"During that therapy process, neither the patient nor the therapist has a clear idea of how patients are responding to the treatments that they're receiving," Lee said. "Currently, clinicians aren't able to see if patients are responding to the prescribed exercises, and patients have no way of knowing how they are progressing."

What does the wearable actually measure?

The device's accelerometer — a sensor that registers motion — records arm movement all day long. A machine-learning algorithm developed by Lee and his graduate student Ryan Wang, the paper's lead author, then converts that raw motion data into an estimate of motor severity, meaning how strongly the stroke has impaired the patient's movement.

That distinction matters. Lee cautions that movement and impairment are related but not identical. A patient can choose to use their arm more, but they cannot instantly change the underlying severity of their impairment through short-term effort.

"Increasing the use of the limbs — yes, we can encourage the person to make use of the limb more," Lee said. "But patients cannot make instant changes to motor severity through short-term behavior change."

The team trained the algorithm on accelerometer data and clinician assessment scores from healthy individuals and from patients in the subacute phase of recovery — the window from one week to six months after a stroke. When tested, the model reflected a patient's true condition 40%–50% better than clinician evaluations.

Could it also shrink clinical trials?

Beyond personalizing treatment, the device showed promise as a research tool. When the researchers recreated a previous study using their digital biomarker — a measurable indicator of a biological condition — instead of clinician observations, they reached statistically significant results with 50% fewer participants.

"We can get a clear idea of the effectiveness of the intervention using a lower sample size and far fewer resources," Lee said. Smaller trials could mean faster research and lower costs.

What happens next?

Lee has filed a provisional patent and is pursuing commercialization through a startup called Lumid Health. His research partners are now recruiting stroke patients for a study at Spaulding Rehabilitation Hospital in Boston.

Because the device captures movement in everyday environments rather than at a single clinic visit, the data may better reflect how patients genuinely function. Continuous feedback could also help patients stay engaged: seeing measurable progress, Lee suggests, may keep motivation up and improve therapy outcomes.

As with any preliminary technology, wider use will depend on further validation in larger, more diverse patient groups. For now, the UMass Amherst team has demonstrated a proof of principle: a sensor small enough to sit on a wrist can see recovery — or its absence — more clearly than the human eye.