Functional unblinding is the problem this desk keeps returning to, in psychedelic trials, in accelerated TMS, and now, with a new analysis published in the Journal of Affective Disorders, in a home-use transcranial direct current stimulation device for depression. The pattern is familiar: patients receiving an active intervention can often tell they got it, which threatens to inflate a subjective outcome with expectation rather than treatment effect. What makes this particular paper worth attention is not that it found unblinding, nearly every trial like it does, but that it built a method to test whether that unblinding actually biased the result, instead of stopping at the admission that it occurred.
The trial and the question
The analysis draws on data from the Empower trial, a sham-controlled study of a home-use tDCS device for depression originally published in Nature Medicine, reanalyzing 149 participants’ data specifically to interrogate the blind. At the ten-week endpoint, researchers had three pieces of information for each participant: whether they actually received active stimulation or sham, what they guessed they received, and how their depression symptoms and side effects had changed. The question was not whether unblinding happened. Trial guesses were already known to be unbalanced, meaning participants could guess their assignment at better than chance. The question was whether that better-than-chance guessing was actually driving the measured antidepressant effect, or running alongside it without touching it.
What the analysis found
Receiving active tDCS was associated with correctly guessing one had received it, confirming the blind was compromised, as expected. It was also associated with more reported adverse events and with greater depressive-symptom improvement at ten weeks. The interesting part is how those three threads relate to each other. Adverse events are the most plausible mechanism by which unblinding would bias an outcome, a participant who feels a tingling sensation on their scalp infers they got the real device, and that inference itself could inflate their reported improvement through expectation. But the analysis found that participants who reported more adverse events actually improved less, the opposite direction unblinding-driven bias would predict. Using a pathway-style regression approach connecting treatment assignment, guessing accuracy, adverse events, and outcome, the authors concluded it is unlikely that unblinding explains the treatment effects observed in the trial.
Why the technique matters more than the finding
A trial team has, broadly, two honest options when unblinding shows up in the data. One is to disclose it as a limitation and move on, which is standard practice and better than ignoring it, but leaves the reader unable to distinguish a genuine effect from an inflated one. The other, harder option is what this paper does: build a specific test for whether the unblinding is biasing, meaning it distorts the outcome, or non-biasing, meaning participants can tell their assignment without that knowledge meaningfully changing what gets reported. That distinction has been implicit in how this desk has read psychedelic and device trials all year, treating a broken blind as a reason for caution rather than automatic disqualification, but rarely does a paper actually build and run the statistical test that could tell the two apart rather than leaving it to interpretation.
Why this belongs next to the FDA’s new guidance
FDA’s recently finalized psychedelic clinical trial guidance devotes real, specific attention to exactly this problem, recommending blinding questionnaires, expectancy evaluations, and central raters blind to treatment allocation, tools built to detect and characterize unblinding rather than pretend it away. This tDCS analysis is close to a working demonstration of what the next step after detection could look like: once you know a blind has failed, a specific analytic method for testing whether that failure actually moved the result. Device trials and drug trials share this problem because they share its root cause, an intervention with a subjective, hard-to-mask signature, tested against a subjective, self-reported outcome. A technique developed on a tDCS headset is, in principle, exportable to a psilocybin capsule.
The caveats
This is a single exploratory reanalysis of one trial’s data, not a validated general method, and the authors describe it as exactly that, exploratory. The specific finding, that adverse events tracked with less improvement rather than more, is somewhat counterintuitive and worth treating as this trial’s result rather than a general law about device trials. And a technique that clears one trial of unblinding-driven bias says nothing directly about a different trial with a different device, population, or effect size; each application would need its own analysis, not a borrowed conclusion.
The frame
Most trials in this general category stop at disclosure: the blind broke, here is the effect size, interpret with caution. This one asks the next question and builds a real method to answer it. Whether or not the specific pathway-analysis approach used here becomes standard, the instinct behind it, don’t just admit a blind failed, test whether the failure mattered, is exactly the discipline this desk has been applying by hand across a year of trial-by-trial reporting. Seeing a research team formalize that instinct into an actual statistical test, on a device trial rather than a drug trial, adds a genuinely useful tool to the kit for reading unblinded data across the whole neuromodulation and psychedelic-adjacent field.