone software is considered as a medical device when used for medical purposes. Fewer than a dozen submissions have been made to the FDA’s Drug Development Tool qualification program (launched in 2016) and none of the submissions have advanced beyond preliminary review.⁶ The FDA’s Innovative Science and Technology Approaches for New Drugs (ISTAND) pilot program⁷ was created to evaluate potential drug development tool candidates that may not fit within existing qualification programs but could be beneficial for drug development (eg, those that combine DHTs with artificial intelligence [AI]) for the promotion of decentralized clinical trials.

Bakker and colleagues⁸ offer a comprehensive review of the current US regulatory landscape concerning DBs in clinical trials. The European Medicines Agency (EMA) has a similar program for DB development, recently issuing guidance on applications in accelerometry, ingestible drug sensors, and AI tools.⁹ In 2023, the EMA set a precedent for qualified DBs with their approval of “stride velocity 95th centile” as a primary end point for trials in DMD.⁸ The ActiMyo (Sysnav; Vernon, France) movement sensor records the fastest 5% of strides during routine walking, is approved for people age ≥4 years with DMD, and represents maximal ambulatory performance in the home environment.¹⁰ Its approval represents decades of research in sensor and algorithm development applied to the specific context of measuring decline or improvement in walking speed of people with DMD over time. 

Speech Analysis in ALS Spectrum Disorders

Other methods have built a large evidence base from small to medium-sized observational trials, but lack regulatory approval as validated endpoints. Digital monitoring of speech in ALS research has seen exceptional growth. Driven by the need for objective, rapid, and less burdensome assessment of bulbar dysfunction as well as the seeming omnipresence of microphones on account of the prevalence of smartphones, speech has emerged as a viable biomarker. Speech features can be divided into acoustic, linguistic, and paralinguistic features which represent the auditory properties of speech (eg, variations in frequency, duration, intensity), language (eg, vocabulary, syntax), and nonphonemic aspects of speech (eg, respiration, volume), respectively.¹¹ Common speech tasks used to elicit these features include rapid syllable alternation, passage reading, sustained vowel phonation, and picture descriptions. Basic features of speech (eg, duration, rate, and percent pause time) have been demonstrated consistently to correlate with disease progression assessed by bulbar and respiratory subscales of the Amyotrophic Lateral Sclerosis Functional Rating Scale–Revised and have sometimes detected changes in disease progression prior to detection using this clinical scale.^12,13

Clinicians have successfully used speech DBs to detect effects of medication treatment on pseudobulbar affect, even when experienced clinicians and validated scale assessments were unable to do so.¹⁴ Speech DBs have also been utilized to differentiate primary upper versus lower motor neuron involvement using syllable rates,¹⁵ and speech fundamental frequencies have been used to differentiate ALS–frontotemporal dementia from symptomatic bulbar ALS.¹⁶ Although no speech DB for ALS has been approved by a regulatory agency, there is a large and growing body of evidence supporting the use of novel metrics to quantify features of speech which will likely be important for diagnosis, management, and drug discovery.

Image Analysis in Myasthenia Gravis

Continuous monitoring may be especially useful in NMJ disease given the unique, potentially highly fluctuating nature of the disease course because the ability to collect data remotely makes frequent or continuous monitoring possible. One such technology is being developed by Garbey and colleagues⁵ using commercial videoconferencing to digitize the Myasthenia Gravis Core Exam. The authors quantified ptosis, ocular alignment, motor movements, and respiratory function with varying levels of accuracy and consistency. Subjective measurements can be converted from qualitative to quantitative measurements through digitization while also eliminating interrater variability. However, this must also be balanced with quality data acquisition to prevent unwanted variability as the data quality will be variably dependent on how effectively the patient participates. The use of manual or automated methods for identifying and adapting for low-quality data is essential to the collection of high quality data, and AI tools may assist with this process. The authors highlight the potential role of AI to integrate data from multiple sources to result in meaningful composite scores that outperform traditional observations.⁵

Unique Advantages of DB

There are a number of benefits associated with incorporating DBs into health care delivery. DBs allow for monitoring outside of formal health care settings, reducing barriers to participation in research or clinical care related to travel cost, distance, effort, logistics, infectious exposure, time, and transportation. They may also allow for more representative participation. These factors are especially important given the mobility limitations of people with NMD. 

Passive and remote monitoring are likely to remove or attenuate observer biases, can provide continuous data collection, and can do so within one’s lived environment. Even if not continuous, more frequent data collection can be particularly helpful in clinical trials by reducing measurement uncertainty and thus the required sample size to observe a treatment effect, which allows for more efficient and likely accurate clinical trial results.¹⁷

Rapidly scalable DBs can be used to collect data that might otherwise only be collected at highly specialized centers and by trained personnel. As such, travel restrictions, specialist shortages, and interrater variability can be easily overcome as long as the necessary tools for collection are present. DBs can provide greater sensitivity to change than traditional clinical outcome measures, allowing for earlier diagnosis and intervention, more accurate prognostic information, and optimization of participant enrollment and stratification in clinical trials.^4,12,17,18

Challenges to Implementation 

The diversity within individual NMDs and among the hundreds of NMDs presents several challenges to DB design, validation, and adoption. Neuromuscular impairment can be multifactorial and can include motor, sensory, and extrapyramidal components. Individuals with NMDs can present with exclusively lower motor neuron dysfunction, upper motor neuron dysfunction, or a mixture. In addition, these symptoms manifest differently and may vary over time. NMDs are frequently nonstatic in course. They usually worsen, but may plateau or improve. For example, NMJ disease can vary between worsening and improvement from week to week or hour to hour. Pediatric-onset NMDs require clinicians to utilize innovative strategies such as using sensors in game-based play to acquire patient data.¹⁸ Similar considerations must be made for individuals with comorbidities affecting thinking, behavior, speech, or movement. 

Several non–disease-specific challenges also exist. Determining the appropriate burden-to-information ratio is important.¹⁹ Monitoring which is too frequent may negatively affect compliance, and an unused or disused biomarker is ineffective. The FDA’s Biomarker Qualification Program (BQP) recognizes the importance of patient compliance and requires rigorous demonstration of patient acceptability. Third-party vendors may charge thousands of dollars for devices, data storage, and algorithm use. For DBs to be used within the continuum of care, health care stakeholders (including third-party vendors) should make considerable investment in the use of global identifiers and data standards for interoperability across health systems.²⁰

As with other digital capture of patient information, data privacy is paramount. Developers face the challenge of creating robust data security protocols which meet regulator requirements and satisfy user concerns. Other ethical challenges include the equity of health care access, systemic biases in DB development and validation, data ownership, and accountability in health care decision-making.²¹ This process of DB approval varies by country and can be lengthy. For example, in the United States the BQP²² aims to process applications from submission to a decision in 19 months. Modifications to biomarkers or context of use require reengaging the BQP. Without changes in regulatory pathways, timelines will struggle to match the pace of advances in DB technologies. 

Future Directions

The future is bright for the development of DBs and their incorporation into clinical practice. However, before clinical acceptance and adoption, DBs must first demonstrate validity and viability in clinical trials by comparison with known standards. DBs have the potential to provide actionable insights for neuromuscular medicine in the near future. As large-language models have revolutionized natural language processing within the field of AI, multimodal models will integrate digital sensor, laboratory, imaging, electrodiagnostic, and clinical data to improve health care decision-making, efficiency, and care delivery.²³ Including patients as stakeholders is important for optimizing the convenience, comfort, and interoperability of these tools, the success of which depend heavily on their investment. Just as MRI revolutionized the practice of medicine, DHTs in combination with AI are poised to redefine the way we conduct research and deliver care.