Published on 19 September 2026

Analysis · Depression · Neuromodulation

Can we predict, before the first session, whose anhedonia will respond to rTMS?

▬ Publication Translational Psychiatry · 2026, online ahead of print · Abram et al. DOI 10.1038/s41398-026-04248-3 PMID 42425953 Scientific 63 Editorial 77

Editorial collections Tomorrow Psychiatry A Study Under the Microscope PEB Decode PEB Academy

The essentials

Thirty-two patients with treatment-resistant depression completed a behavioral reward-seeking task, adapted from rodent research, before and during a course of repetitive transcranial magnetic stimulation. Among the twenty-three patients with measurements both before and at the end of treatment, the number of preferred rewards obtained before the course correlated with the decrease in the anhedonia item of the PHQ-9, with a Spearman coefficient of −0.58 and a p-value adjusted for multiple comparisons of 0.007. The measure the authors presented as primary, the gap between preferred and non-preferred rewards, does not survive this same correction, with a coefficient of −0.45 and an adjusted p-value of 0.06. The work is publicly funded, but the senior author discloses consulting work for industry and unlicensed patents related to neurostimulation. This is an observational cohort with no control group, no randomization, and no confidence interval reported around the coefficient. No clinical use follows from it today.

Context

Repetitive transcranial magnetic stimulation is part of the therapeutic arsenal for treatment-resistant depression, but response varies considerably from one patient to another, a point the authors recall to justify the search for response markers. A course represents several weeks of daily sessions, thirty-six sessions for a complete series at the center studied. Knowing beforehand who is most likely to respond is therefore a matter of a patient’s time as much as of care resources.

The prediction attempts cited by the authors mostly involve functional imaging and connectivity between the prefrontal cortex and the subgenual anterior cingulate cortex, both costly and poorly suited to routine practice. The approach explored here is different: a roughly thirty-minute behavioral task, adapted from a food-seeking paradigm used in rodents, in which the subject decides whether to wait for or forgo a reward of varying value.

Mechanism

Why measure a decision, not a symptom?

Element	What the study reports
The paradigm	A virtual reward-seeking task
FindingThe subject has a limited amount of time to move between four categories of videos: kittens, dance, landscapes and bicycle accidents. Each offer comes with a wait time of three to thirty seconds, to accept or decline. The behavior measured is willingness to wait, not a self-report of mood.
Preference	Revealed by choices, not self-reported
FindingThe ranking of the four categories is established, session by session, from the number of videos actually obtained in each. The authors also tested a ranking based on the ratings patients gave the videos: with this self-reported ranking, the association with the decrease in anhedonia disappears.
Animal homology	Adapted from rodent research
FindingThe same decisional pattern is observed in mice on the twin Restaurant Row task, which allows a back-and-forth between the animal model and the clinic. This homology is an argument for how the task was built; it does not by itself validate its predictive value.
Clinical target	Anhedonia, not depression as a whole
FindingThe hypothesis assumes that prefrontal stimulation acts on reward circuits. A behavioral measure of reward pursuit would therefore be closer to the target than an overall depression score.
What isn’t measured	No brain measurement
FindingThe document reviewed reports neither imaging nor paired electrophysiology in this cohort, and the authors further note that targeting was not guided by neuronavigation. The mechanism remains an interpretation of observed behavior, not a direct observation of the circuit.

The study at a glance

Population
32 adult patients with treatment-resistant major depressive disorder, aged 20 to 84 years, mean age 50.5 years, 18 women and 14 men, described as 100% Caucasian. Recruitment at a single site, a university interventional psychiatry clinic in a suburb of Minneapolis, Minnesota. A convenience sample, whose size reflects the number of patients who could be recruited during the study period.
Predictive measure
Behavioral Web-Surf task, completed before the course and then weekly, distinguishing preferred categories (ranks 1 and 2) from non-preferred categories (ranks 3 and 4). The number of videos obtained is normalized to session duration.
Intervention
Repetitive transcranial magnetic stimulation of the dorsolateral prefrontal cortex, five days a week, thirty-six sessions for a complete series, delivered as part of routine care. Figure-eight coil in 21 patients, targeting F3, F4 or bilateral by the Beam F3 method, at frequencies ranging from 1 to 50 Hz depending on the patient, including 15 receiving theta-burst stimulation. H1 coil in 11 patients, at 18 Hz. The protocol was set by the treating clinician and could change during the course.
Comparator
No untreated group. A community sample of 140 volunteers recruited online serves as a reference for the task’s properties, not as a therapeutic comparator. The main analysis is a longitudinal correlation within a single group.
Criteria
A single clinical instrument, the PHQ-9 questionnaire, collected as part of routine care and averaged weekly. Anhedonia is measured with a single item, the one on little interest or pleasure in doing things. Change is calculated between baseline (week 0) and the end of treatment (weeks 6 and 7).
Design
Naturalistic observational cohort with no randomization, Oxford level of evidence 3b. Spearman correlations and linear mixed models, with false discovery rate control. Main analyses limited to the first seven weeks of the course.

Quality control

Point checked	Judgment
Multiple testing correction	Applied
FindingFalse discovery rate control across families of correlations and their follow-up tests. The adjusted p-value of 0.007 accounts for the declared multiplicity, which is the minimum expected for this type of analysis. The authors themselves flag the results that do not survive this correction.
Specificity analyses	Convergent but partial
FindingThree contrasts point in the same direction: preferred versus non-preferred rewards, the anhedonia item versus the total depression score and versus the depressed-mood item, and the trajectory of gains in patients whose anhedonia improves versus the others. But the measure presented as primary, the gap between preferred and non-preferred rewards, loses significance after correction, and the ranking based on self-reported ratings does not reproduce the association.
Robustness to covariates	Verified
FindingThe correlation on preferred rewards holds after adjustment for baseline anhedonia, baseline depression, age, sex and coil type, and when the analysis is restricted to sessions of at least twenty minutes, with coefficients ranging from −0.54 to −0.62.
Independence	Senior author’s industry ties
FindingFunding is public and non-profit, mainly from the Department of Veterans Affairs, the U.S. National Institutes of Health and foundations. But the senior author discloses consulting for Abbott Laboratories and several unlicensed patents on neurostimulation and cognitive biomarkers, which cover exactly the subject of this work. The other authors report no relevant financial interest.
Sample size of the main analysis	23 subjects
FindingThe cohort includes 32 patients, but the predictive correlation rests on the 23 patients with measurements both before the course and at the end of treatment, as indicated by the reported degrees of freedom. At this sample size, a correlation coefficient is estimated with considerable uncertainty, and one or two outlying subjects can shift the result substantially.
Precision of the estimate	No interval reported
FindingNo confidence interval accompanies the correlation coefficients, although the authors do report one for the test-retest reliability of the measure. Without a bound, the reader cannot judge the range of values compatible with the data.
Recruitment	Highly selected
FindingA single specialized center in Minnesota, patients already referred for stimulation, a sample described as 100% Caucasian. The authors themselves note that the demographics of the Minneapolis suburb differ from those of the country as a whole. Transposition to other clinical populations or demographic settings has not been tested.

Results

−0.58

Spearman correlation coefficient between the number of preferred rewards obtained before the course and the change in the PHQ-9 anhedonia item at the end of the course, adjusted p-value 0.007. Twenty-three patients, no confidence interval reported.

Result	What the source reports
Measure presented as primary	Spearman −0.45, adjusted p 0.06
PEB readingThe gap between preferred and non-preferred rewards, designated by the authors as the measure of the main analyses, does not correlate significantly with the change in anhedonia after correction for multiple testing. The result that is highlighted comes from a follow-up analysis.
Correlation highlighted	Spearman −0.58, adjusted p 0.007
PEB readingIt concerns preferred rewards only. The more of them a patient obtained before the course, the more the anhedonia item decreased between baseline and the end of treatment. The coefficient is negative because a decrease in anhedonia corresponds to a negative change in score.
First specificity analysis	Preferred rewards only
PEB readingThe correlation holds for preferred rewards, coefficient −0.58, but not for non-preferred rewards, coefficient 0.23 and p 0.28. It disappears, however, when the ranking of categories is based on the ratings patients reported rather than on their choices.
Second specificity analysis	Anhedonia rather than the overall score
PEB readingNo significant correlation with the change in total PHQ-9 score, coefficient −0.35 and p 0.10, nor with the depressed-mood item, coefficient −0.16 and p 0.46. The signal is concentrated on the anhedonia item, which supports the idea of a dimensional effect rather than a general improvement. An absence of significance in a sample of twenty-three is not, however, an absence of effect.
Third specificity analysis	A trajectory of gains, not a correlation
PEB readingThis is not a correlation calculated separately in two subgroups, but a mixed model testing how gains evolved across weeks. The interaction between week and anhedonia improvement is significant, chi-square 4.65 and p 0.03: gains increase in the 18 patients whose anhedonia improves, p 0.009 and adjusted p 0.02, and do not change in the 7 patients with no improvement or worsening, p 0.43. The group with no improvement includes seven subjects, which rules out any conclusion about a non-responder profile.
Overall clinical change	PHQ-9 from 14.8 to 11.2
PEB readingMean PHQ-9 score decreased from 14.80 (standard deviation 5.90) before the course to 11.15 (standard deviation 6.99) at the end of treatment, effect size d 0.80. In the absence of a control group, this change cannot be attributed to stimulation.

Critical appraisal

Domain	Risk of bias
Absence of a comparator	Structural
FindingWithout a control arm, nothing distinguishes what is attributable to stimulation from what reflects spontaneous evolution, closer follow-up, or maintained treatments. The authors explicitly acknowledge this and write that they cannot rule out that the improvements observed reflect non-specific treatment factors.
Treatment heterogeneity	Variable protocols
FindingTwo coil types, three lateralization approaches, frequencies from 1 to 50 Hz, and a protocol sometimes modified during the course at the treating clinician’s discretion. No neuronavigation-guided targeting. What this cohort receives under the name of stimulation is not a single intervention.
Regression to the mean	Addressed
FindingCorrelating a baseline measure with a change score exposes the analysis to a classic artifact. The authors anticipated the objection: the correlation persists as a partial correlation adjusted for baseline anhedonia, coefficient −0.56 and adjusted p 0.007, and for baseline depression, coefficient −0.60 and adjusted p 0.006. The reservation therefore concerns statistical power, not an absence of control.
Outcome measurement	A single self-reported item
FindingAnhedonia is measured with a single item from a depression screening questionnaire, in patients who know they are receiving active treatment. The authors acknowledge that the PHQ-9 was not designed to study anhedonia and that this item likely conflates loss of pleasure with loss of interest.
Analytic degrees of freedom	Exploratory
FindingThe document reviewed mentions no prior registration of the analysis protocol. Several definitions of preference ranks are tested, along with several session-duration thresholds and several sets of covariates. False discovery rate control limits the declared multiplicity, not the undeclared one.
External validity	To be established
FindingA single center in the American Midwest, no independent replication, no European data. The authors themselves estimate that 73 to 133 participants would be needed to detect the effect sizes observed in their community sample. The task would require adaptation and validation before any use outside the research setting.

Level of evidence

Scientific63

Editorial77

Oxford level of evidence 3b, an observational cohort with no control group and no randomization. Confidence is reasonable regarding the existence of an association between the number of preferred rewards obtained before the course and the change in the anhedonia item in this sample, an association that withstands several adjustments. It is weaker regarding the measure the authors designated as primary, which does not survive correction for multiple testing, weak regarding the size of the link, for lack of a confidence interval, weak regarding its predictive nature, since a correlation in a single group of twenty-three subjects does not translate into individual predictive value, and absent regarding any use in practice. The authors themselves describe their marker as promising for tracking treatment response and call for larger trials, a formulation this analysis does not harden.

The colleague test

What an experienced colleague would say if shown this study in two minutes, between two consultations.

It’s an appealing idea, and I can see exactly what they’re getting at: patients who still want something before the first session tend to do better. But twenty-three patients in the analysis that counts, no control group, a single item measuring anhedonia, and the measure billed as primary that falls apart after correction. This is the kind of result that tends to shrink on replication. I’ll file it away, but it won’t change how I select patients for stimulation.

Translation for practice: no decision about referring a patient for stimulation should rest on this result. What it offers is a working hypothesis, that anhedonia may be a privileged target of prefrontal stimulation, worth watching if a replication in a larger sample appears.

What you can do with this

Change nothing in the selection of patients referred for stimulation, which remains based on the usual clinical criteria.
Note the value of assessing anhedonia separately from the overall depression score, before and after a course, with a dedicated scale. This study uses none, and its authors count that among its limitations.
Be ready to explain to a patient or family why such a result, if presented as a predictive test, is not one.
Keep in mind the distinction between a correlation observed within a group and a prediction usable for an individual, which requires validation in an independent cohort.
Watch for a replication with a larger sample. The authors themselves place the useful threshold between 73 and 133 participants, before giving this line of research a place in clinical reasoning.

Frequently asked questions

Is this test usable in consultation?

No. It is a research task lasting about thirty minutes, not validated as a clinical tool, and its individual predictive value has not been established.

Is a correlation of −0.58 strong?

It is considered moderate to strong in absolute value, but here it rests on twenty-three subjects and no confidence interval is reported. The same study conducted in another sample of this size could yield a substantially different result.

Does this mean stimulation acts on reward circuits?

That is the authors’ hypothesis, and it is consistent with the result. The study includes no brain measurement that would establish this directly.

Do non-responders have a collapsed pursuit of reward?

The source does not show this. It reports that gains in preferred rewards increase across weeks in the 18 patients whose anhedonia improves, and do not change in the 7 patients with no improvement. Seven subjects are not enough to describe a non-responder profile.

Does the result hold if patients are simply asked what they prefer?

No. When the ranking of categories is based on the ratings patients give the videos rather than on their choices, the association with the decrease in anhedonia is no longer found.

Annotated bibliography

Source study. Abram SV, McInnes AN, Sullivan CRP, Cooper DC, Sweis BM, Widge AS. Reward pursuit during a translational reward task correlates with anhedonia reductions following rTMS in patients with major depressive disorder. Translational Psychiatry, 2026. Accepted version published online ahead of print, with no volume or page numbers assigned in the document reviewed. Received October 13, 2025, revised June 8, 2026, accepted June 30, 2026. DOI 10.1038/s41398-026-04248-3. PMID 42425953. Funding: Department of Veterans Affairs (CX002355), Brain and Behavior Research Foundation (32773 and 32856), National Institute of Mental Health (R01MH136230, L40MH127601, R01MH051399-31S1, R21MH120785, K23MH112867), National Institute on Aging (R24AG065172), Burroughs Wellcome Fund Career Award for Medical Scientists, Leon Levy Scholarship, Minnesota’s Discovery, Research, and Innovation Economy Initiative, Minnesota Medical Discovery Team on Addictions. Conflicts of interest: Alik S. Widge discloses consulting for Abbott Laboratories and several unlicensed patents related to neurostimulation and cognitive biomarkers; the other authors report no relevant financial interest; Samantha V. Abram is a U.S. government employee, and the content of the article does not represent the views of the Department of Veterans Affairs. Protocol approved by the University of Minnesota ethics committee (STUDY00010052). Data available on request from the authors.

Editorial collections