VR Biofeedback for Sleep and Emotional Distress

A new emerging research study in the Journal of Medical Internet Research examined whether virtual reality (VR)-based biofeedback could improve sleep quality in adults with depressive symptoms, anxiety symptoms, or both over a 4-week period. The question is timely. Sleep disturbance is not a decorative symptom in mood and anxiety presentations; it is often part of the engine. When sleep becomes fragmented, delayed, or chronically nonrestorative, emotion regulation usually gets shakier, stress reactivity tends to climb, and day-to-day resilience becomes much harder to maintain.

This is one reason biofeedback remains so interesting. In broad terms, biofeedback refers to training that gives a person real-time information about physiological processes such as breathing, heart rate, temperature, or electrodermal activity so they can learn to regulate them more effectively. Neurofeedback is the EEG-based subset of that family, where the feedback signal comes from brain activity. In both cases, the larger clinical idea is self-regulation: giving the nervous system practice, not just advice.

What makes this paper especially relevant for the neurofeedback and biofeedback world is that it does not only ask whether regulation training helps. It asks whether the delivery environment matters. Can VR make the intervention more engaging, more immersive, or more effective than a conventional therapist-guided biofeedback session? That is not a trivial question. In real clinics, adherence and attentional engagement are not side issues; they are often the difference between a protocol that looks elegant on paper and one that a tired, stressed human can actually sustain.

The headline finding is appealing but nuanced: both active interventions improved subjective sleep quality in the symptomatic group, but VR was not clearly superior to conventional biofeedback. That makes this a promising paper, though not quite a “VR wins” paper.

Methods

This was a single-center, 3-arm, open-label randomized controlled trial conducted at Samsung Medical Center in South Korea. Adults who endorsed depressive symptoms, anxiety symptoms, or both were placed into the DAS cohort if they reported subjective depression or anxiety, had not received psychiatric treatment in the past 6 months, and scored at least 10 on the PHQ-9 or at least 9 on the PDSS. Individuals with psychosis, bipolar disorder, personality disorders, recent substance abuse, major neurological disease, serious medical illness, brain injury, or prior psychiatric medication exposure were excluded. A healthy control cohort was also recruited to broadly resemble the symptomatic VR group in age, sex, and education.

A total of 120 participants were enrolled, and 118 were included in the final analysis after 2 early dropouts. The DAS cohort was randomized 1:1 into either a VR-based biofeedback group (n=40 analyzed) or a conventional biofeedback group with a therapist (n=38 analyzed). The healthy control group (n=40) received the VR intervention.

Participants attended intervention visits at weeks 0, 2, and 4. Sleep was assessed with the Pittsburgh Sleep Quality Index (PSQI) at baseline and week 4. The PSQI yields a global score as well as seven component scores: subjective sleep quality, latency, duration, efficiency, disturbance, sleep medication use, and daytime dysfunction.

The VR intervention used a Samsung Odyssey Plus headset with stereo audio while participants sat in a motion chair. Each session lasted 5 minutes and presented four nature scenes paired with calming music and environmental sounds. A psychiatrist guided slow breathing and relaxation during the experience. The script emphasized muscle relaxation, slow abdominal breathing, and gradual exhalation while participants moved through virtual natural environments such as a river crossing, open sky, and meadow.

The conventional biofeedback condition also lasted 5 minutes and used the ProComp Infiniti system. Participants viewed physiological signals on a screen with a therapist, including heart rate or blood pressure, skin conductance, respiration, and skin temperature. The therapist delivered feedback when physiological markers changed by 15%.

One methodological wrinkle deserves attention. The manuscript describes the VR arm as “VR-based biofeedback,” but the protocol description emphasizes immersive relaxation, guided breathing, and calming sensory content more than explicit real-time physiological feedback embedded into the VR environment itself. That does not invalidate the intervention, but it does complicate mechanistic interpretation. The safest reading is that this was an immersive VR relaxation-biofeedback package rather than a tightly specified closed-loop physiological feedback system.

Results

The baseline sleep burden was clearly heavier in the symptomatic groups than in the healthy control group. Mean baseline global PSQI scores were 9.70 (SD 2.49) in the DAS/VR group, 10.76 (SD 2.76) in the DAS/BF group, and 5.85 (SD 2.39) in the healthy VR group. Since PSQI scores above 5 are generally interpreted as poor sleep quality, both symptomatic groups began the study in a clearly disturbed range.

Over 4 weeks, both active interventions in the symptomatic cohort produced significant within-group improvements in global PSQI score. The DAS/VR group improved from 9.70 to 7.20, a mean reduction of -2.50 points (SD 2.89, P<.001). The DAS/BF group improved from 10.76 to 7.37, a mean reduction of -3.39 points (SD 2.80, P<.001). The healthy control VR group also improved, though more modestly, from 5.85 to 4.90, a mean reduction of -0.95 points (SD 2.09, P=.01).

At the component level, the symptomatic groups showed significant gains in subjective sleep quality, sleep latency, sleep disturbance, and daytime dysfunction. Sleep duration and sleep efficiency did not significantly improve in either symptomatic intervention arm, and medication use was essentially unchanged. That pattern is clinically interesting. It suggests the intervention may have primarily influenced how quickly participants settled, how restless sleep felt, and how impaired they felt during the day, rather than clearly extending sleep time or improving efficiency.

The main head-to-head question was whether VR outperformed conventional biofeedback in the symptomatic cohort. Statistically, it did not. The difference in change in global PSQI between DAS/VR and DAS/BF was not significant after adjusting for age and sex (P=.14). Sleep disturbance improved by -0.58 (SD 0.75) in DAS/VR and -0.66 (SD 0.75) in DAS/BF, again with no significant adjusted between-group difference (P=.49).

When DAS/VR was compared with healthy controls receiving the same VR intervention, the symptomatic group showed significantly greater improvement in sleep disturbance (adjusted P=.0014) and global PSQI score (adjusted P=.01). That sounds impressive, but it also likely reflects greater baseline impairment and therefore more room to improve.

Adverse effects in the VR group were generally minor: visual discomfort, difficulty adjusting the headset, mild dizziness, neck strain, and occasional anxiety triggered by certain scenes such as height-related environments.

Discussion

The central clinical message of this paper is encouraging and slightly humbling at the same time. Encouraging, because three very brief sessions delivered across 4 weeks were associated with meaningful improvements in subjective sleep quality among adults with depressive symptoms, anxiety symptoms, or both. Humbling, because the flashier technology did not clearly beat the simpler therapist-guided comparator.

That result matters. In many corners of digital mental health, there is an understandable temptation to assume that more immersive delivery automatically means stronger clinical outcomes. This trial does not support that assumption. It suggests that the active ingredients may lie less in the headset itself and more in the broader regulatory package: slowed breathing, guided relaxation, autonomic downshifting, repeated practice, therapist structure, expectancy, and the experience of doing something purposeful for one’s nervous system.

At the same time, VR should not be dismissed. The study was not a failure for VR; it was a reminder that engagement and superiority are not the same thing. In actual practice, a person who struggles to remain present during standard training may do much better inside an immersive environment that captures attention and dampens external distraction. For some clients, that may improve adherence or willingness to return, even if the average group outcome is not statistically better than conventional biofeedback.

The methodological limitations, however, keep the interpretation on a short leash. First, the trial was open-label. Participants knew what kind of intervention they were receiving, and the main sleep outcome was subjective. That creates ample room for expectancy effects, demand characteristics, and nonspecific improvement. Second, the symptomatic cohort was defined by screening thresholds rather than formal diagnostic interviews establishing major depressive disorder or an anxiety disorder for the purposes of this outcome paper. Third, participants were self-referred and medication-free, which likely produced a relatively selective and motivated sample rather than the full complexity of everyday psychiatric practice.

Fourth, and perhaps most important for mechanism, the VR condition is not described in a way that makes the feedback loop fully transparent. The conventional biofeedback arm clearly displayed physiological parameters and used therapist feedback tied to measurable changes. The VR arm, by contrast, appears to have centered on guided breathing and immersive relaxation scenes. That may still be clinically useful, but it makes the study harder to interpret as a clean test of “VR-based biofeedback” versus conventional biofeedback. It may be more accurate to think of it as immersive regulation training compared with standard physiological feedback training.

There is also the issue of outcome specificity. The PSQI is valuable and widely used, but it remains a self-report measure. Without actigraphy, polysomnography, or at least some objective sleep-wake estimate, we cannot know whether the intervention changed sleep architecture, total sleep time, awakenings, or sleep efficiency in a physiological sense. What we can say is that participants reported better sleep, faster sleep onset, less disturbance, and less daytime dysfunction. That is clinically meaningful, but it is not the same as demonstrating objective sleep normalization.

For clients and referring clinicians, the practical take-home is fairly straightforward. This kind of intervention may represent a low-burden, nonpharmacological adjunct for people whose sleep is entangled with emotional distress, especially if they are reluctant to begin medication or are looking for a body-based complement to psychotherapy. The intervention appears safe overall, though not completely frictionless: headset comfort, dizziness, visual strain, and scene-specific anxiety should be screened for rather than discovered the hard way.

For neurofeedback and biofeedback professionals, the paper offers a useful design lesson. Regulation protocols do not operate in a vacuum. Sensory framing, attentional load, therapist presence, and tolerability are part of the intervention, not decorative packaging around it. This is particularly relevant for sleep-oriented work, where over-effort is often the enemy. A protocol that quietly facilitates parasympathetic settling may be more valuable than one that looks technologically sophisticated but demands too much cognitive effort.

There is also a deeper conceptual point here. Sleep-focused regulation work may benefit from targeting the nervous system’s transition dynamics rather than sleep itself in a narrow sense. This study’s strongest changes were not in duration or efficiency, but in sleep latency, disturbance, and daytime dysfunction. That pattern fits the idea that these brief interventions may help people enter a less hyperaroused state, rather than transforming all aspects of sleep physiology. For many clinically distressed clients, that is not a minor shift. It may be the exact entry point that makes broader treatment possible.

Future trials would be stronger with longer follow-up, objective sleep measures, tighter diagnostic characterization, and a clearer physiological description of the VR feedback loop. It would also be useful to know whether home practice, bedtime timing, or higher session dose changes the picture. Right now, the paper supports cautious optimism rather than technological triumphalism.

Brendan’s perspective

My first instinct with a client like this is usually the path of least resistance: start with HRV biofeedback before moving to neurofeedback. Not because neurofeedback is less valuable, but because HRV work is often faster to set up, easier to explain, and surprisingly effective for a large subset of poor sleepers, especially those whose insomnia is driven by hyperarousal, worry, shallow breathing, somatic tension, or that familiar “tired but internally still on” feeling. A few sessions of breathing-paced autonomic training can sometimes shift sleep latency, nighttime settling, and next-day stress reactivity more quickly than expected. In practice, that matters. When clients get an early win, even a modest one, their motivation changes. They stop feeling like their body is a mystery that has betrayed them. If HRV-based training produces a clear response, wonderful: we may not need to complicate the treatment. But when the body settles somewhat and sleep still remains fragmented, inconsistent, or strangely effortful, that is often when neurofeedback becomes the missing piece. In other words, I do not see HRV biofeedback and neurofeedback as competitors. I see them as sequenced tools. Start with the easiest lever that often works. If that lever is not enough, then brain-based training may be exactly what helps the client move from partial regulation to real stability.

Translating this study into EEG neurofeedback for sleep

If I were translating the spirit of this paper into EEG neurofeedback practice, I would think less in terms of “sleep protocol” as a fixed recipe and more in terms of the physiology standing between the client and sleep onset. For the anxious, cognitively overactivated sleeper, I am often thinking about excessive fast activity, poor inhibitory settling, and difficulty transitioning out of a vigilant cortical mode. Clinically, that may point toward protocols aimed at reducing high-beta excess when clearly present, while reinforcing calmer, more stable rhythms that support downshifting without producing daytime dullness. Depending on the presentation and assessment data, central or sensorimotor placements may be useful when the problem looks like restless arousal, motor tension, or difficulty disengaging. In other cases, especially when rumination and top-down overcontrol dominate, frontal strategies may make more sense.

The important caution is that sleep complaints are not neurophysiologically uniform. One client cannot fall asleep because the system is hot and busy. Another falls asleep but wakes repeatedly because the regulation is shallow and unstable. Another is exhausted all day yet becomes alert the moment the lights go out. Those are not the same treatment problem. This is where qEEG, symptom patterning, and plain old clinical observation become important. If a person becomes calmer with slower breathing but still cannot sustain sleep, I start wondering whether the residual issue is cortical instability rather than purely autonomic activation. If, on the other hand, neurofeedback makes them feel more organized during the day but does nothing for pre-sleep agitation, then the autonomic layer may still be under-addressed.

I also think session sequencing matters more than we sometimes admit. For some clients, daytime neurofeedback to improve regulatory flexibility followed by evening HRV work at home is a very elegant combination. For others, especially highly activated clients, starting the session with a few minutes of breathing-based settling may make the subsequent neurofeedback training more learnable. Sometimes the success of a sleep protocol is not in the target itself, but in the order in which the nervous system is invited to learn.

Sleep protocols beyond the headset

What I like about this paper is that it quietly reminds us not to become too device-centric. The therapeutic effect may not live in the shiny object. It may live in the coordinated package: breathing, pacing, therapist presence, environmental tone, repetition, and a manageable sense of agency. That is exactly how many strong clinical sleep protocols work outside VR.

In practice, I often think of sleep work as layered regulation. The autonomic layer includes paced breathing, HRV training, temperature awareness, and reducing pre-sleep sympathetic charge. The cortical layer includes neurofeedback strategies aimed at improving stability, reducing overactivation, or restoring flexibility where the daytime brain seems unable to transition efficiently. The behavioral layer includes timing, light exposure, evening routine, stimulants, exercise timing, and the sometimes underappreciated issue of over-efforting for sleep. Then there is the sensory layer: sound, visual input, body positioning, muscle tension, and how much external salience remains in the environment.

This layered model is particularly useful because many clients do not fail sleep interventions for lack of effort. They fail because the intervention mismatches the layer where their bottleneck lives. An autonomically overloaded client may not need a complex EEG strategy on day one; they may need to feel their exhale lengthen and their chest stop acting like it is still in a meeting. A client with long-standing dysregulation, trauma-related vigilance, or a very “busy” EEG profile may improve with breathing but plateau until neurofeedback is added. And some clients need both from the beginning, but in small, well-tolerated doses.

The larger lesson is that sleep is rarely improved by forcing it directly. It is improved by making sleep more likely. That distinction matters. The nervous system falls asleep; the conscious mind mostly gets in the way.

Which clients may benefit most

The easiest wins, in my experience, often come from the anxious high-arousal sleepers. These are the people who are visibly tired but physiologically unconvinced it is safe to power down. Their systems often respond well to HRV biofeedback, respiratory pacing, and simple neurofeedback approaches that reduce overactivation and increase tolerance for quiet. They are often the best argument for starting with the least invasive, fastest-feedback intervention first.

The depressed or mixed anxious-depressed sleeper can be more complicated. Some of these clients do benefit from calming protocols, but others are not simply hyperaroused. They may have fragmented circadian timing, low energy, emotional blunting, and inconsistent state regulation that requires a more individualized approach. In those cases, sleep complaints may improve only when daytime regulation improves more broadly.

I would also keep neurodivergent clients very much in mind here. Some autistic or ADHD clients may find immersive environments more tolerable and engaging than standard clinic feedback displays, while others may find VR overstimulating, disorienting, or just too much. Likewise, some clients who do poorly with traditional “relaxation” instructions are not resistant; they simply need regulation training that is more concrete, less abstract, and better matched to their sensory profile. This is where flexibility matters. A protocol is only good if the client can actually use it.

Do we really need VR?

Not always. And that is perfectly fine. The strongest argument for VR is not that it is inherently more therapeutic, but that it may improve adherence, motivation, attentional capture, and willingness to engage in practice long enough for regulation learning to happen. For some clients, that is a major advantage. But the downside is worth taking seriously too. A highly salient immersive environment may sometimes reduce transfer if the learning becomes too context-bound, too dependent on novelty, or too different from the ordinary bedroom environment where sleep actually needs to occur. So yes, VR may help some clients get started. But I would still want the long-term goal to be simple, portable self-regulation that works when the headset is off.

Conclusion

This emerging study adds a welcome piece to the sleep-regulation literature. In adults with depressive symptoms, anxiety symptoms, or both, both VR-based and conventional biofeedback approaches were associated with improved subjective sleep quality across a short 4-week window. The improvements were not trivial, especially in the symptomatic groups, and they appeared in clinically relevant domains such as sleep latency, sleep disturbance, and daytime dysfunction.

At the same time, the most responsible interpretation is a measured one. VR did not clearly outperform conventional biofeedback, the study relied on self-report sleep outcomes, and the exact “biofeedback” mechanics of the VR condition remain somewhat opaque. So this is not yet evidence that immersive technology is inherently the better clinical tool.

What it does suggest is something more practical: carefully structured regulation training, whether delivered through a headset or a standard therapist-guided setup, may help distressed sleepers feel meaningfully better. For the neurofeedback and biofeedback field, that is good news. It reminds us that sleep interventions do not always need to be elaborate to be useful. Sometimes the real therapeutic win is helping the nervous system relearn how to settle.

References

Seong, S., Kim, H., Cho, Y., Kim, M.-J., Park, K. R., Choi, J., Lee, S., Kim, D. J., Kim, S. J., & Jeon, H. J. (2025). Impact of virtual reality–based biofeedback on sleep quality among individuals with depressive symptoms, anxiety symptoms, or both: 4-week randomized controlled study. Journal of Medical Internet Research, 27, e65772. https://www.jmir.org/2025/1/e65772