Virtual reality exercise to help COVID patients with refractory breathlessness

Introduction

Chronic breathlessness syndrome has recently been delineated as a disabling sensation of difficult breathing despite optimized treatment of underlying medical conditions¹. Although chronic breathlessness, as a syndrome, has originally been described in chronic medical conditions, refractory breathlessness (RB) despite optimized treatment, can also be observed in acute settings². The condition has a non-negligible deleterious impact on patients’ quality of life and autonomy, by impacting cognition, locomotion, or mental health^3,4. RB’s definition emphasizes the importance of a complete work-up and identification of the underlying condition for the appropriate pathophysiological treatment; it also underlines the current failure to study an under-recognized respiratory suffering^1,2 and the need for development of evidenced-based interventions for RB. Indeed, breathlessness remains under-recognized in the hospital setting, although it can be easily collected and rated by trained staff and is associated with short and longer-term mortality⁵. Neuroimaging evidence suggests that the perception of breathlessness is underpinned by a neural network involving the insula, dorsal anterior cingulate cortex, amygdala, and medial thalamus and shares important pathways with other brain functions such as pain processing⁶ and bodily self-consciousness⁸. Prior interventions using immersive virtual reality (iVR)-based Digital Therapeutics (DTx), also referred to as digiceuticals⁹, have demonstrated alleviation of chronic pain in patients with complex regional pain syndrome or spinal cord injury^10,11. In the respiratory domain, visuo-respiratory stimulation (based on visuo-respiratory feedback) has been associated with an increased feeling of breathing control (breathing agency)¹², a reduced negative emotional state related to experimental dyspnea¹³, as well as changes in physiological measures of breathing^7,14.

Long COVID is a complex condition with multiples symptoms occurring beyond the initial period of COVID-19 infection recovery and affects patients who were managed in the community or in the acute care setting⁸. Like general weakness, malaise, fatigue, and concentration impairments, breathlessness has consistently been reported in long COVID cohorts with a high prevalence of around 25% (CI95 18% to 34%)⁷. As the subjective experience of breathlessness is poorly associated to specific physiological markers such as pulmonary function tests or lung imaging in long COVID ⁸, treatment is challenging and is most often focused on pulmonary rehabilitation ¹⁷. This also implies that a cognitive intervention using a neuro-rehabilitation approach could be tested to understand and alleviate this debilitating symptom.

The present COVid-19 Virtual Reality (COVVR) clinical study was performed to determine the effects of an iVR-based DTx on breathing comfort in patients recovering from severe COVID-19 pneumonia. We also quantified patients’ subjective benefit related to such intervention and the feasibility of using this new breathing rehabilitation tool at home or in the wards in this specific population.

Method

Patients

Thirty-one patients were recruited for the study (see Figure 1). Clinical inclusion criteria were that patients i) were recovering from severe COVID-19 pneumonia, and ii) presented with RB with a self-rated intensity of five or higher (out of ten) on a visual analogic dyspnea scale. Furthermore, patients had to be able to give consent and to understand and speak French or English. Patients that presented with unstable respiratory, neurological, or cardiac conditions, or psychiatric illness were excluded. Patients that scored below 25 on the Montreal Cognitive Assessment (MoCA) were also excluded.

• Download figure
• Open in new tab

Figure 1:

Study profile

Procedure

Screening

Patients were screened by a respiratory physician during morning rounds. Individuals were referred to the study. It was then verified that the patients met the inclusion criteria by performing an anamnestic interview as well as the MoCA cognitive assessment.

Setup

Eligible patients were installed in a semi-seated position in their hospital bed and wore a belt-mounted linear force sensor (Go Direct® Respiration belt, Vernier, Beaverton (OR), USA) fitted on the abdomen to allow proper recording of respiratory movements. They were also equipped with a head-mounted display (Zeiss VR ONEPLUS, Oberkochen, Germany) that holds a VR compatible smartphone (Samsung Galaxy S8, Seoul, South Korea). The smartphone ran the VR simulation and connected via Bluetooth® to the respiration belt. Software developed by the laboratory in collaboration with MindMaze allowed to collect and filter respiratory data and to render a computer-generated virtual environment in real-time.

Intervention Conditions

In the VR environment, patients could look around and see, next to them, a gender-matching virtual body lying on a bed in a similar position as theirs (Figure 2). The virtual body was flashing in a waxing and waning visual effect. This visual effect could be in synchrony (synchronous condition) or in massive asynchrony (asynchronous condition) with the patient’s respiratory movements. In the synchronous condition, the radiance of the visual flash was maximal at the end of inspiration and minimal at the end of the expiration. In the asynchronous condition, at the end of each visual flash a duration between 2.5 and 33.3 seconds is randomly generated for the next visual stimulation, such that the feedback is both phase-shifted and frequency-modulated with respect to the actual respiration.

• Download figure
• Open in new tab

Figure 2:

Portable setup and virtual reality feedback. (A) A respiratory belt captures the respiratory movements of the chest and sends the signal to a smartphone via Bluetooth. A custom software generates the virtual environment. (B) A matched-gender avatar is displayed and observed by the patients by slightly turning their head to the side. The virtual body is illuminated synchronously or asynchronously with respect to the patient’s chest movements. The top image represents the end of the expiration with a low flashing intensity, while the bottom image shows the end of the inspiration corresponding to the maximal luminosity in the synchronous condition.

Intervention Procedure

Once the patient was installed, a baseline assessment was performed, where patients closed their eyes (no visual feedback, NVF) and their respiratory movements recorded for two minutes.

Participants were then asked to rate their current breathing comfort by answering two questions according to a 7-point Likert scale (from −3= Strongly disagree to +3= Strongly agree): 1) I have difficulty breathing and 2) My breathing is enjoyable. These two items were also included in the following questionnaires and used to evaluate changes in 1) breathing discomfort, and 2) breathing comfort.

After baseline assessment, participants were randomly assigned in a 1:1 ratio to receive either the sequence “synchronous/asynchronous” or “asynchronous/synchronous”. In each block, patients were asked to first look around in the VR room, and then to orient their gaze towards the virtual body while relaxing for 5 minutes. They were not informed that the flashing of the virtual body was related to their breathing. Each block was followed by a subjective questionnaire of six questions (7-point Likert scale) derived from previous visuo-respiratory studies^4–6, (see Table 1). Question 1 evaluated the awareness of the visuo-respiratory experimental manipulation, while question 3 pertained to the breathing agency. Question 2 and 5 were proposed as control. Breathing comfort was evaluated with the same questions used in the baseline questionnaire (respectively Q4 and Q6, Table 1). Finally, to assess the acceptance and feasibility of the iVR intervention patients completed a feasibility questionnaire (7-point Likert scale, see supplementary section Table 4).

View this table:

• View inline
• View popup
• Download powerpoint

Table 1

Subjective questionnaire items

Results

From November 2020 to April 2021, patients’ enrollment, randomization and testing took place at the division of Pneumology at Geneva University Hospital. Twenty-six patients were randomly assigned either to the “asynchronous/synchronous” sequence (n=12) or the “synchronous/asynchronous” sequence (n=14). At the time of database lock in May 2021, data was available for all except two (7.7 %) of 26 patients (Figure 1).

Differences in subjective and physiological measures between conditions are reported in Table 2. At baseline, the mean breathing comfort rating was 0.08 ±1.6 and the mean breathing discomfort rating was −0.04 ±1.8. Mean values and standard deviations for each experimental condition, in function of the experimental sequence are provided in supplementary Table 5.

The main outcome measures of this study were breathing comfort ratings. The mean comfort rating was 0.75 ±1.78 during the asynchronous condition and 1.29± 1.37 during the synchronous condition, with an estimated difference of 0.54 (95% CI 0.05 to 1.04) (p<0.05) between conditions (Figure 3.A). The order of conditions did not impact these findings (no significant interaction for comfort rating between experimental conditions (asynchronous or synchronous) and experimental sequence (starting the experiment with asynchronous stimulations or starting it with synchronous stimulations)). Moreover, post-hoc paired one-sided t-tests, confirmed a significant difference between breathing comfort ratings during the experimental (synchronous) condition compared to baseline. Such difference was absent for the control (asynchronous) condition, excluding a mere effect of VR distraction (see supplementary section for statistical details). For the assessment of discomfort, even though a similar trend was observed in the data, neither the main effect of experimental condition nor its interaction with the experimental sequence reached significance.

• Download figure
• Open in new tab

Figure 3. Breathing comfort (A), Agency (B) and Awareness (C) test results.

The boxplots depicting subjects’ ratings during asynchronous condition compared to the synchronous condition, independent of experimental sequence. The thick line within a box plot represents the median, the diamond represents the mean, the upper boundary of the box indicates the 25th percentile and lower boundary the 75th percentile. The whiskers above and below the box indicate the minimal and maximal values, while points above the upper or below the whiskers indicate outliers. Subjective ratings were measured using a 7-point Likert scale with −3 = Strongly disagree, −2 = Disagree; −1 = Somewhat disagree; 0 = Neither agree nor disagree; 1 = Somewhat agree; 2 = Agree; 3= Strongly agree.

The secondary outcome measures of this study were the embodiment-related subjective ratings and the physiological measures. The mean agency rating was −0.67±2.26 during the asynchronous condition and 0.92±2.26 during the synchronous condition, with an estimated difference of 1.58 (95% CI 0.34 to 2.83) (p<0.05) between conditions (Figure 3.B). The order of conditions did not impact agency ratings (no interaction between experimental conditions and experimental sequence). The mean awareness rating was −0.67 ±2.14 during the asynchronous condition and 1.5± 1.8 during the synchronous condition, with an estimated difference of 2.17 (95% CI 1.07 to 3.27) (p<0.0001) between conditions (Figure 3.C) and the order of conditions did not impact awareness ratings. For the first control item, neither the main effect of experimental condition nor its interaction with the experimental sequence reached significance. For the second control item, ratings did not differ between the two experimental condition (−2.79 ± 0.51), leading to models not converging. For both RR and its variability, neither the main effects of experimental condition nor its interactions with the experimental sequence reached significance; suggesting that the effect of the intervention on comfort rating was not related to respiratory signal modulations.

The evaluation of the iVR intervention feasibility was the second main outcome measure of this study. As depicted on Figure 4 & Figure 5, 91.2% of the patients were satisfied by the intervention (Satisfaction: mean score = 1.75 ± 1.65; t = 5.20, p < 0.0001, 95% CI 1.17 to inf). In addition, 66.7% rated the iVR intervention as beneficial for their breathing (Respiratory benefit: 0.71 ± 1.92, t = 1.81, p < 0.05, 95% CI 0.04 to inf). 50% reported that the VR intervention made them feel better (Well-being benefit: 0.17 ± 2.3, t = 0.36, p > 0.05, 95% CI −0.64 to inf). Also, 45.8 % of the sample would like to continue using the device during their recovery (Rehabilitation: 0.04 ± 2.03, t = 0.10, p >0.05, 95% CI −0.67 to inf) and when at home (Home use: −0.34 ± 2.22, t = −0.74, p > 0.05, 95% CI −1.11 to inf). Finally, 37.5 % of the sample would have liked to use the intervention earlier during their stay at the hospital (Hospital use: −0.083 ± 1.86, t = −0.22, p > 0.05, 95% CI − 0.74 to inf). Descriptive statistics and statistical tests are described in the supplementary section (Table 6).

• Download figure
• Open in new tab

Figure 4. Feasibility scores for all items.

The boxplots depicting subjects’ ratings for feasibility items. The thick line within a box plot represents the median, the diamond represents the mean, the upper boundary of the box indicates the 25th percentile and lower boundary the 75th percentile. The whiskers above and below the box indicate the minimal and maximal values, while points above the upper or below the whiskers indicate outliers. Subjective ratings were measured using a 7-point Likert scale with −3 = Strongly disagree, −2 = Disagree; −1 = Somewhat disagree; 0 = Neither agree nor disagree; 1 = Somewhat agree; 2 = Agree; 3= Strongly agree.

• Download figure
• Open in new tab

Figure 5. Percentage of feasibility scores for all items, rated on a 7-point Likert scale,

with −3 = Strongly disagree, −2 = Disagree; −1 = Somewhat disagree; 0 = Neither agree nor disagree; 1 = Somewhat agree; 2 = Agree; 3= Strongly agree

Discussion

In COVVR, an iVR-based immersive DTx improved breathing comfort in patients with RB recovering from severe COVID-19 pneumonia. During synchronous visuo-respiratory stimulation, breathing comfort significantly improved compared to the asynchronous control condition and to a baseline condition. A similar effect of the intervention was observed for breathing agency, i.e. the feeling of being in control of one’s breathing. Finally, the vast majority of the sample reported a global satisfaction regarding the VR intervention and, more importantly, indicated that the iVR feedback improved their breathing. Our findings were neither modulated by the experimental sequence, nor can they be explained by VR distraction, or by changes in the respiratory measures during the intervention.

RB, a common but underreported condition, is defined as breathlessness that persists despite optimal treatment of the underlying pathophysiology. This results in important disabilities impacting cognition, locomotion and mental health^3,4,20. Mounting evidence using functional neuroimaging suggests that patients suffering from RB may exhibit brain “hypersensitivity” to afferent respiratory signals as a result of learned expectations during daily physical activities. ²⁰ Perception and anticipation of breathlessness are known to share neural correlates such as brainstem areas or the insular cortex ^21–23. It is likely that our manipulation interacts with such systems. Consequently, once treatment of the underlying pathophysiology has been optimized, the brain may become a potential target for pharmacological and non-pharmacological interventions. Pharmacological treatments may also be useful; low dose oral sustained-release morphine administered for RB is associated with improved health status in COPD without affecting PaCO2 or causing serious side-effects, especially in patients with mMRC stage 3 to 4²⁰. Pulmonary rehabilitation, an evidence-based multidisciplinary non-pharmacological intervention has also been shown to modify neural responses to learned breathlessness associations, which corresponds from a clinical point of view to central desensitization to dyspnea as an improvement in lung mechanics is most often not observed with this intervention ²¹.

Using iVR, the impact of visuo-respiratory stimulations on breathing has already been documented in healthy subjects, where the exposure to visuo-respiratory feedback led to increased breathing agency ¹², and to changes in respiratory rate²⁵, tidal volume variability¹⁴, and respiration amplitude²⁶. Moreover, a greater positive emotional state, during experimentally induced dyspnea (loaded breathing), has also been observed during a similar iVR intervention¹³.

iVR DTx are becoming popular in the field of chronic pain management as standard therapies have many shortcomings such as the prescription of controversial opioid medication ²⁵. Of note, breathlessness and pain share striking similarities⁶. They engage similar brain networks⁶, are best characterized by multidimensional models²⁸ and both respond to opioid treatment. As the global COVID-19 pandemic has progressed, a significant proportion of patients experience prolonged symptoms beyond the initial period of acute infection, such as RB. Such clinical cases are known as “Long COVID”¹⁵. The increasing number of patients isolated for prolonged periods has stressed an urgent need to develop multidisciplinary rehabilitation strategies that can be individualized and adapted to accommodate patients’ needs in intensive care or at home²⁴. Given the COVVR findings, we propose that our iVR DTx is a feasible and safe neuro-rehabilitation tool that could be considered for use to improve breathing comfort in patients experiencing refractory breathlessness after COVID-19 infection. Of note, while our intervention has been tested in patients recovering from COVID-19, its use could be extended to numerous respiratory conditions characterized by refractory breathlessness, as iVR DTx intervention directly impacts central processing of respiratory signals to alleviate breathlessness.

Our study comes with certain limitations. The main limitation is the small sample size, although it has been based on an adequate power calculation for a proof-of-concept study. Another limitation is that we included only a homogeneous population of patients recovering from severe COVID-19 infection. Our intervention should now be tested in a larger cohort of patients with refractory breathlessness to improve generalizability. Another important unanswered question is whether central desensitization to breathlessness while training with iVR DTx can persist when patients are off-treatment. Longer term studies with breathlessness ecological momentary assessments are therefore needed before such treatment can be widely accepted in a clinical setting.

One important innovation this study demonstrates is the improved efficacy linked to real-time biofeedback, going beyond the classically admitted distraction from suffering offered by iVR³⁰. Relying on the recent expansion of Internet of Things (IoT) technologies and of wearable physiological sensors in particular, our approach highlights how iVR DTx interventions can now more specifically target the neural mechanisms of pain and RB. Although in its current form our technology did not fulfill the expectations for home-use in our patients’ cohort, it paves the way for future e-health products that would not only provide monitoring but also a continuum of interventions from hospital bed to patient’s home.

In conclusion, our study shows that an immersive VR-based digital therapeutic can improve breathing comfort in patients recovering from severe COVID-19 pneumonia. Global satisfaction and respiratory benefit from the patients are reported, attesting to the feasibility of this simple intervention. Although more clinical studies are needed, our iVR-based DTx may become a key factor of multi-dimensional treatment of refractory breathlessness.

Contribution

Every author named at the start of the article contributed to the study: Conceptualization (AD, BO, BS, HB, KO, ST), Data curation (BS, FJ), Formal Analysis (FJ, BS), Funding acquisition (AD, BO, BS), Investigation (FJ, SA), Methodology (AD, BO, BS, FJ, HB, KO), Project administration (AD, FJ, SA), Resources, Software (CS, HB, LF), Supervision (AD, BO, BS, KO), Visualization (BS, FJ), Writing – original draft (BS, FJ), Writing – review & editing (AD, BO, BS, CS, HB, KO, ST). All authors confirm that they had full access to all the data in the study and accept responsibility to submit for publication. FJ and BS verified the data.

Funding

Dr Betka Sophie’s salary has been funded by a Marie Skłodowska-Curie Individual Fellowship (H2020-MSCA-IF-2019 894111/ RESPVR), awarded by the European Commission. The project has been partly funded by the Bertarelli Foundation.

Supplementary

Supplementary Results

Means and standard deviations in function of the experimental sequence

View this table:

• View inline
• View popup
• Download powerpoint

Table 5

Means and standard deviations for the asynchronous and asynchronous conditions, in function of the experimental sequence.

Breathing comfort - Tests against baseline

Using post-hoc paired one-sided t-tests, we found a significant difference between breathing comfort ratings during the synchronous condition compared to baseline (Difference: 1.25±0.431, t = 2.901, p < 0.01, 95% CI 0.511 to inf). This was not observed between breathing comfort ratings during the asynchronous condition compared to baseline (Difference: 0.708±0.547, t = 1.296, p > 0.05, 95% CI −0.229 to inf), excluding a mere effect of VR distraction.

Additional figures

• Download figure
• Open in new tab

Figure 6.

Subjective measures for which the main effect of the experimental manipulation was not significant. The boxplots depicting subjects’ ratings during asynchronous condition compared to the synchronous condition, independent of experimental sequence. The thick line within a box plot represents the median, the diamond represents the mean, the upper boundary of the box indicates the 25th percentile and lower boundary the 75th percentile. The whiskers above and below the box indicate the minimal and maximal values, while points above the upper or below the whiskers indicate outliers. Subjective ratings were measured using a 7-point Likert scale with −3 = Strongly disagree, − 2 = Disagree; −1 = Somewhat disagree; 0 = Neither agree nor disagree; 1 = Somewhat agree; 2 = Agree; 3= Strongly agree. Discomfort item:”I had difficulty breathing”

• Download figure
• Open in new tab

Figure 7.

Objective measures for which the main effect of the experimental manipulation was not significant. The boxplots depicting subjects’ physiology signal during asynchronous condition compared to the synchronous condition, independent of experimental sequence. The thick line within a box plot represents the median, the diamond represents the mean, the upper boundary of the box indicates the 25th percentile and lower boundary the 75th percentile. The whiskers above and below the box indicate the minimal and maximal values, while points above the upper or below the whiskers indicate outliers. Subjective ratings were measured using a 7-point Likert scale with −3 = Strongly disagree, − 2 = Disagree; −1 = Somewhat disagree; 0 = Neither agree nor disagree; 1 = Somewhat agree; 2 = Agree; 3= Strongly agree.

Feasibility

View this table:

• View inline
• View popup
• Download powerpoint

Table 6 Descriptive statistics and statistical tests of the feasibility items.

Ratings were measured using a 7-point Likert scale with −3 = Strongly disagree, −2 = Disagree; −1 = Somewhat disagree; 0 = Neither agree nor disagree; 1 = Somewhat agree; 2 = Agree; 3= Strongly agree. N=number of observations, SD = standard deviation, df = degree of