The ritual, repeated 32 times across half a year in a neurology department in Madrid, took about 53 minutes. First a figure-8 coil, guided by neuronavigation onto the left dorsolateral prefrontal cortex, delivered 600 pulses at 50 Hz in bursts of three, roughly three minutes at 120 percent of resting motor threshold. Then, immediately, 50 minutes of therapy: five nouns per session, drawn from a list of 261 words across eight categories (animals, foods, objects, places, furniture, clothing, transports, body parts), each word one the person had already failed to name twice before the trial began. Homework followed, copying and recalling. Ten of these sessions in the first fortnight, then one a week for 22 weeks. Half the room did not know whether the coil above their scalp was live. Nor did the person assessing them.
That is the RECONNECT trial, run at Hospital Clinico San Carlos, published in JAMA Network Open on August 11, 2025 by Lucia Fernandez-Romero and colleagues, registered as NCT05842473. Sixty-three people with primary progressive aphasia took part (24 nonfluent, 12 semantic, 27 logopenic), mean age 71.8 years, all with a Clinical Dementia Rating of 0 or 1, randomised two to one to active or sham intermittent theta-burst stimulation. It matters because of a gap in the field: previous stimulation studies in this syndrome ran for days or weeks, and the authors state that longer-term effects had not been examined. Six months of blinded, sham-controlled stimulation in a rare, relentlessly progressive language disorder is genuinely new evidence, publicly funded (Instituto de Salud Carlos III and European Union grants), with no conflicts of interest reported and no device manufacturer named as sponsor.
And here is what it bought. The primary endpoint was not speech. It was the standardised uptake value ratio on FDG-PET, a measure of glucose metabolism, in a left-hemisphere region tied to disease progression. At six months the adjusted mean was 0.78 (95% CI, 0.77 to 0.79) in the active group and 0.77 (95% CI, 0.75 to 0.78) in the sham group. The adjusted difference was 0.02 (95% CI, 0.01 to 0.03), F = 4.17, P =.046, partial eta squared 0.08. Not a tenth of a point; two hundredths, with a confidence interval whose lower bound touches 0.01 and a P value that clears the threshold by four thousandths. The direction is less metabolic decline in the stimulated group. The magnitude is a scan finding, and nobody in this trial reported that a scan finding was audible at a kitchen table.
The secondary outcomes are where the human interest sits, and they are secondary. On the Mini-Linguistic State Examination, a 100-point language battery, the adjusted six-month means were 79.06 for active and 71.35 for sham, an adjusted difference of 7.71 points (95% CI, 3.06 to 12.35; P =.002), the sham group having worsened. Confrontation naming of the trained words favoured active stimulation by 23.81 points (95% CI, 11.85 to 35.77), a gain on the specific nouns that were drilled, which is not the same thing as finding words in conversation. Daily-living function and neuropsychiatric symptoms both moved in the active group's favour by modest margins. Words per minute in spontaneous speech did not separate at all (3.41; 95% CI, -8.86 to 15.68). That last null is the one to sit with: the outcome closest to talking freely is the one that did not move.
Safety and stickiness were the quiet successes. Adherence was complete in 58 of 63 participants, 92.1 percent, across six months of hospital visits by people whose disease was advancing throughout. There were no significant differences in adverse events between groups, and the two serious adverse events both occurred in the sham arm and were judged unrelated to it. Two people in the active group declined the final PET scan, three withdrew early, and the missing data were not imputed; the analysis was a modified intention-to-treat. The paper carries a correction published November 19, 2025, fixing y-axis labels in one figure. And the registry entry, first posted on May 6, 2023, arrived after the enrolment window the paper describes as beginning in December 2021, with ClinicalTrials.gov listing an actual start of February 1, 2023: a discrepancy worth naming in a trial whose whole value is its rigour.
So, should a clinic buy a coil? The trial's authors decline to make that leap, and their restraint is the most useful sentence in the paper: whether the imaging changes are "clinically meaningful remains unknown". They also note that functional communication measures, the kind that capture whether someone can order coffee or argue with a grandchild, were not used, and belong in future studies. This is one site, one target, one stimulation protocol, 63 people, six months, with a metabolic surrogate as the thing being proved. For a Prevention reader the honest translation is narrow: nothing here prevents primary progressive aphasia, and nothing here yet shows that adding magnets to word therapy changes how a person's language holds up in their own house.