Somewhere in a research clinic, a person with Parkinson's disease sits under a coil or a pair of electrodes and does not know whether anything is being delivered. Over the following weeks their tremor and their walking are measured by someone who also does not know. Some of those people improve. For decades that improvement has been treated as a nuisance to be subtracted, a smudge on the lens. A paper published on 27 February 2026 in the peer-reviewed journal Biomedicines takes the smudge seriously and asks what it is actually made of.
The study, by Aybike Reyhanli, Jorge Ortega-Marquez and colleagues with Felipe Fregni of the Spaulding Neuromodulation Center and Harvard Medical School as senior author, is a systematic review and meta-analysis: no new patients, no new stimulation, just a disciplined re-reading of trials that already exist. It followed PRISMA reporting standards, was registered in advance with PROSPERO (CRD1272381), and searched PubMed/MEDLINE, Embase, Web of Science and the Cochrane Library from inception through September 2025 for randomized, sham-controlled trials of non-invasive neuromodulation in adults with Parkinson's disease. Seventeen trials qualified, 654 participants between them, all of them using repetitive transcranial magnetic stimulation or transcranial direct current stimulation. One detail deserves flagging early, because the field's headline technology is often assumed to be in the frame: no deep-brain stimulation trial met the eligibility criteria. This is a paper about coils and scalp electrodes, and its lessons for surgery are argued, not measured.
What the pooled numbers showed is more interesting than a simple verdict. Sham stimulation was associated with small but statistically significant improvement on Part III of the Unified Parkinson's Disease Rating Scale, the part where a clinician examines movement, both immediately after the intervention and at follow-up. On the UPDRS total score, which folds in daily living and other domains, there was no significant placebo-related improvement at all. The placebo response, in other words, was not a general glow of feeling better; it landed on the motor examination and largely stayed there. It also did not swell with time. Active stimulation, by contrast, produced larger improvements that held on both the total score and Part III at follow-up. The between-group comparison for Part III favoured active stimulation just after treatment (standardized mean difference -0.68, 95% confidence interval -1.32 to -0.04), a result whose interval brushes against zero and sits on top of substantial variation between trials (I2 of 85.2 per cent), and that difference was no longer statistically significant at follow-up. The authors' own reading is careful: placebo effects contribute to short-term improvement, especially on motor examination, but "do not fully account for the sustained benefits observed with active stimulation".
For a family sitting with this, the most useful number in the paper is a boundary rather than a result. In their discussion the authors note that the minimal clinically important difference is roughly 3 to 3.5 points on UPDRS Part III and about 4 to 10 points on the total score, and that the placebo-related effects they pooled are likely below the threshold at which a change is clearly perceptible in a life. That is not a dismissal of expectation. It is a statement about scale: something measurable was happening in the sham arms, and it was probably smaller than the difference between a good morning and a bad one.
The methodological argument follows from the shape of the data. If the placebo response is outcome-specific and time-stable, then a trial's answer depends heavily on which scale it chose, when it measured, and what it compared against. The paper presses for exactly that: careful outcome selection, deliberate timing of assessments, and comparator structures built so that placebo and active effects can be pulled apart rather than netted off. The review's discussion also revisits the mirror image of the problem, the lessebo effect described by Tiago Mestre and colleagues in Neurology in 2014, in which knowing you might be assigned to placebo appears to shrink measured benefit: in that analysis, the same treatments produced improvements on UPDRS motor scores about 1.6 points smaller in placebo-controlled trials than in active-controlled ones, with wide uncertainty (95% CI 0.2 to 3.0). Expectation, that is, can cut in either direction, and a two-arm design quietly bakes both cuts into the result.
None of this is new territory for Parkinson's, which is why the field keeps returning to it. In 2001, in Science, Raul de la Fuente-Fernandez and colleagues at the University of British Columbia used PET imaging in 12 patients, studied in two groups of six, and found that placebo administration was followed by substantial release of endogenous dopamine in the striatum, the same system the disease erodes. That is a mechanism, in people, for why a sham can move a motor score. Twenty-five years later the question has shifted from whether expectation has a biology to how trials should be built around a biology they cannot switch off.