Published on 16 September 2026
Virtual reality cognitive remediation: a large effect on cognition, and on daily life?
In brief
A team in Copenhagen randomised 62 adults aged 18 to 55 years, clinically stable outpatients with unipolar disorder, bipolar disorder or a psychosis spectrum disorder, all with cognitive impairment that was both objectively measured and subjectively reported. Half received four weeks of virtual reality cognitive remediation: two two-hour sessions a week with a psychologist, supplemented by training in the headset at home. The other half received an active control in the same headset, one two-hour session a week: free games and versions of the same scenarios stripped of their training elements. On the primary outcome, a cognitive score itself measured in virtual reality at week 5, the difference is large: d = 1.55, p < 0.0001, and it holds at week 17 (d = 1.53). The ability to carry out activities of daily living, rated blind by an occupational therapist observing the tasks, also improves (d = 0.80, p = 0.0088), but the difference stays below the threshold the authors regard as clinically relevant and was not measured again at week 17. The design has real strengths, an active control and blinded assessors. But 62 patients in a single centre, a primary effect of a size rarely seen for a cognitive intervention, twice as many sessions with the therapist and longer headset exposure in the active arm, and retrospective trial registration all call for replication before any firm conclusion is drawn.
The context
Cognitive impairment weighs on functional outcome in mood disorders and psychoses, and no medication corrects it. Cognitive remediation is the best-studied response, but its weak point is well known: gains made on training tasks transfer poorly to real life. Virtual reality promises to narrow that gap by exercising cognition in situations that resemble everyday life.
The question the trial asks is therefore twofold. Does virtual reality remediation improve cognition more than using the headset without targeted training, and does that benefit show up in the tasks of daily living?
The study at a glance
| Question (PICO) | |
|---|---|
| Population | |
| Adults aged 18 to 55 years, outpatients, clinically stable, Danish-speaking: unipolar or bipolar disorder in full or partial remission (HDRS-17 and YMRS scores ≤ 14), or a psychosis spectrum disorder (ICD-10 F20) judged relatively stable by the treating clinician. Objective cognitive impairment on the SCIP (at least 0.5 standard deviation below the score expected for age, education and sex, on the total score or on at least two of the five subtests) and subjective cognitive complaints (COBRA ≥ 14, CODEL ≥ 7 in at least two of the four domains). 62 patients randomised, 31 per arm: 28 with bipolar disorder, 10 with unipolar disorder, 24 with a psychosis spectrum disorder (10 schizophrenia, 11 schizotypal disorder, 3 unspecified non-organic psychosis) | |
| Intervention | |
| Virtual reality cognitive remediation (Meta Quest 2 headset) for 4 weeks: two individual 2-hour sessions a week with a clinical psychologist (8 in total), combining psychoeducation with adaptive-difficulty training in four daily-life scenarios (kitchen, supermarket, restaurant, office), plus two 30-minute home training sessions in the headset a week and real-life exercises. On average 18.0 hours completed out of a possible 20, of which 10.7 hours in the headset | |
| Comparator | |
| Active virtual reality control for 4 weeks: one individual 2-hour session a week with a clinical psychologist, consisting of free games (simple reaction time, attention, navigation in the virtual environment) and versions of the four scenarios without strategies, feedback or adaptive difficulty; no home training. On average 7.2 hours completed out of a possible 8 | |
| Outcomes | |
| Primary: change in the global functional cognitive capacity score on the CAVIR (a virtual kitchen test) from baseline to week 5, also collected at week 17. Secondary: process ability in activities of daily living (AMPS, in a test apartment, at week 5 only) and verbal learning and memory (RAVLT). Tertiary: CAVIR subtasks, neuropsychological domains and global score, functioning (FAST, UPSA-B), subjective cognitive complaints (COBRA, CODEL). Symptoms (HDRS-17, YMRS, BNSS, SAPS) in post hoc analysis | |
| Design | |
| Single-centre, parallel-group randomised trial (1:1), block randomisation stratified by age and diagnosis, double-blind (participants and assessors), active control, intention-to-treat analysis with linear mixed models, Benjamini-Hochberg correction for secondary and tertiary outcomes. Oxford level of evidence 1b. Funded by five foundations, including TrygFonden |
The quality check
| Criterion | Status |
|---|---|
| Comparator | Active, but less intensive |
| FindingSame headset and same scenarios, without the training elements, which controls for part of the habituation to virtual reality and for expectation effects. But one session a week instead of two and no home training: 7.2 hours completed against 18.0 in the active arm, of which 10.7 in the headset | |
| Blinded assessment | Planned |
| FindingAssessors were informed neither of allocation nor of treatment content, and the AMPS was rated by a blinded occupational therapist. The article does not report any check that this blinding was maintained | |
| Participant blinding | Partial |
| FindingAsked at week 17, 18 of 28 participants (64%) in the active arm identified their arm, against 9 (35%) in the control arm. Perceived usefulness for daily life was markedly higher in the active arm (77% against 48% on a visual analogue scale, p < 0.0001) | |
| Sample size | Small |
| Finding62 patients in a single centre, classic ground for an overestimated effect. The power calculation targeted 54 patients to detect a difference of 0.6 in Z score on the primary outcome | |
| Registration | Retrospective |
| FindingRegistered on ClinicalTrials.gov (NCT06038955) in September 2023, almost a year after the first participant was enrolled (October 2022) but before enrolment ended (August 2024). The article also refers, contradictorily, to preregistration | |
| Follow-up | Limited attrition at week 5 |
| FindingCAVIR data missing for 7 of 62 participants at week 5 (3 against 4) and for 13 at week 17 (total recalculated from the notes to table 2); AMPS missing for 12 participants at week 5. The figure of 93% refers to treatment adherence in the active arm (28 of the 30 who started it), not to completeness of follow-up | |
| Independence of the tools | Tools developed by the team |
| FindingThe training scenarios were developed by the authors’ group with a co-author who is a virtual reality developer, and the CAVIR was validated in publications by the same team. The funding foundations had no role in the study | |
The findings
| Outcome | Result |
|---|---|
| CAVIR, primary outcome | Week 5: effect 0.98 in Z score (95% CI 0.65 to 1.32), d = 1.55, p < 0.0001; week 17: effect 0.91 (0.61 to 1.21), d = 1.53, p < 0.0001 |
| PEB readingLarge, needs replication measured in a virtual kitchen, in the same modality as the training | |
| Activities of daily living (AMPS) | Week 5: effect 0.57 in Z score (95% CI 0.14 to 1.00), d = 0.80, p = 0.0088; that is 0.23 logit (0.06 to 0.40), against a clinically relevant difference set at 0.3 logit. Not measured at week 17 |
| PEB readingModest transfer, unknown durability rated by observation, outside the headset, analysed in 58 participants | |
| Verbal memory (RAVLT) | Week 5: effect 0.66 (95% CI 0.35 to 0.97), d = 0.99, p < 0.0001; week 17: effect 0.67 (0.29 to 1.05), d = 0.76, p = 0.0009 |
| PEB readingConsistent a conventional test, independent of virtual reality | |
| Tertiary outcomes, week 5 | 7 of 20 significant after Benjamini-Hochberg correction (12 of 20 before correction), including the global neuropsychological score and the FAST total score; both secondary outcomes remain significant after correction |
| PEB readingExploratory the effect on subjective cognitive complaints (COBRA, CODEL) does not survive correction, which does not demonstrate an absence of effect | |
The primary outcome is the most striking result, and also the most fragile. The CAVIR assesses cognition in a virtual kitchen, that is, in the very modality in which the active arm trained for four weeks, in sessions and at home. The authors point out that the test kitchen differs from the training kitchens (layout, recipes, higher difficulty) and that the control arm also practised the same scenarios; they nonetheless acknowledge that an effect of intensity, of habituation to virtual reality or of learning the test cannot be excluded. Part of the difference may therefore reflect familiarity with the task rather than a general cognitive gain. For clinical practice, the result that matters is rather the AMPS, rated outside the headset: its effect is more modest, and more credible, but it stays below the clinically relevant difference adopted by the authors (0.23 logit against 0.3), with a confidence interval that includes it, and its durability was not measured.
Critical appraisal
| RoB 2 domain | Judgement |
|---|---|
| D1, randomisation | Low risk |
| FindingBlock randomisation stratified by age and diagnosis, sequence generated by a person outside the trial and concealed from the investigator responsible for enrolment; groups balanced at baseline | |
| D2, deviations from intended intervention | Reservation |
| FindingParticipants in the active arm often able to identify their arm, twice as many sessions with the psychologist and longer headset exposure in the active arm (10.7 hours against 7.2) | |
| D3, missing outcome data | Low risk |
| FindingPrimary outcome missing for 7 of 62 participants at week 5, spread across the arms (3 against 4), and mixed-model analysis of all randomised participants | |
| D4, measurement of the outcome | Reservation |
| FindingBlinded assessors, but primary outcome measured in the same virtual modality as the training | |
| D5, selection of the reported result | Reservation |
| FindingRetrospective registration, though it took place before enrolment ended, and a protocol published in 2024; the documents consulted do not allow the registered primary outcome to be compared with the one reported | |
Level of evidence
PEB appraisal: moderate confidence in the direction of the effect, low confidence in its size. The design is better than that of the average cognitive remediation trial, with a genuine active control. What the trial shows is an improvement on a cognitive score measured in virtual reality and on an observer-rated functional scale. What remains to be established is the true size of these effects across several centres and in less selected patients, and how much of them is due to targeted training rather than to time spent in the headset and with the therapist.
The colleague test
What an experienced colleague would say if you put this study to them in two minutes, between two consultations.
“ A d of 1.55 in cognitive remediation, in 62 patients, I don’t take that at face value. There is a genuine active control, even if it is less intensive, and the gain on activities of daily living is what interests me. If I worked in rehabilitation, I would want to try it. In an office practice, it changes nothing on Monday. ”
What this means in practice: an encouraging result for rehabilitation services, one that justifies multicentre trials against a control of equal intensity, but not yet grounds for telling patients that a benefit to their daily life is established.
What you can do with this
- Look for cognitive impairment, both objective and subjective, in stabilised patients who complain of it, in bipolar disorder and depression as well as in psychosis spectrum disorders: this is the profile the trial selected.
- Refer to a rehabilitation service offering cognitive remediation when this impairment weighs on daily life, whether the modality is virtual or not.
- Present virtual reality as a promising and still experimental modality, not as an established treatment.
- Read the next trials with three questions in mind: does the effect hold across several centres, does it hold on measures taken outside the headset, and does it hold against a control of equal intensity?
Frequently asked questions
Is virtual reality cognitive remediation available in routine clinical practice?
Not in routine practice. For now it belongs to specialised services and research protocols.
Why be wary of such a large effect size?
Because small trials often produce overestimated effects, and because the primary outcome is measured in the same virtual modality as the trained task. The more modest effect on activities of daily living is more plausible.
Did the control group in the virtual reality trial really do something comparable?
Partly. Same headset and same scenarios, but without strategies, feedback or adaptive difficulty, and supplemented by free games. Above all, only one two-hour session a week instead of two, with no home training: 7.2 hours completed against 18.0, of which 10.7 in the headset. At the end of the study, 64% of participants in the active arm identified their arm, against 35% in the control arm.
Does virtual reality cognitive remediation work equally for bipolar disorder, depression and schizophrenia?
The trial enrolled bipolar disorder (28 patients), unipolar disorder (10) and psychosis spectrum disorders (24) together, on the basis of cognitive impairment. Exploratory analyses find an effect on the primary outcome in both subgroups (mood disorders, psychosis spectrum disorders), but the effect on the AMPS does not reach significance in the 38 patients with a mood disorder (p = 0.069), which does not allow an absence of effect to be concluded. The authors acknowledge that the trial was underpowered for these subgroups: it cannot say whether the effect differs from one diagnosis to another.
Annotated bibliography
Source study. Jespersen AE, Lumbye A, Vinberg M, Glenthøj LB, Nordentoft M, Bruun CF, von Bülow C, Makransky G, Wæhrens EE, Miskowiak KW. Virtual reality-based cognitive remediation versus virtual reality control in people with mood or psychosis spectrum disorders in Denmark: a single-centre, double-blind, randomised controlled trial. The Lancet Digital Health. 2026 (June); 8(6): 101002, volume, issue and article number established from the Crossref and PubMed records. DOI: 10.1016/j.landig.2026.101002 · PMID 42106222, identifier established from the PubMed record on 10 September 2026. Open access article. Funding: TrygFonden (grant 150128), Axel Muusfeldts Foundation (2022-0097), Jascha Foundation (2022-0134), Ivan Nielsen Foundation (grant number not available) and Familien Hede Nielsen Foundation (2023-1741); the funders had no role in design, data collection, analysis, interpretation or writing. Declared competing interests: K. W. Miskowiak, honoraria from Lundbeck, Angelini, Gedeon Richter and Janssen-Cilag, and funding from the European Research Council, Innovations Fund Denmark, Independent Research Fund Denmark, the Svend Andersen Foundation and the KID Foundation; M. Vinberg, honoraria from Lundbeck, Eli Lilly and Johnson & Johnson, travel support and participation on a data safety monitoring board or advisory board for Johnson & Johnson, unpaid roles at the Danish Psychiatric Society Research Chapter, the Danish Chapter International Society for Bipolar Disorders and in two ECNP bodies; L. B. Glenthøj, honoraria from Lundbeck, Heka-VR and the mental health services of the Capital Region of Denmark, consulting fees from Boehringer Ingelheim, and funding from Independent Research Fund Denmark, the Lundbeck Foundation, Danish Life Science Strategy, the Danish Parliament budget, the Sofus Carl Emil Friis and Wife Foundation and EIT Health; C. von Bülow, royalties from Munksgaard Publishing; E. E. Wæhrens, funding from the Oak Foundation (OFIL-24-07), royalties from Munksgaard Publishing and membership of the REHPA advisory board; A. E. Jespersen, honoraria from Lundbeck; no competing interests declared by the other authors. Registration: ClinicalTrials.gov NCT06038955, completed retrospectively in September 2023; protocol published in Trials (2024; 25(1): 82, DOI 10.1186/s13063-024-07910-7); approved by the Committee on Health Research Ethics of the Capital Region of Denmark (H-22004153). The authors’ supplementary appendix was consulted.