Mismatch Negativity: A Prediction-Error Index That Flags iTBS Responders in Depression
- In 46 patients with major depression and 64 healthy controls, baseline mismatch-negativity (MMN) amplitude did not differ between groups and did not correlate with symptom severity, so it does not work as a diagnostic marker of depression.
- In the active stimulation arm, a larger baseline MMN amplitude predicted greater symptom reduction after two weeks of dorsomedial prefrontal iTBS: r = 0.48, p = 0.021 on the self-rated MADRS-S and r = 0.46, p = 0.026 on the BPRS affective subscale.
- No such correlation appeared in the sham arm, and MMN amplitude itself did not change after stimulation, so the signal is a stable baseline moderator rather than a target-engagement readout.
- The dissociation, prediction present in the active but not the sham arm, points to MMN as a treatment-specific plasticity gauge rather than a generic prognostic or placebo-response marker.
Neurostimulation for depression works, but only for a fraction of patients, and today we commit them to a two-to-six-week course before we learn whether they are in that fraction. A cheap, passive, pre-treatment index that separates likely responders from likely non-responders is one of the field's most wanted tools. This study, from a Swedish group in the Journal of Affective Disorders, proposes an unlikely candidate: the mismatch negativity, the oldest and best-characterised prediction-error signal in cognitive neuroscience.
What the numbers say
The mismatch negativity is elicited by a passive auditory oddball: a stream of identical "standard" tones is occasionally interrupted by a "deviant" that differs in duration, and the brain emits a frontocentral negative deflection roughly 150–250 ms later. No task, no attention, about fifteen minutes of EEG. In predictive-coding terms the MMN is the precision-weighted prediction error the auditory cortex broadcasts when an incoming tone violates the regularity it had learned, and its generation is NMDA-receptor-dependent, which is why it doubles as a proxy for cortical plasticity.
The plasticity link is the whole point here. Intermittent theta-burst stimulation (iTBS) is thought to act through long-term-potentiation-like plasticity, so the authors reasoned that a brain still capable of vigorous prediction-error signalling should have more plastic reserve for the protocol to mobilise. The data fit: within the group that received two weeks of twice-daily prolonged iTBS over the dorsomedial prefrontal cortex, a larger baseline MMN predicted a bigger drop in depressive symptoms (r = 0.48 on MADRS-S, r = 0.46 on the BPRS affective subscale). That is a moderate correlation, accounting for roughly a quarter of the outcome variance. Crucially, the same relationship was flat in the sham arm.
Two negative results sharpen the picture rather than weaken it. Baseline MMN did not distinguish patients from controls, unlike the reliably reduced MMN seen in schizophrenia, so this is not a diagnostic aid. And MMN amplitude did not shift after treatment, so it is a trait-like starting condition, not a readout of whether the stimulation "took." What the index captures is a patient's plastic reserve going in.
Where it fits in practice
The natural use is pre-treatment stratification: a fifteen-minute passive EEG before starting a plasticity-dependent stimulation course, read as one input among clinical predictors, not a gate. A clinician or trial designer could plausibly use it to enrich a cohort for likely responders or to flag patients who may need an alternative first. But clinical readiness is not here yet. This is a pilot, with the 46 patients split across active and sham arms, so the active correlation rests on roughly two dozen people. There is a single moderate correlation, no validated cutoff, no AUC, sensitivity or specificity, and no replication. The target was the dorsomedial prefrontal cortex, not the standard left dorsolateral site used in most rTMS for depression, so the finding cannot yet be transported to a conventional protocol. Treat it as a research-grade moderator worth measuring in the next trial, never as a reason to withhold treatment from a patient with a small MMN.
A cortex still emitting a vigorous prediction error has more plastic reserve for iTBS to mobilise, and its mismatch negativity shows it before the first pulse.
Small single-site pilot with the 46 patients divided across active and sham arms, a single moderate correlation (r ≈ 0.48) with no validated cutoff, AUC or replication, and a dorsomedial prefrontal target that differs from standard left-DLPFC rTMS, so generalisation is unproven.