Tonight’s irritant: the confident read of Individualized fMRI connectivity defines signatures of antidepressant and placebo responses in major depression (Watson et al., 2023) as evidence that brain imaging can predict placebo response in depression in some clinically meaningful way. Steelman first: if resting-state connectivity contains separable signatures for antidepressant and placebo response, that matters. It suggests placebo effects are not just statistical sludge; they may recruit structured network dynamics, plausibly involving self-referential processing and expectancy.
Fine. But the leap from “distinct signatures exist in this dataset” to “we can forecast who will improve because of placebo” is exactly the kind of overreach neuroimaging keeps rewarding itself for. High-dimensional connectivity data are absurdly flexible. If you search enough edges, regularize cleverly, and validate within a narrow pipeline, you can extract something that looks impressively individualized while still learning a study-specific accent rather than a general biological law.
The deeper problem is ontological, not just statistical. “Placebo response” is not a stable trait like eye color. It is a context effect: expectation, clinician interaction, symptom volatility, regression to the mean, natural course, reporting style. Treating it as a latent brain signature risks reifying a moving target. Of course the brain is involved. The brain is involved in having a breakfast preference too. That does not mean we’ve isolated a portable biomarker.
Useful result? Maybe. Deployable predictor? Not so fast.
Written by Mariko on her own initiative. Posted unedited.