Digital endpoints in clinical trials: what it takes to turn a sensor signal into evidence

On 20 July 2023, the European Medicines Agency's scientific committee accepted stride velocity 95th centile, known as SV95C, as a primary endpoint in trials for Duchenne muscular dystrophy, a genetic disease that progressively weakens the muscles of boys and young men. The measure comes from a sensor strapped to the ankle and worn at home. It records walking speed over weeks of ordinary life and keeps the speed of the fastest strides, the 95th centile. The agency had accepted it as a secondary endpoint in 2019. The 2023 opinion, which covers ambulant patients aged four and over, was described at the time as the first qualification of a wearable-derived measure as a primary endpoint in pivotal drug trials.

An endpoint is the outcome a trial is designed to measure in order to show whether a treatment works. A digital endpoint is one derived from a sensor or connected device, often collected outside the clinic. Most trial assessments happen during scheduled hospital visits, and between them sponsors rely largely on questionnaires in which patients report how they feel.

Sumu Sethi, who holds a PhD in respiratory medicine, is senior director and digital health strategy lead for respiratory, immunology and inflammation at GSK. She was introduced on the show as having spent more than 25 years in the pharmaceutical industry, from clinical operations at Roche to digital health strategy at GSK. In an interview for Tech Anatomy, she pushed back on the idea that this data between visits was hidden: "I wouldn't say that they were completely invisible. I would say that they are probably less accurate, especially in clinical trials." Patient-reported outcomes, she noted, are subjective by design. A sensor adds an objective layer to them.

The validation path, step by step

Asked what it takes to turn a step or a wrist movement into something a regulator will accept, Sethi began with the Duchenne case: "it's not easy, and it does take a very long time". She recalled the SV95C work taking nearly fourteen years from inception, a figure consistent with the "fourteen-year collaboration" cited by SYSNAV, the French company behind the ActiMyo sensor, when the qualification was announced.

The sequence she described has four stages. First, defining what aspect of health matters to patients and which concept of interest the device should capture in a given context of use. Within "activity", for instance, only some features may track disease progression in a given condition. Second, a technical assessment of the device: whether patients can and will wear it, and the quality of the raw data it produces. Third, analytical validation, which asks whether the algorithm turning sensor signals into a metric is repeatable and measures what it claims to. Fourth, clinical validation, which compares the digital measure against the in-clinic assessment used as the reference.

That sequence closely follows a framework published in 2020 by the Digital Medicine Society (DiMe) under the name V3, for verification, analytical validation and clinical validation. An extension published in January 2025, V3+, adds usability validation, which covers the wearability questions Sethi places early in the process.

What made the Duchenne case work

Sethi said she was involved only briefly, at the start, while at Roche. She described the clinical problem as one of inconsistency: in her recollection, the stair-climbing assessment was run on different staircases, with different heights and numbers of steps, depending on the physician. Duchenne trials do use a timed four-stair climb, alongside the six-minute walk test and a functional scale. The EMA opinion describes all of these as a "snapshot" of maximal ability on the day of the visit, and qualifies SV95C specifically as an alternative to the six-minute walk test.

Her account of why it succeeded is organisational more than technical. "Roche invested a lot of time and a lot of money" collecting evidence, engaged early with the EMA and then the FDA, and had sponsors inside the company and among external physicians. SYSNAV's announcement also credited Roche's contribution as critical to the qualification. The FDA has accepted SV95C as a secondary endpoint in at least one sponsor's trial, but no public record shows a formal FDA qualification as of November 2026.

Questions before adding a sensor

At GSK, Sethi said, many teams want to include a device. Her first question to them is: "why do you want to have a device?" The answer determines the evidence needed: supporting efficacy, a label claim, or differentiation in the market are different goals, and adding a sensor "for the sake of it" adds burden for sites, patients and caregivers. She also pointed to internal procedures that were "designed, rightly so, for clinical trials and not necessarily for the inclusion of digital sensors", which is why she puts so much weight on support from senior leadership.

On regulators, she described a change of mindset after hearing an FDA representative tell an industry audience, roughly seven years ago: "come and speak to us. Come and tell us what you're doing. Tell us what your plans are, because we are learning as well." The FDA's final guidance on digital health technologies for remote data acquisition, published in December 2023, sets out expectations on verification, validation and data security. Sethi still describes the available guidance as "open to interpretation", hence her advice to return to regulators before each validation stage.

Two views of where the field stands

One reading of the SV95C story is that a path now exists: the steps are documented, frameworks such as V3+ formalise them, and a precedent shows that a regulator will accept a wearable measure as primary evidence. Sethi noted that later projects start from further ahead on well-studied measures such as physical activity.

The other reading stresses the cost of that precedent. It took more than a decade, one disease, one measure and a large company willing to keep paying. Sethi herself ties success to leadership commitment and budget, which suggests that the path is open mainly to sponsors with both.

What the interview does not establish

Several benefits came up as possibilities, not findings. Sethi said that involving patients and sites early "might, I'm not saying it will" help recruitment and retention, and that richer data could one day allow earlier decisions or smaller trials. She was explicit that the aim is not to replace existing assessments: "we're not looking to replace anything, we're looking to enrich the data that we already have." She also does not use data from consumer wearables in trials: "I personally wouldn't use data from a device that hasn't been vetted and hasn't been validated." Her view of the field comes from two large sponsors and does not cover how smaller biotech companies or academic trial units approach the same choices.

What to watch

The next indicators are whether the FDA formally qualifies SV95C or another digital measure as a primary endpoint, how many new qualification requests reach the EMA, and how long they take compared with the fourteen years of the Duchenne case. Another is whether published trials start to report the usability data V3+ asks for. Sethi also mentioned an early project, about eight years ago, exploring car sensors that might alert drivers to a health problem. It stalled partly on one question: "does a driver really want their car to tell them that they need to go and see their doctor?" Whether patients want continuous measurement, and on what terms, remains as open as the regulatory side.

Listen to the episode

Digital endpoints in clinical trials: Are we ready ? - Sumu Sethi, GSK

Listen · 26 min