The setup was almost cruelly simple. A coloured cue appeared for four seconds. Then, after a wait of two and a half to four and a half seconds, a 0.2 second electrical pulse. Then a number: how much did that hurt, on a scale of nought to a hundred? What the participants did not know at first was that the interesting trials were the rigged ones. A cue promising high pain, or a cue promising low pain, followed by the identical medium shock. Same current, same skin, same nerve. Different number afterwards.

That much has been shown before; it is the engine of the placebo and nocebo effects, and it is why the wording of a consent conversation is not cosmetic. The question Hsin-Yun Tsai, Ming-Tsung Tseng and colleagues at National Taiwan University set out to answer, in a paper published in the Journal of Neuroscience on 24 April 2024, was narrower and stranger: what decides how big the effect is? The dominant account is statistical. The brain is said to blend a prior (what you expect) with the evidence (what your nerves report), weighting each by its precision. A tight, consistent expectation should win the argument; a vague one should be overruled by the skin. Tsai's team built that model, fed it the data, and it failed. Neither the relative precision of the expectation nor the effect it predicted lined up with what actually happened to people's pain ratings.

What did line up was feeling. Before each shock, participants were sometimes told simply to attend to the cue, and sometimes to regulate: to deliberately turn down the emotion the cue provoked. On attend trials, the high-pain cue produced anxiety averaging 58.1 out of 100, against 33.0 for the medium cue. When participants damped that anticipatory anxiety down, the expectation's grip on their pain loosened in step. The same held on the pleasant side: turning down the relief a low-pain cue provoked shrank the pain reduction it produced. A Bayesian comparison of the two accounts favoured the emotion model over the precision model in the scanner group (individual fits favoured emotion in eight participants for negative expectations against two for precision, and in fourteen for positive expectations), and a separate behavioural experiment reproduced it, with expected posterior model probabilities of 0.906 for negative and 0.913 for positive expectations. In the brain, the anterior cingulate cortex ramped up as negative expectations amplified pain, and the right anterior insula as positive expectations eased it. Writing a Journal Club appraisal in the same journal on 8 January 2025, Chloe Cheung and Phivos Phylactou of Western University put the upshot plainly: the emotion attached to an expectation was the stronger determinant, not the consistency of it.

Hold the caveat firmly, because it is large. These were 31 pain-free volunteers, aged 20 to 31, mean age 22.6, screened for neuropsychiatric conditions, receiving brief laboratory shocks. Thirty-six started; five were excluded, one of them because he or she could not tell the three shock intensities apart. Nobody in this study had dementia. Nobody had chronic pain. This is mechanism work in young, healthy brains, and it is not a treatment finding.

And yet it lands on a sore spot in dementia care. In 2006, Fabrizio Benedetti's group in Turin reported in the journal Pain that expectation-related analgesic mechanisms appear to break down in Alzheimer's disease, and that analgesic therapies are correspondingly less effective when that component is gone: the first demonstration, they wrote, of such a disruption occurring in a clinical condition rather than being engineered in a lab by hiding the drug. Read alongside Tsai's result, that becomes a sharper question than it used to be. If the part of the expectation that does the work is the feeling it carries, then what is lost when a person can no longer hold a sentence about the future in mind long enough to feel anything about it? A 2015 review in Neurologia noted that the absence of a placebo analgesic response found in Alzheimer's disease has not been studied in other dementias, despite the extensive frontal involvement in frontotemporal and vascular disease. That gap is still there.

None of this tells a family what to do, and none of it tells a clinician what to prescribe. What it does is put a different emphasis on an old instruction. The advice has long been to explain treatment clearly, to be consistent, to make the expectation reliable. Tsai's data suggest that reliability, at least in healthy young adults under laboratory pain, was not the active ingredient. Warmth was. The look on the face of the person holding the syringe, the tone that says this will help, the presence that makes a dressing change feel survivable: on this evidence those are not the soft accompaniment to the pharmacology. They may be part of the mechanism. Which is worth knowing on a ward where the person in the bed can no longer follow the explanation but can still read the room.