January 7, 2026 | After less than three years of collecting voice samples from individuals with amyotrophic lateral sclerosis (ALS), and a control group of healthy volunteers, the Peter Cohen Foundation awaits approval from the Food and Drug Administration (FDA) of its algorithm as a secondary biomarker for tracking the progression of the devastating neurological disease. The drug development tool, a “listener effort” prediction model, will be used in a multitude of clinical trials as soon as the first quarter of 2026, according to Indu Navar, CEO and founder of the nonprofit operating as both EverythingALS and sister brand EverythingAD.
EverythingALS is already actively working with some of the 22 companies composing its consortia of pharma partners to implement the algorithm and that is poised to quickly quadruple in scale, she says, citing the longstanding need for an objective and clinically meaningful measure of progression. The reliance on subjective questionnaires is why trials up to now have required a lot more time and participants—and one big reason why they ultimately fail to meet the regulatory mark for market entry.
The newly developed machine learning model represents the first-ever objective, scalable digital biomarker for any neurological disease, reports Navar. It predicts effort scores, previously a subjective clinician rating, from speech recordings, and is highly sensitive to subtle changes in dysarthria (motor speech disorder) as ALS progresses on a 0-to-100-point scale.
It has been 70-year wait for any significant change in the method of detecting or eliminating ALS, says Navar, a former tech entrepreneur whose husband Peter Cohen died of ALS in 2019 without the benefit of a speedy diagnosis. “We need to make a change in the way we do our clinical trials; otherwise ... we’re in the caveman’s world. It’s like asking someone, ‘How do you feel about the sugar in your blood?’ It’s ridiculous.”
This was what motivated Navar to launch the nonprofit and pursue development of a clinical predictive AI model using speech, much as has been successfully done to rate disease severity based on X-ray and CT images. Only here, the experts serving as performance benchmarks were three unaffiliated speech pathologists.
The trio initially rated 6,000 speech samples, and the algorithm was found to agree with their assessment 96% of the time, she says. Similarly impressive performance was seen when the model was tested on about 5,000 speech samples from the Radcliffe Study of EverythingALS and a third large dataset from the HEALEY ALS Platform Trial led by Massachusetts General Hospital.
This was strong evidence that the machine learning model could reliably predict how speech pathologists would rate speech samples, and it can make that call of progression given only one minute of speech, says Navar. Its deployment in the clinical trials is intended to “move the needle” on stalled progress in ALS by productizing the algorithm.
Since clinical trials require participants, EverythingALS also helps people make that journey. The same trial-matching technique will be used for patients with Alzheimer’s disease (AD) via the EverythingAD.org website, Navar points out. People who have been or are in the process of being diagnosed with either disease are encouraged to register to learn of new trial matches and opportunities.
ALS and AD have a lot of crossovers in terms of pathology, she says. The TDP-43 protein is a key feature of ALS, causing motor neuron death, and is also increasingly recognized in AD, where it worsens cognitive decline. The protein likewise plays a role in Parkinson’s disease and other non-AD dementias.
The data collection phase for development of the ALS algorithm involved over 1,200 study participants, recruited over about 14 months. Individuals who were progressing were asked to continue giving speech samples for as long as possible to aid understanding of how dysarthria happens, says Navar.
The statistical power of all that big data wasn’t good enough for subtyping patients based on the progression of their disease, she continues. Instead, AI was trained with a small set of data to determine the features that matter for progression tracking. Various machine learning models were used for unstructured, structured, and audio data to pinpoint the most reliable feature of them all—the amount of effort listeners put into understanding someone’s speech.
Thousands more patients will be needed to enroll in upcoming ALS studies of consortia partners, she adds. As part of these studies, participants will be asked to recite three rotating sentences representing one minute of speech every other week for five months.
The same sort of approach will be used for assessing the breathing and gross motor skills of patients, using physical therapists to validate those disease progression algorithms. “We’ve collected data from sensors in the feet and insoles and are also collecting video data from people, how they’re walking, [so] we can now predict people’s falls and their balance issues,” Navar says. For purposes of assessing speech, excellent signals were captured from the audio recordings alone.
With each new software product, the development process is expected to shrink—to 18 months initially and eventually under a year, she notes.
The speech-based prediction model, and subsequent algorithms, will all be disengaged from how the data gets collected, says Navar. EverythingALS has variably used third-party brands like Modality.ai and Linus Health for the job as well as its own internally developed collection device focused on the minimum data required to make an accurate prediction. “We leave it to the clinical trial companies or pharma companies to decide.”
The advantages of participation have expanded considerably over time, from “please donate [a voice sample] for the science” to the ability to readily identify the most beneficial trial for any one individual, she says. ALS varies greatly, so the eligibility criteria tend to be strict based on a patient's unique disease profile.
To properly match patients to studies, says Navar, EverythingALS developed an AI chatbot called SAVA named after one of the patients who created a spreadsheet to manually aggregate clinical trial data to help himself and others make informed decisions. Humans are also available to help patients look for trials. “Over 100 trials are going to be opening up in the next 12 to 18 months in ALS,” and many of them are larger phase 2 or 3 studies requiring hundreds of patients.
Navar will be conducting a baselining study for ALS, going live early in 2026, focused on defining the initial characteristics, disease severity, and rate of progression in patients to facilitate their matching to experimental trials as they open. Patients are encouraged to participate once they’ve signed up with EverythingALS to become a citizen scientist.
Information participants share about their date of diagnosis, where they live, and how far they’re willing to travel for a trial will be used to help direct them to enrollment opportunities with the full range of studies listed in more technical language on the government platform, she explains. A baselining study for AD is scheduled to go live in the second quarter of 2026.
Although only 10% of individuals with ALS are currently enrolled in clinical trials, when surveyed, close to 80% of that population report that they want to get into a study, says Navar. Diagnostic delays result in many missed opportunities because trials generally exclude people whose symptoms have been ongoing for more than 16 or 17 months. Eligibility is also tied to breathing requirements as measured by forced vital capacity, which gradually declines throughout the disease course.
Another major bottleneck is that available information, notably on ClinicalTrials.gov, is inaccessible—meaning it is not easily found, understood, and used—and non-specialty physicians are often unfamiliar with clinical trials in their local area or lack the time and inclination to manage a referral, she says.
Currently, the EverythingALS community comprises about 7,000 families who attend regularly scheduled online educational talks and supportive group discussions, says Navar. But anyone in the process of being diagnosed and wants to know their clinical trial options can join the network, and that invitation will be extended via multiple websites.
The goal of EverythingALS is to reduce the enrollment time of patients into trials from the current 12 to 18 months to two months and, ultimately, less than 24 hours, says Navar. People registering on the site and baselined will be handed over to the right clinical trials, based on an AI-powered sweep of their medical records to spot issues that might exclude their participation in some studies, such as a recent heart attack or diabetes that requires treatment with insulin.
EverythingALS works with some of the top ALS experts, including Eduardo Locatelli, M.D., at Nova Southeastern University, Jinsy Andrews, M.D., at NYU Langone Health, and Mass General’s Suma Babu, M.D. They’re all interested in reducing the often complex, confusing, and stressful process of finding and enrolling in a clinical trial under the all-too-real time constraints of disease progression.
Individuals affected by ALS “have the power to change the status quo,” says Navar. “We’re not the victims. But we all have to come together to do it ... there is power in the crowd.”
It’s a passion born of personal experience after her husband was diagnosed, Navar says. “I know how isolating it was and how alone we felt.”
ALS is a “system problem” needing objective measures, quicker matching of individuals into trials, and better monitoring, so they produce the quality of data needed for patient subtyping, Navar stresses. But only if tackled with the entrepreneurial zest of companies like Tesla and Amazon will the world see long-overdue updates to “gold standard” drug development practices for this and other neurological diseases.