Acceptance of immersive head-mounted virtual reality in older adults

Immersive virtual reality has become increasingly popular to improve the assessment and treatment of health problems. This rising popularity is likely to be facilitated by the availability of affordable headsets that deliver high quality immersive experiences. As many health problems are more prevalent in older adults, who are less technology experienced, it is important to know whether they are willing to use immersive virtual reality. In this study, we assessed the initial attitude towards head-mounted immersive virtual reality in 76 older adults who had never used virtual reality before. Furthermore, we assessed changes in attitude as well as self-reported cybersickness after a first exposure to immersive virtual reality relative to exposure to time-lapse videos. Attitudes towards immersive virtual reality changed from neutral to positive after a first exposure to immersive virtual reality, but not after exposure to time-lapse videos. Moreover, self-reported cybersickness was minimal and had no association with exposure to immersive virtual reality. These results imply that the contribution of VR applications to health in older adults will neither be hindered by negative attitudes nor by cybersickness.


Results
participants. Seventy-six volunteers between ages 57 and 94 years participated in this study. Half of the participants (n = 38) was allocated to the HMD-VR group and the other half was allocated to the control group. One participant of the HMD-VR group dropped out after the first session for an unknown reason. One participant of the control group refused further participation during the first session (prior to completing a single test or questionnaire) and was replaced by a new participant to complete the sample of 38 individuals. The HMD-VR and control group were matched on age (BF 01 = 4.2) according to a Bayesian independent samples t-test, and matched on education level (BF 01 = 8.6) according to a Bayesian contingency table test. The groups were also matched on gender and independent living status. Participant characteristics of both groups are reported in Table 1. More details on participant recruitment are reported in the Methods section.
Changes in attitudes after a first HMD-VR exposure. We observed neutral attitudes towards HMD-VR prior to a first exposure to HMD-VR. In the HMD-VR group, attitudes increased from 3.4 (SD = 0.6) to 3.9 (SD = 0.8) after a first HMD-VR exposure, while in the control group initial attitudes were neutral (M = 3.0, SD = 0.8) and remained neutral after time-lapse video exposure (M = 3.0, SD = 0.9) (Fig. 3a). According to an analysis of covariance (ANCOVA) that modeled the post-pre attitude difference as a function of group (HMD-VR vs. control) and the self-reported user experience, there was inconclusive support for a main effect of group on the attitude difference (F(1, 71) = 2.56, P = 0.11, BF 10 = 1.02). There was also inconclusive evidence for a main effect of self-reported user experience (F(1, 71) = 4.49, P = 0.04, BF 10 = 1.01). Finally, there was anecdotal evidence for an interaction between self-reported user experience and group (F(1, 71) = 5.0, P = 0.03, BF 10 = 2.6). The interaction between self-reported user experience and group suggests that a more positive self-reported user experience was associated with a larger post-pre attitude difference in the HMD-VR group, but not in the control group (Fig. 3b). Levene's test indicated no significant heteroscedasticity, (F(1, 73) = 2.15, P = 0.15) and visual inspection of residuals showed no violations of the ANCOVA assumptions (Supplementary Materials 1).

Figure 2.
Schematic representation of the study protocol including the order of the test instruments and questionnaires that were administered to participants. In the first session, all participants completed the Montreal Cognitive Assessment 40 , the praxis scale of the Birmingham Cognitive Screen 43 , computer selfefficacy, computer proficiency and attitude towards HMD-VR scales. In a first recruitment phase, participants were allocated to the HMD-VR group (n = 38). In a second recruitment phase, participants (n = 38) were allocated to the control group. The two groups were matched on age, education, gender and independent living status. After exposure to HMD-VR or time-lapse videos in a second session, the user experience of the HMD-VR or time-lapse video condition was measured. Afterwards participants completed a second administration of the attitude scale and completed the simulator sickness questionnaire 41 . A subset of 44 participants also completed the Marlowe-Crowne social desirability scale 42 .
Factors predicting initial attitudes. A path model was estimated to study the association between age and initial attitudes and the extent to which this association was mediated by global cognitive status, computer proficiency and years of formal education. The pairwise correlations and descriptive statistics of all variables considered for the path model of all 76 participants are reported in Table 2. The variables included in the model and results of the model are visualized in Figs 4 and 5. The model could explain 23% of variance in initial attitudes, 43% of variance in computer proficiency, 45% of variance in the MoCA and 22% of variance in years of formal education. Computer proficiency (β = −0.66, P < 0.001, BF 10 = 700,139.5, Fig. 4d), MoCA (β = −0.67, P < 0.001, BF 10 = 33,695,113.2, Fig. 4b) and years of formal education (β = −0.47, P < 0.001, BF 10 = 866.53) were each significantly associated with age. There was strong support for the absence of a residual correlation between initial attitudes and the MoCA according to the BF when correcting for years of formal education, computer proficiency and age (β = 0.02, P = 0.90, BF 01 = 11.0, 95% CI [−0.25, 0.28], Fig. 5). In addition, the BF supports the hypothesis of no residual correlation between initial attitudes and years of formal education when correcting for age, computer proficiency and the MoCA score (β = 0.13, P = 0.26, BF 01 = 6.15, 95% CI [−0.09, 0.35], Fig. 5). There was evidence for a residual correlation between initial attitudes and computer proficiency (β = 0.31, P < 0.001, BF 10 = 3.96, 95% CI [0.06, 0.57]), and there was no residual association between initial attitudes and age when correcting for computer proficiency, MoCA and years of formal education (β = −0.12, P = 0.49, BF 01 = 6.8, 95% CI [−0.44, 0.21]). There was substantial support for the absence of mediation of age and initial attitudes by the Cybersickness. None of the participants in the HMD-VR group reported any severe discomfort on the SSQ scale, while two participants reported one severe discomfort and one participant reported two severe discomforts in the control group. The proportions of participants reporting moderate, mild or no complaints for each SSQ item in the HMD-VR and control group were low and similar between both groups (Table 3). There were more mild and moderate complaints in the HMD-VR group on some of the single symptoms than in the control group (Table 3), but exploratory analyses showed no significant differences at the single-item level of the SSQ (all uncorrected p-values larger than 0.13, Supplementary Materials 2). Across all SSQ items, there was substantial evidence for independence between self-reported discomfort and the participant group for severe (   www.nature.com/scientificreports www.nature.com/scientificreports/

Discussion
Given the rising popularity of VR for the assessment and treatment of a variety of health conditions that are common in older adults [1][2][3][4][5][6] , and the emergence of new low-cost high-quality immersive head-mounted displays, it is important to understand acceptance of HMD-VR in this population. In this study, we investigated older adults' attitudes towards HMD-VR and assessed changes in attitudes after a first HMD-VR exposure. We tested older adults of a broad range of ages, education levels, technology experience, global cognitive statuses and levels of independence (Table 1), as we assumed that these participants may be more representative of many immersive VR health applications' end-users than samples mainly consisting of community dwellers or baby-boomers.
The results showed that older adults without prior experience with HMD-VR had a neutral attitude towards this new technology. This result corresponds to previous findings on technology attitudes in older adults 26,45 . Furthermore, we found no association between initial attitudes towards HMD-VR and social desirable response styles in a subgroup who completed a social desirability scale (Supplementary materials 3). In addition, we found evidence that attitudes became more positive after a first exposure to HMD-VR, while attitudes remained stable in a group of participants who watched time-lapse videos on a standard notebook computer. This interaction effect suggests that the change in attitudes in the HMD-VR group was not merely the result of a positive experience, but was due to HMD-VR exposure itself. This result is compatible with previous findings on the effect of www.nature.com/scientificreports www.nature.com/scientificreports/ computer experience on computer attitudes 31 . The observation that attitudes towards HMD-VR can be improved by a first HMD-VR exposure further supports the use of HMD-VR in the older population and suggests that HMD-VR acceptance in the current cohort of older adults should not necessarily be of concern to HMD-VR health application developers. However, future studies are needed to reveal which features of the HMD-VR application itself and which methods of introducing older adults to HMD-VR applications affect the acceptance and user experience of HMD-VR. In addition, the type of help needed while using HMD-VR and the extent to which older adults can learn to use HMD-VR independently remains an important question.
In addition, we explored whether certain characteristics of older individuals predicted initial attitudes towards HMD-VR. Previous research revealed that computer interest was associated with age, education, and computer knowledge 32 , that technology usage depended on age and that the latter was mediated by computer anxiety and intelligence 33 . Our results expanded this knowledge base by investigating the relationship between age, attitudes towards HMD-VR and the mediating role of computer proficiency and global cognitive status in a heterogeneous sample of community dwellers and adults in assisted living. As predicted, age had a significant negative association with initial attitudes without correcting for computer proficiency, global cognitive status and years of formal education, which is in line with earlier findings 25 . Moreover, the age-attitude association was not mediated by global cognitive status when controlling for computer proficiency and years of formal education. Thus, although older adults experiencing cognitive decline may experience difficulties in learning to use new technology 33 , our results suggest that they are equally willing to try out new technology as their peers with a better cognitive status.
Cybersickness symptoms did not occur frequently in our sample of participants, and, importantly, were not significantly associated with the exposure to HMD-VR in a between-subject design. Furthermore, although there were more reports of mild to moderate complaints of certain cybersickness symptoms in the HMD-VR than the control group (Table 3), none of these differences were statistically significant (Supplementary Materials 2). Noteworthy is that research in healthy participants and vestibular patients has shown a decline in motion sickness susceptibility with increasing age 46 . More research is needed to clarify how age affects susceptibility to motion sickness induced by different types of HMD-VR applications.
We did not observe major safety concerns when applying HMD-VR in older adults following the HMD-VR device's safety guidelines. However, we must note that this does not imply that all HMD-VR applications are safe for these end-users. Foremost, we did not systematically assess in what proportion of older adults HMD-VR can be safely used (e.g. proportion of older adults meeting the safety exclusion criteria). The safe applicability of the technology in a large proportion of older adults is an important consideration for HMD-VR health application developers and should be further investigated. Additionally, we cannot make statements about falling risks as participants remained seated during the HMD-VR exposure. It is known that HMD-VR affects dynamic balance in young adults without balance disorders 47 , making it likely that HMD-VR health applications for older adults may best be developed to be used in a seated position. Furthermore, prevalence of cybersickness symptoms depends on the design of the HMD-VR application, with more severe cybersickness symptoms when motion is not under control of the observer 48 . In this study, we used an HMD-VR application with motion that is under control of the observer to minimize the chance of observing certain cybersickness symptoms. Therefore, future research should investigate to what extent older adults are sensitive to motion that is not under their control and how the design of the HMD-VR device and application could mitigate possible cybersickness issues in older adults.
To conclude, we showed that older adults are willing to use HMD-VR and have more positive attitudes towards HMD-VR after a first positive HMD-VR experience. Moreover, the negative association between age and initial attitudes was not mediated through cognitive status or years of formal education. Furthermore, cybersickness was not significantly associated with HMD-VR exposure. These results support the development and use of HMD-VR health applications for older adults.  Table 2. Pairwise correlations, descriptive statistics and 95% bootstrapped confidence intervals of all variables considered for the path analysis. ***Significant at the 0.001 level after Holms correction for multiple comparisons. CP = computer proficiency, CSE = computer self-efficacy, MoCA = total score on the Montreal Cognitive Assessment, YFE = years of formal education, In attitudes = mean score on the first administration of the attitude scale, M = mean, SD = standard deviation. Computer self-efficacy was not included in the model, since it was strongly correlated with computer proficiency and would result in multicollinearity (>0.80). We did include years of formal education, as it was only moderately correlated with age, and we wanted to dissociate the effects of age and education. www.nature.com/scientificreports www.nature.com/scientificreports/ Methodology participants. Participants were recruited through nursing homes and participant contact lists of previous studies of our research group. A medical implant or epilepsy were grounds for exclusion based on the HMD-VR device's EU safety regulations. Poor vision or hearing that could not be corrected by glasses or a hearing aid, inability to provide informed consent and previous HMD-VR exposure were also grounds for exclusion. In the first recruitment phase of the study, all participants were allocated to the HMD-VR group (n = 38). Afterwards,  Predictors of initial attitudes. The standardized regression coefficients for each path are depicted. The association of age with computer proficiency, education and the MoCA were significant. There was no residual association of age and attitudes corrected for computer proficiency, education and MoCA. There was no residual association of years of formal education and attitudes corrected for computer proficiency, age and MoCA. There was no residual association of MoCA and attitudes corrected for years of formal education, computer proficiency and age. The association between age and attitudes was not mediated through education and MoCA. The mediation role of computer proficiency for the relation between age and attitudes was inconclusive.
www.nature.com/scientificreports www.nature.com/scientificreports/ in a second recruitment phase of the study, 38 participants were recruited and allocated to the control group. The participants of the control group were matched on age, education, gender and independent living status to the participants of the HMD-VR group. In each group, 45% of participants were residing in assisted living, while 55% were community dwellers. The majority of participants (89%) had no knowledge that the study would involve technology, and 74% of participants were not recruited by any means of technology (e.g. no e-mail contact, no online advertisements). All study procedures were approved by the Social and Societal Ethics Committee of the KU Leuven (G-2017 01 733) and executed in accordance with the committee's ethical guidelines. Each participant provided written informed consent.

Instruments. HMD-VR exposure and time-lapse videos. The HMD-VR group experienced the application
Perfect of nDreams 49 using the Oculus Rift and Touch Controllers. By means of the Touch Controllers, the user can interact with the HMD-VR environment, for instance by picking up or throwing objects. The environment is artificially made, but the scenes look natural and familiar (e.g. mountain, lake). This HMD-VR application has no observer-independent background motion in the scenes. The HMD-VR exposure was audio-taped. The control group watched six pre-selected time-lapse nature videos on the YouTube broadcasting system available under the standard YouTube License. While watching the videos they listened to the audio through headphones to mimic wearing a head-mounted device in the HMD-VR condition. The videos were selected to resemble the scenery of the HMD-VR environment, to elicit a similar aesthetic appreciation as the HMD-VR environment, and to have an equal duration as the HMD-VR exposure.
Neuropsychological Assessment. The MoCA 40 is a short, sensitive screen for mild cognitive impairment that consists of pen-and-paper tasks measuring executive functions, memory, language and reasoning. The MoCA has good internal consistency (α = 0.83) 40 . The BCoS Praxis measures the ability of participants to execute purposeful actions with their upper limbs 43 . The percentage of exact score agreement from test to re-test ranges from 50% to 60% in neurologically healthy controls 50 . The subscale consists of multistep object use, figure copying and gesture production, recognition and imitation. All tasks were administered and scored according to the test manual instructions, and scores were compared to the age-adjusted cutoff scores of the test manual 50 .
Questionnaires. As to date, there is no validated Dutch questionnaire to measure attitudes towards HMD-VR. Thus, we developed an attitude scale consisting of questions gauging the perceived ease of use, perceived usefulness and anticipations about using HMD-VR. Eighteen items were constructed: six based on a questionnaire assessing attitude towards computers 51 , five on a questionnaire assessing attitude towards internet 52 and seven new items were added (Supplementary Table S2).
To measure user experience, a 23-item scale gauging enjoyment and immersion was designed (Supplementary Table S3). Six of the 11 enjoyment items were based on the intrinsic motivation inventory 53 and 10 of the 11 immersion items were translations of the International Test Commission Sense of Presence Inventory items 54 .
Computer proficiency (CP) (Supplementary Table S4) was measured with a 22-item scale of which 10 new items were added onto 12 items that were translations of the CP items of Boot and colleagues 55 .  Table 3. Proportion of participants that reported moderate, mild or no discomfort on the SSQ. Values represent the proportion of participants who rated each SSQ item as moderate, mild or no discomfort. Ratings of severe discomfort on the SSQ items are not included in the table as they did not occur in the HMD-VR group.
www.nature.com/scientificreports www.nature.com/scientificreports/ Computer self-efficacy (CSE) (Supplementary Table S5) was measured with a 14-item scale gauging the confidence and anxiety in performing beginner and advanced computer activities. Eleven items were direct translations of the items reported in Barbeitte and Weiss 56 .
In addition, we measured openness with the 12-item openness subscale of the validated Dutch Neuroticism Extraversion Openness Five-Factor Inventory 3 57 . This scale was excluded from further analyses given its insufficient internal consistency (Supplementary materials 4).
The attitude, user experience and computer self-efficacy scale items were rated on a 5-point Likert scale ranging from "totally disagree" (1) to "totally agree" (5), with a neutral position (3). All items were scored in such a way that the mean score on each scale ranges from a low score of "1" to a high score of "5". The computer proficiency items were rated on a 5-point scale going from "I have never tried to do this task" (1) to "very easy" (5).
Cybersickness was measured with the Simulator Sickness Questionnaire (SSQ) 41 that was translated by our research team. Each of the 16 SSQ items were rated with four levels representing no, mild, moderate or severe discomfort.
Social desirability was measured with a validated short version of the Marlowe-Crowne Social Desirability Scale (SDS) 42 , which was translated by our research team. Each SDS item is designed to elicit a socially desirable response and has to be evaluated as true/untrue by the participant. The total score is the proportion of items on which the participant responded socially desirable.
procedure and design. The study consisted of two sessions with on average one day in between the two sessions (range: 0-5 days, Fig. 2). The first session took approximately 60 minutes and the second session took approximately 90 minutes. In the first session, a demographic interview, the MoCA and BCoS Praxis were administered. Then, participants completed the CSE, CP, attitudes and NEO-FFI 3 openness scales. In the second session, the HMD-VR group first received an explanation about the HMD-VR device and how they could interact with the virtual environment. Then, they were exposed to the virtual environment for an average of 26 minutes (SD = 5.5 minutes, range: 8-36 minutes) and participants were allowed to take breaks. After a few minutes of free exploration in the virtual environment, the experimenter assisted participants to let them perform interactions with the virtual objects (e.g. throwing a stone in the lake). Assistance could take the form of reminding participants what they could do in the environment, explaining participants how they could perform actions with the touch controllers, and manually assisting participants to execute these actions. The examiner explained participants how to perform actions on average 5 times (SD = 3, range: 1-18) and manually assisted participants on average 1 time (SD = 3, range: 0-17) during the HMD-VR exposure. Participants in the control group also received an explanation about the HMD-VR device, but were not exposed to HMD-VR. Instead, they watched time-lapse videos. In both groups, the experimenter stayed with the participants and participants were allowed to interact with the experimenter freely and ask as much help as needed to operate the HMD-VR or video application. The participants remained seated at all times. After exposure to the HMD-VR or time-lapse videos, participants were asked about their experience using the user experience scale. The attitude scale was re-administered after the user experience scale, thereafter followed by the SSQ. The SDS was added to the study protocol for the last 44 participants (6 of the HMD-VR and 38 of the control group). All questionnaires were administered with pen and paper as a semi-structured interview supervised by a trained clinical psychologist and in the same order for each participant. Special care was taken to ensure that all participants, including participants with mild cognitive impairment (MoCA < 26) understood each question. In a supplementary analysis on the response style of the participants to the questionnaires, we did not find evidence that participants with mild cognitive impairment had issues understanding the questionnaire items (Supplementary materials 3). Data analysis. The dataset and data analysis scripts are available on FigShare 58,59 . Data analysis was performed in R 60 . First, an item analysis to detect items that negatively impacted the scales' reliability was conducted. Items with an item-total correlation lower than 0.20 were removed. Based on the remaining items, the Cronbach's alpha, its 95% confidence interval (CI) and the mean inter-item correlation were calculated 61 . If a scale had a Cronbach's alpha lower than 0.75, the scale was excluded from further analyses 61 .
To test whether a first HMD-VR exposure affected HMD-VR attitudes, we used an ANCOVA to model the post-pre attitude difference as a function of the main effect of the between-subject factor (HMD-VR versus control group), the main effect of the covariate self-reported user experience and the interaction between both variables using the lm function in R 62 . Type III sum of squares were estimated. Thus, we tested whether there was substantial evidence for an effect of group on top of the self-reported user experience and vice versa with inclusion of the interaction term and we tested whether there was substantial evidence for an effect of the interaction term in addition to the two main effects. The evidence for an effect was evaluated based on the BF, which was computed with the Bayes Factors package 63 according to the same model comparisons used for computing the F-statistics. A BF 10 larger than 3 was considered substantial evidence in favor of the alternative hypothesis.
The assumption of heteroscedasticity was checked with Levene's test and we visually inspected the association between the fitted values and residuals of the ANCOVA model ( Supplementary Fig. S1).
To assess whether initial attitudes were associated with age and whether this association was mediated through years of formal education, computer proficiency and MoCA, we conducted a path analysis. For predictors that had strong collinearity with each other (>0.80), we chose to include only one predictor. Standardized regression coefficients, accompanying 95% CIs and BFs based on the method described in Wetzel and Wagenmakers 64 were estimated. Path analysis was performed with the Lavaan package 65 . Power analysis in G*Power 3.10 66 revealed that 60 participants were needed to detect correlations of 0.40 evaluated at a threshold of 0.01 to obtain 80% power.
To compare cybersickness between the HMD-VR and control group, we tested the dependency of the occurrence of mild, moderate, severe and no physical complaints on the between-subject condition (HMD-VR versus (2019) 9:4519 | https://doi.org/10.1038/s41598-019-41200-6 www.nature.com/scientificreports www.nature.com/scientificreports/ control group) across all SSQ items using Bayesian contingency table tests 67 . The 95% credible interval of the difference in the probability of reporting a complaint between the HMD-VR and the control group was estimated using Markov Chain Monte Carlo sampling. Additionally, the proportion of participants who reported severe, moderate, mild or no discomfort for each SSQ item separately was calculated. In addition, we calculated the Frequentist Chi-square and Bayesian contingency table test for each item of the SSQ separately (Supplementary materials 2).
The number of verbal and manual interventions during the HMD-VR exposure were counted by one observer who was blind to the participant characteristics and test scores. These counts were based on audiotapes of the HMD-VR sessions.
All reported statistics are based on two-sided hypothesis tests and all BFs are based on uninformative prior distributions 63 . BFs larger than 3, which correspond to a 75% confidence level in the decision 68 , are interpreted as substantial evidence in favor of either the null or alternative hypothesis 44 and BFs in between 2 (67% confidence level) and 3 are interpreted as anecdotal evidence 69 .