Virtual Reality (VR) refers to technology capable of generating immersive experiences of being present in virtual, computer-generated environments, typically achieved through the use of a head mounted display that withholds the outside world, continuously tracks head rotation (and possibly position), and updates the stereoscopic display as the user looks around and interacts with environment (1). Although recognizable VR technology first appeared in the 1960's, it is only in the last 5 years that consumer VR devices have become common (standalone versions of which now cost only a few hundred USD), alongside mature ecosystems for development and dissemination of VR applications (2). This presents a paradigm shift also for clinical and public health applications of VR (3), numerous successful examples of which have accumulated in the last 25 years (4).

One such example is VR exposure therapy: here, VR technology is used to create and present virtual equivalents of phobic stimuli (e.g., animals, heights, and situations) to users in an immersive way (5, 6), to allow for graded, controlled, and systematic exposure until the fear response habituates and inhibitory learning occurs (7). This treatment format is attractive to both patients (8–10) and therapists (11, 12), is feasible also with adolescents (13)—although considerably less studied with this population (14)—and is associated with low rates of deterioration, i.e., unlikely to have negative effects (15). Recent research has shown that VR exposure can be self-directed (16), and even packaged as automated interventions relying on gamification components rather than a real-life therapist directing treatment (17–20), and even a virtual therapist (21, 22).

As VR continues to grow and establish itself as a consumer technology (23), automated and self-directed interventions have the potential to make a significant public health impact by addressing the large treatment gap (24) and delay in treatment-seeking for anxiety disorders (25), phobias in particular. Although some methodological concerns have been raised (26), meta-analyses have demonstrated efficacy of VR exposure therapy for anxiety disorders (27–29), even when only considering trials featuring in-vivo behavioral approach tests as outcomes (30). This latter finding reveals that a reduced fear response to virtual phobic stimuli generalizes to the in-vivo equivalents that users will ultimately face in real life. This generalization likely explains why many trials report continued symptom improvements at follow-ups 1 to 6 years after completing VR exposure therapy (18, 31–34). Notwithstanding the immediate fear reduction (30), the full clinical effect may come from lowering the threshold for users to subsequently engage in in-vivo exposure opportunities as they appear in everyday life, breaking the vicious circle of avoidance that maintains anxiety disorders (35).

Although the idea that post-treatment, continued exposure predicts continued improvement likely applies equally well to traditional exposure therapy (36), VR exposure therapy is unique in that here, continued exposure entails not only continuing to do what was taught in treatment, but doing it in a different way. Despite having been noted as a limitation of the extant literature (4, 17), the necessary transition from in-virtuo to in-vivo exposure—and implicitly the lowered threshold hypothesis—has received surprisingly little research attention. Indeed, we are not aware of any study that has measured and attempted to statistically model the moderating role of post-VR, in-vivo exposure in explaining long-term continued symptom improvement. Indirect evidence, however, comes from two trials featuring the same automated VR exposure intervention for spider phobia: in one trial (18), participants in the VR arm were given written and oral information explaining the rationale for transitioning to in-vivo exposure in everyday life, as well as instructions for planning, execution, and evaluating exposure tasks; while in the other trial (17), no such written material was provided. In line with the lowered threshold hypothesis, the former trial saw continued improvement during the follow-up period, while the latter did not. In another trial, completing in-vivo exposure exercises during an explicitly framed transition period following a single session of VR exposure therapy for public speaking anxiety, was independently associated with additional symptom decrease (16).

These findings warrant further investigation, since the lowered threshold hypothesis has important implications for the design of automated VR exposure interventions that show such a great public health potential. In the current study, using data from one of the trials on automated VR exposure therapy for spider phobia, and using complementary statistical modeling techniques, we report a first such investigation.

Meta-analytic research has shown that fear reduction following VR exposure therapy generalizes also to in-vivo equivalent stimuli; moreover, several studies report that individuals who have completed VR exposure therapy continue to improve in the year(s) that follow. The lowered threshold hypothesis proposes that this post-VR continued improvement is explained in part by individuals now being able to engage in in-vivo exposure in everyday life, breaking the vicious circle of avoidance that define phobias. The current study offers a first test of this hypothesis, the results of which are in partial support of the hypothesis: while reported planned (but not involuntary) in-vivo exposure occasions was independently associated with continued improvement over the course of 1 year (with one exception), a sequential model revealed the hypothesized bidirectional association between symptom decrease and extent of exposure applied only to self-reported phobia symptoms, not behavioral avoidance, and only during the first few months after treatment.

Finding no outcomes associations with number of involuntary exposures is expected given that this measure presumably captures life circumstances (e.g., type of residence and line of work) rather than treatment adherence, and was included to control for this possible source of bias. Associations with planned exposures were however largely robust across analyses and in the expected direction. While these findings need to be replicated in independent samples, our results are consistent with the observed difference in long-term outcomes across the two trials on this particular VR intervention (17, 18), and suggest that it is indeed beneficial to explicitly promote transition to in-vivo exposure after completing VR exposure, and that this can be accomplished with brief oral and written instructions. Even in the broader field of cognitive behavioral therapy (CBT), where numerous studies have examined predictors of relapse (44, 45), there has been very relatively little research on the associations between long-term symptom trajectories and post-treatment adherence to the therapeutic strategies taught and practiced during active treatment. This lack of comparable research also makes it difficult to interpret whether the self-rated compliance of 61% found in the current study is high or not. One rare previous study found that adherence to different CBT components for insomnia varied extensively between components, and that the most reportedly used component (74%) was in fact not associated with improvement, while behavioral strategies like stimulus control and sleep restriction (41% reported use) were (46). While such findings are obviously of great clinical interest, examining this particular research question entails facing several methodological barriers. First, in lieu of automatically collected measures using wearables (47), ecological momentary assessments (48), or even momentary reporting of involuntary spider exposures (49), measurements of adherence and/or exposure need likely be retrospective, at least to some temporal degree. Second, since adherence is typically not a parameter that participants can be randomly assigned to, special statistical techniques are required to estimate the causal effect of adherence in itself (50). Including a second-stage randomization after the VR intervention to either a transition component or not, with explicit adherence targets (frequency of exposures, habituation threshold etc.), would not only allow more advanced and appropriate statistical modeling, but would also resolve the need for outlier adjustment, as was the case in the current study.

Interestingly, binarized exposure counts were not significantly associated with BAT improvements and barely significantly associated with FSQ improvements. This suggests that one exposure occasion is not enough for continued improvement and that the numeric range (covered by the other variable version) is in itself clinically pertinent. A seemingly counter-intuitive finding was that immediate self-rated phobia symptom improvement, but not behavioral approach improvement, predicted number of subsequent in-vivo exposure exercises. This could be explained by individuals with great behavioral improvement seeing no need for further in-vivo exposure, and/or that improved performance for many came at the cost of high distress that discouraged further in-vivo exposure. The discrepancy between models, along with the corresponding weak correlation between improvement metrics (yet stable correlations at each time point) may also be due to a delay in improvement insight: a similar, reversed discrepancy—rapid behavioral improvement without simultaneous decrease in self-reported fear—has been observed in fear memory reconsolidation disruption experiments (51). Although we cannot rule out the influence of demand characteristics (52), the remarkable stability over time of the correlation between the objective and subjective outcome measures, suggests no or negligible effects only. It should also be noted that participants did not have access to prior FSQ nor were provided with any interpretation guidelines; further, since no further treatment was offered at follow-ups, participants had no incentive to exaggerate symptom ratings. Regardless of source of the diverging results, the findings of the current study shows the value of capturing both aspects of phobia presentation.

The sequential model revealed that planned exposure exercises did not influence continued symptom reduction (self-rated phobia symptoms specifically) beyond the 3 month follow-up, yet planned exposure exercises were associated with both types of symptom reduction in longitudinal models (with few exceptions) that included exposure exercises as a time-varying predictor, i.e., assuming equal effect of exposure exercises across time. While the former approach breaks down the presumed process in individual steps, the latter approach examines whether variance in observed outcomes that is not explained by time, is instead explained by frequency of exposure exercises. In addition, statistical power varies between approaches. Of interest, participants reported the same number of planned exposure exercises at the 3 and 12 month follow-ups, despite covering different durations, revealing that this became less common over time. However, over the same 12 months, treatment effects continued to grow (18), suggesting that other factors likely contribute to continued symptom improvement at a later stage. Examining a broader range of predictors of long-term outcomes in VR exposure therapy should be considered an important topic for future research.

Our findings offer preliminary, partial support of the lowered threshold hypothesis of how Virtual Reality exposure therapy promotes continued symptom improvement. In longitudinal models with time-varying predictors, number of preceding in-vivo exposure occasions was associated with greater symptom decrease. In a sequential model, immediate post-treatment symptom reduction was associated with self-reported frequency of subsequent in-vivo exposures which in turn was associated with continued symptom reduction. However, this applied only to self-reported phobia symptoms, not behavioral avoidance, and was limited to the first 3 months of the follow-up. If these findings can be replicated in independent samples, it appears beneficial to actively promote in-vivo transition in order to maximize long-term effect of VR exposure therapy.

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

The studies involving human participants were reviewed and approved by Stockholm Ethical Review Board. The patients/participants provided their written informed consent to participate in this study.

PL, PD, AM, GA, LR, WH, and PC designed the study. PL, PD, and WH collected the data. PL and PD analyzed the data. PL drafted the manuscript. PD, AM, GA, LR, WH, and PC revised the manuscript critically for important intellectual content. All authors contributed to the article and approved the submitted version.

WH is the founder and chief technology officer of Mimerse, a company specializing in developing VR interventions for mental health. PL has consulted for Mimerse but holds no financial stake in the company. The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.