Anatomy of a trial registration NCT03582397.

What does the registry record actually contain?

Filed by Columbia University with Joel Stein as principal investigator, the record describes a device feasibility study of VR mirror therapy for upper limb rehabilitation after stroke. The intervention was the WiseMind software, developed by collaborator Realiteer Corp., delivered on an Oculus Rift: two 15-minute sessions per visit, three visits a week, for four weeks.

The design line is the most informative part: single group, pre-test/post-test, no allocation, no masking. Eleven participants actually enrolled. The study ran from October 2017 to July 2018 at Columbia University Medical Center and is marked completed.

Primary outcomes are all feasibility measures: adherence to visits, tolerance via the Simulator Sickness Questionnaire, usability via the System Usability Scale, adverse-event tracking, and cognitive feasibility via the Montreal Cognitive Assessment. The clinical motor measures, the Fugl-Meyer arm/hand section and the Action Research Arm Test, appear as secondary outcomes.

Why are no results posted, and where did they go?

The registry itself contains no outcome data, which is common: results-posting on ClinicalTrials.gov lags and is inconsistently enforced, particularly for small feasibility studies. The scientifically important question is whether the trial reached the peer-reviewed literature, and this one did.

The identifiers line up cleanly with the 2019 paper by Weber, Nilsen, Gillen, Yoon, and Stein in the American Journal of Physical Medicine & Rehabilitation: same institution, same PI, same intervention format and outcome battery, with 11 enrolled on the registry and 10 completers in the paper. We analyse that paper in detail in a companion commentary; in brief, tolerability was excellent and the Fugl-Meyer change (21.7 to 22.8) was not statistically significant.

How do you read a trial registration critically?

A few fields do most of the work, and NCT03582397 illustrates each:

• Design and masking. Single group with no control or blinding means no causal claim about efficacy will ever be supportable from this study, no matter the results.
• Actual enrollment. Eleven people. Any percentages or p-values downstream of that number inherit its fragility.
• Primary versus secondary outcomes. What a study is powered and designed to answer is the primary outcome, feasibility here. Motor recovery findings from secondary outcomes are exploratory by construction.
• Sponsor and collaborators. A software company collaborating on a trial of its own product is a disclosed interest worth carrying into your reading of any later claims.
• Status and results. “Completed” with no posted results means: go find the publication, and note whether one exists at all.

None of these are red flags in themselves; small unblinded feasibility trials are how device research is supposed to begin. They become misleading only when downstream marketing quotes such a trial as though it demonstrated efficacy.

Where does this trial sit in the VR rehabilitation evidence?

It is an early rung on a ladder the field is still climbing. Feasibility pilots like this one establish that immersive mirror therapy can be delivered to stroke patients; mechanism studies such as the mirror-feedback fMRI work and VR-guided motor imagery experiments establish that the brain responds along the theorized pathways; and pooled analyses like the 2021 subacute stroke meta-analysis test whether the accumulated trials add up to a clinical effect.

For chronic pain, the relevance is by analogy of mechanism: the same visual-feedback principle underpins mirror therapy and its VR descendants, including the embodiment training in Karuna's program. The registry habit travels too, when evaluating any program, ours included, asking “what was the design, what was the n, what was primary” is the right reflex.