100-participant RCT of commercial VR cognitive training (Beat Saber) in chronic TBI: null primary outcome on sustained attention, secondary gains in processing speed, executive function, and quality of life

Background

Traumatic brain injury (TBI) is a leading cause of disability worldwide, and impaired attention, processing speed, and working memory are among the most common cognitive sequelae. Attention also underpins memory and executive functions, so even small impairments can affect social or occupational participation. The INCOG 2.0 guidelines identify attention training as an important rehabilitation goal, but historically much of cognitive training has been delivered via decontextualized computerized exercises whose clinical relevance has been questioned for lack of ecological validity. Virtual reality has emerged as a possible alternative because of its immersive capabilities and the engagement that gamified content provides - commercial games like Beat Saber and Fruit Ninja have been explored as cognitive training media because they place demands on response speed, vigilant attention, and concurrent target tracking. However, TBI-specific evidence has been limited, with prior trials characterized by small samples and outcome measures that closely resembled the training tasks.

The Johansen trial set out to address that gap with a large, methodologically rigorous parallel-group RCT testing 5 weeks of commercial-VR-game training (Beat Saber on Meta Quest 2) against a nonspecific counseling-and-activity-suggestion control condition in adults in the chronic phase of TBI, with sustained attention as the pre-registered primary outcome.

What the researchers did

One hundred participants aged 18-65 with radiologically verified complicated mild-to-severe TBI and documented objective impairments in sustained attention, processing speed, or working memory were enrolled from Norway’s largest rehabilitation hospital (Sunnaas) between October 2022 and September 2024. Randomization with block sizes of 4-6 was generated by a statistician not actively involved in the study; an allocation error resulted in 51 VR and 49 control. The trial was registered at ClinicalTrials.gov (NCT05443542) and on Open Science Framework (osf.io/6gphy); the protocol was pre-published (Johansen et al. 2024, Trials).

VR intervention (n=51). Unsupervised home-based use of the commercial rhythm game Beat Saber on an Oculus Quest 2 head-mounted display: 30 minutes per day, 5 days per week, for 5 weeks. The VR group received tutoring in equipment management before starting. Weekly phone calls covered technical difficulties, game-difficulty adjustment, adverse events, cybersickness symptoms, fatigue (Fatigue Severity Scale), and motivation (Situational Motivation Scale).

Control intervention (n=49). Nonspecific counseling plus an information booklet with activity suggestions resembling general advice on everyday activities (compensatory strategies, energy conservation, nutrition, sleep hygiene, everyday activities, physical activity, hobbies, games). Participants chose 1-2 themes to work on; no specific time recommendation was given. Most frequently chosen themes: physical activity and energy-conservation techniques. Weekly phone calls covered fatigue and motivation. This was NOT active cognitive rehabilitation - the authors flag the control as nonspecific and heterogeneous in their limitations section.

Outcomes. Primary: Coefficient of Variation (CoV) on Conners Continuous Performance Test-3 (CPT-3) as a measure of sustained attention. Lower CoV = better. Secondary: CPT-3 Hit Reaction Time (HRT, processing speed); CPT-3 commissions and omissions (attentiveness/impulsivity); WAIS-IV Digit Span Backward and Sequencing (working memory); BRIEF-A self-report and informant-report (executive function); Patient Competency Rating Scale (everyday functioning); Fatigue Severity Scale; QOLIBRI (quality of life). Assessments at baseline (T1), post-intervention 5 weeks (T2), and 16 weeks post-baseline (T3, primary endpoint).

Analysis. Intent-to-treat using linear mixed-effects models with time, time-by-group interaction as fixed effects and random intercept; estimated mean changes with 95% confidence intervals; alpha 0.05 two-sided. Outcome assessors and the statistician were blinded; the only person who knew the allocation sequence was the principal investigator.

Sample size. Powered for the primary outcome (CoV) with 3% mean change difference, 5% SD, 80% power, 5% alpha, 10% attrition: 45 per group required, 50 per group enrolled.

What they found

Primary outcome - sustained attention (CPT-3 CoV). NO significant between-group difference at 5 weeks (mean difference -0.01, 95% CI [-0.02, 0.01], p=.743) or 16 weeks (mean difference 0.01, 95% CI [-0.01, 0.02], p=.473). BOTH groups showed within-group CoV reductions from baseline at both timepoints (VR all p≤.001; control all p≤.001), but the reductions did not differ between groups.

Processing speed (CPT-3 HRT). At 16 weeks, there was a significant between-group difference (mean difference 14.67 ms, 95% CI [1.07, 28.26], p=.035), but in an unexpected direction: VR reaction times became LONGER (+5.30 ms from baseline) while control reaction times became SHORTER (−9.37 ms from baseline). Neither within-group change was significant on its own.

Commission errors. At 16 weeks, the VR group made significantly fewer commission errors than the control group (mean difference −7.87, 95% CI [−12.11, −3.63], p<.001). Within-group reduction was −9.60 in VR vs −1.73 in control. The authors note commission reduction was the only outcome to reach their pre-defined clinically relevant change (0.5 SD from baseline).

Working memory. No significant between-group or within-group differences on WAIS-IV Digit Span Backward or Sequencing at any timepoint.

Self-reported executive function (BRIEF-A). Significant between-group differences at 16 weeks favoring VR on Metacognitive Index (mean difference −4.54, p=.003) and Global Executive Composite (mean difference −6.08, p=.017). Effects were already present at 5 weeks (MI p=.003; GEC p=.023). However, INFORMANT BRIEF-A scores (rated by a close-other) did NOT show a between-group difference at 5 weeks (p=.667) or 16 weeks (p=.219). The self-reported gains were not corroborated by people who know the participant - a pattern the authors explicitly flag in their interpretation.

Quality of life (QOLIBRI). Significant between-group difference at 16 weeks favoring VR (mean difference 4.52, 95% CI [0.23, 8.80], p=.039), but NOT at 5 weeks (p=.369).

Everyday functioning (PCRS). No significant between-group differences at 5 weeks (p=.167) or 16 weeks (p=.132).

Fatigue (FSS). No significant differences at any timepoint.

Post-hoc Inverse Efficiency Score (IES). Combining speed and accuracy, the VR group had significantly lower (better) IES than control at both 5 weeks (between-group p=.035) and 16 weeks (between-group p=.003). The authors interpret this as objective evidence of a speed-accuracy shift - the VR group is sacrificing some speed for substantially better accuracy.

Adverse events and tolerability. Twenty of 51 VR participants reported mild cybersickness symptoms (temporary dizziness, eyestrain, fatigue, mild headaches), most resolving in weeks 1-2 (n=8 in week 1, n=7 in week 2). One VR participant withdrew due to unspecific adverse events, one reported VR-related fatigue increase, and one experienced reoccurrence of benign paroxysmal positional vertigo. Weekly motivation and fatigue did not differ between groups across timepoints.

Why this matters

This is the largest RCT to date investigating commercial-VR-game training for cognitive rehabilitation after TBI. The pre-registered primary outcome of sustained attention was null - the VR intervention did NOT outperform the nonspecific counseling control on CoV. However, the secondary findings - longer reaction times paired with significantly fewer commission errors, better self-reported executive functioning and quality of life, and a significantly lower (more efficient) inverse-efficiency score - converge on the authors’ interpretation that the VR group developed an improved speed-accuracy tradeoff, prioritizing precision over speed. The authors argue this reflects a shift in executive strategy rather than a change in basic attentional capacity.

Three important caveats for clinical interpretation:

1. The control condition was nonspecific advice, not active cognitive rehabilitation. The authors explicitly flag this as a heterogeneous control that may have affected the between-group comparison. The trial does NOT show that commercial-VR-game training outperforms structured evidence-based cognitive rehab; it shows that it outperforms a counseling-and-booklet condition.

2. Self-reported gains were not corroborated by informants. The BRIEF-A self-report showed significant between-group differences favoring VR; the BRIEF-A informant-report did NOT. The authors note that blinding of participants was not possible, so demand characteristics may inflate self-reports. The objective CPT-3 commission and IES findings provide independent support for the speed-accuracy interpretation, but the self-vs-informant discrepancy is worth flagging.

3. Functional communication outcomes were not assessed. If cognitive-communication is the clinical target (e.g., for an SLP working with someone post-TBI), this trial does not directly inform that goal.

For Therapy withVR specifically: this trial used Beat Saber on Oculus Quest 2 as a commercial off-the-shelf cognitive training intervention. Therapy withVR is a clinician-controlled speech-language therapy platform - a different design, different population focus, different delivery model. The Johansen trial is included in the Evidence Hub because it adds to the broader immersive-VR-in-rehabilitation evidence base for adults with neurogenic conditions, not because it relates to Therapy withVR.

Limitations

The authors explicitly flag the following:

• Modest baseline neuropsychological deficit. Group averages were within 1 SD of normative means on most baseline cognitive measures, which may imply participants are not fully representative of the broader TBI population.
• Blinding of participants not possible. Self-reported findings may be influenced by social desirability bias and demand characteristics. The self-vs-informant BRIEF-A discrepancy is consistent with this concern.
• Nonspecific control condition. Counseling and activity-suggestion booklet rather than active cognitive rehabilitation; the heterogeneous nature of the control might have affected the between-group comparison.
• Adherence not formally measured. VR equipment did not record valid duration-of-use data; control participants were not given a recommended amount of time. Dosage understanding is incomplete in both arms.
• Multiple secondary outcomes / Type I error risk. The authors explicitly caution: “Given the number of secondary outcomes assessed, caution is warranted in interpreting the positive findings, as the potential for errors increases with multiple comparisons.”
• Cybersickness. 39% of VR participants reported mild symptoms; most resolved in weeks 1-2 but the prevalence is non-trivial.
• Allocation error resulted in 51/49 rather than 50/50; this likely had a negligible effect on the analysis but is a deviation from the pre-registered plan.
• COI to disclose. Co-author Alexander Olsen owns stocks and a pending patent application and is the cofounder/owner of Nordic Brain Tech AS (paper notes none of this is related to the present work). Co-author Dawn Neumann reports multiple HHS/ACL/NIDILRR, DoD/CDMRP, and VA funding/payments. The other 11 authors have nothing to disclose. These relationships are surfaced for transparency.