The study was conducted between December 2022 and February 2024 in 5 nursing homes located in the Auvergne-Rhône-Alpes region of France. Two facilities were situated in Lyon, accommodating 80 and 103 residents, respectively. A third facility, located in a village on the outskirts of Lyon, housed 80 residents. The remaining 2 facilities were in Le Puy-en-Velay, with capacities of 45 and 49 residents. Nursing homes were initially selected from the Korian private group, which had granted study authorization, and recruitment was later extended to facilities located near Lyon and Le Puy-en-Velay to increase the number of participating centers. Participants included in the study lived in one of these 5 nursing homes and had BPSD, measured by the Neuropsychiatric Inventory (NPI) [28,29], with a score ≥9 on at least 1 of the following symptoms: apathy, agitation, anxiety, depression, or aberrant motor behaviors, since the VR intervention was developed to focus mainly on these symptoms. An NPI threshold of ≥9 was used to identify symptoms considered both severe and frequent. Exclusion criteria were BPSD severity preventing group participation, infectious neurological diseases, substance abuse, nonstabilized psychiatric disorders, severe uncompensated sensory deficits, epilepsy, visual hallucinations, pacemakers, severe or uncontrolled heart conditions, scalp burns, and scalp sores. Based on the recruitment capacity and the feasibility of carrying out the interventions in the different centers, it was planned to recruit 12 participants in each center (ie, a planned sample size of 60 participants). Randomization was stratified by center and was conducted electronically (Ennov Clinical) by the investigator, with a maximum of 6 participants per group per center. Resident capacities for each center, as well as the numbers of participants enrolled and included in the final analyses, are presented in Multimedia Appendix 1. The random allocation sequence was accessible only to the data manager and was concealed from personnel responsible for participant enrollment and assignment. Given the nature of the intervention, blinding of participants and intervention facilitators was not feasible.

This study (IDRCB 2021-A00297-34-A) was approved by the local ethics committee (Comité de Protection des personnes EST I, France) on June 30, 2022, and was registered on ClinicalTrials.gov with the title “Virtual Reality Intervention for the Reduction of Behavioral and Psychological Symptoms of Dementia (LUMEENCOMPAD)” with registration number NCT04769024. Participants provided written informed consent, or, if unable, their relative or legal guardian did so, with the participant’s oral consent still considered. Individuals unable to consent and without a relative or legal guardian were excluded.

The primary outcome was the NPI, a French version (entitled NPI-ES for “équipes soignantes”), the equivalent of the NPI for nursing homes [28,29]. Responses to the NPI were collected from health care staff by a trained health care professional (ie, psychologist or medical doctor). The NPI was designed to assess changes in the patient’s behavior by gathering information from someone who knows the patient well (ie, a family or professional caregiver), enabling clinical follow-up and treatment evaluation. It assesses 12 BPSD (delusions; hallucinations; agitation and aggression; depression and dysphoria; anxiety; elation and euphoria; apathy and indifference; disinhibition; irritability and lability; aberrant motor behavior; nighttime behavioral disturbances; and appetite and eating abnormalities) in terms of frequency (range 0‐4) and severity (range 0‐3). The score is calculated by multiplying the frequency and severity for each dimension and adding up each of these subscores (range 0-144; 144 being the maximum score).

Secondary outcomes were immediate well-being, social interactions, and apathy symptoms.

Immediate well-being was assessed using the Immediate Well-Being Evaluation (IWBE), a visual analog scale developed for people with major NCD, even at a severe stage (eg, Mini-Mental State Examination [MMSE] <10) [30]. Participants had to answer the question “How are you feeling right now?” by moving a slider on the scale. On the side presented to the participant, the scale shows pictograms of simple facial expressions (ie, joy, neutral, and sadness) and it goes from “Not Well” (on the left) to “Very Well” (on the right). On the other side, facing the evaluator, the position of the slider is associated with numbers from 0 (the lowest feeling of well-being) to 5 (the highest); Multimedia Appendix 2.

Social interactions were assessed by the Social Observation Behaviors Residents Index (SOBRI), an observation grid evaluating several interaction behaviors (eg, “mutual gaze” and “nodding”) toward peers and health care professionals [31] during the interaction time. It is composed of 16 items, which can be coded as 0 or 1 depending on their occurrence (0 indicates absence; 1 indicates presence of the behavior). Thus, the total score ranges from 0 to 16. An external observer, a clinical research coordinator, completed the SOBRI during each of the 12 sessions, observing each participant separately for 4 minutes.

Apathy symptoms were assessed using the Apathy Interview [32], a scale proposed to health care staff to evaluate 3 symptoms of apathy (ie, emotional blunting, lack of initiative, and loss of interest). For each symptom, its intensity is estimated from 0 (absence of the symptom) to 4 (major). The total score ranges from 0 (absence of apathy) to 12 (major apathy).

Adverse events were systematically recorded after each session, and their seriousness was assessed by the investigator.

The last evaluation of the participants was conducted within 2 weeks following the final session, depending on caregiver and facility availability. The NPI and the Apathy Interview were rated, and a questionnaire was completed to report any major event in the life of the resident that could interact with the results of the study (eg, death of a loved one or deterioration in general condition). The reporting of this randomized controlled trial followed the CONSORT (Consolidated Standards of Reporting Trials) 2025 statement (Checklist 1).

Sociodemographic characteristics (ie, age and sex) were extracted from participants’ medical records by the investigator at inclusion. The MMSE was administered at baseline whenever feasible; otherwise, the most recent score recorded within the preceding 3 months was retrieved from the medical record. As intervention and control participants were equally distributed within each center, any center-specific effects were naturally balanced between groups. Consequently, center was not included as a covariate in the analyses.

Participant characteristics and various measures were compared between the intervention and control groups using the Chi-square/Fisher test for proportions and 2-tailed Student’s t test for means (SDs). For scores assessed preintervention and postintervention, mean (SD) values of all preintervention and postintervention scores were calculated. The effect of the intervention on the outcomes (ie, NPI, SOBRI, IWBE, and Apathy Interview) was evaluated using generalized linear mixed models with group, time, and the interaction between time and group as fixed factors, and a random intercept for participants. The models were adjusted for age, sex, and number of sessions to account for differences in intervention exposure across participants due to participants missing 1 or more sessions. Additional analyses, including MMSE as an adjustment variable, were conducted but could not be included in the primary analyses due to the number of missing MMSE data. Results were presented as standardized β coefficients (effect sizes) for each fixed effect, the standard error, and the P value. For the analyses, participants were excluded if they attended only 1 or no session, or if preintervention and/or postintervention measurements were missing.

All tests were 2-tailed, and a P value of less than .05 was considered statistically significant. For each analysis, the number of participants analyzed is indicated in brackets. This number depends on the presence of missing data. Analyses were performed using SPSS software (version 20.0; IBM Corp) and R software (version 4.1.2; R Foundation for Statistical Computing).

Fifty-six participants were enrolled, but 3 were excluded for not attending any sessions, leaving 53 participants (27 intervention and 26 control). The mean age was 85.7  (SD 8.3) years, with 60.4% (32/53) aged 85 or older. Most were women (42/53, 79.3%), and Alzheimer disease was the most common diagnosis (30/53, 56.6%). No baseline differences were found between groups in demographics or assessment measures (NPI, Apathy Interview, IWBE, and SOBRI); Table 2. A distribution of participants with each NPI symptom by group (intervention/control) and study phase (preintervention/postintervention), according to frequency and severity levels, is also provided in Multimedia Appendix 3.

Of the 53 participants, only 15 attended all 12 sessions, but the majority (n=46, 86.8%) attended at least 6 sessions with a median of 10 (IQR 8‐12) sessions. No significant difference in the number of sessions attended was observed between the intervention (mean 8.59, SD 3.45) and control (mean 9.57, SD 2.48) groups (P=.24). The number of sessions attended was added as an adjustment variable in the models for the analyses, and a summary of attendance and reasons for nonattendance is provided in Multimedia Appendix 4. One participant was excluded from the analyses because he attended only 1 session, and postintervention assessments were missing for 2 participants because they discontinued the study early; see Multimedia Appendix 5 for an equivalent flow diagram adapted to our study. Thus, 50 participants were analyzed for the primary outcome (NPI).

Only 2 research-related adverse events, out of the 481 recorded sessions, were reported in the intervention group, corresponding to known symptoms of motion sickness and remaining at the lowest stage of severity. No research-related adverse events were recorded in the control group.

A trend toward a reduction in BPSD was observed between the beginning and the end of the study, although this reduction was not significant (β=−1.148, P=.07, n=50). Additionally, this reduction did not depend on the group (intervention or control; β=−3.49, P=.32, n=50); see details in Tables 3 and 4.

Social interactions observed during the sessions tended to be lower in the intervention group than in the control group (β=−1.789, P=.07, n=52). Social interactions did not appear to change over the course of the sessions (β=−0.455, P=.16, n=52). After adjustment for MMSE, the association between group allocation and social interactions was significant (β=−3.805, P=.001, n=28 participants with available MMSE). In this adjusted model, higher MMSE scores were also independently associated with a greater number of social interactions during the sessions (β=0.325, P=.007, n=28).

The analysis showed no significant effect of the VR intervention on the feeling of immediate well-being when compared with the control group (β=0.281, P=.22, n=50). However, the IWBE score increased significantly between the beginning and the end of the sessions, regardless of the group (β=0.363, P<.001, n=50). The number of intervention sessions attended was positively associated with well-being (β=0.101, P=.01, n=50; Tables 3 and 4).

Apathy symptoms, as assessed using the Apathy Interview, were not significantly affected by the sessions (β=−0.415, P=.11, n=50), and there was no significant difference between the intervention and control groups (β=0.038, P=.97, n=50), Tables 3 and 4.

The primary aim of this multicenter randomized controlled trial was to evaluate whether a group-based VR intervention reduced BPSD compared with an active cognitive stimulation control in nursing home residents. Contrary to our hypothesis, VR did not significantly improve BPSD, apathy, or immediate well-being compared with the active control. However, both interventions improved immediate well-being, and VR was found to be safe and feasible to implement in routine nursing home practice.

This study did not identify significant differences between the 2 groups in terms of BPSD or apathy, contrasting with previous findings suggesting that VR could reduce aggressiveness [22], apathy [23,27,34], and depression [27]. One explanation may be that many earlier studies lacked a comparison group [23,27] or used standard care as a control [22,34], whereas ours used an active control group with cognitive stimulation. Another possibility is that the NPI, although widely used, may be insufficiently sensitive to detect subtle short-term changes in BPSD following nonpharmacological interventions. Additionally, the postintervention assessment was conducted at varying time points within the prespecified 2-week follow-up period, which may have introduced additional variability in the outcome measures and reduced our ability to detect intervention effects.

Although not the primary objective of the study, a notable finding was that immediate well-being improved following each session, regardless of the intervention type. This aligns with prior research reporting increased short-term pleasure [24,35], and satisfaction [36] following similar interventions.

Some trends in the results warrant further exploration. BPSD tended to decrease after the interventions, regardless of the group. As Gueyraud [37] noted, two 45-minute weekly sessions represent less than 0.5% of a resident’s week, so their clinical impact may be modest. This raises the question of whether such interventions, even if effective, yield clinically meaningful benefits. Larger samples may be required to detect small but meaningful effects, and it could be relevant to use multiple outcome measures to assess BPSD reduction—for instance, by monitoring medication use and its longitudinal changes.

These observations could also suggest the potential need for longer interventions, which raises another important yet unresolved issue concerning the structure of VR interventions. In this study, our schedule—two 45-minute sessions per week for 6 weeks—mirrors the existing literature, in which VR sessions typically last around 27 minutes, with 2.8 sessions per week over 7.6 weeks [17]. However, many studies do not report all parameters (duration, frequency, and total length) [17]. In our sample, many participants missed sessions, with a median attendance of 10 (IQR 8‐12) sessions. Attendance variability may have affected outcomes, as improvements in well-being appeared to be associated with the number of sessions attended.

Cognitive ability could also play a role in the effectiveness of these interventions. Cognitive decline is known to be associated with increased BPSD [38], especially apathy [39,40] and reduced well-being [41,42] and may influence the ability to engage in social interactions [43]. In our study, greater cognitive impairment appeared to be linked to reduced social engagement. This observation is consistent with previous literature and suggests that the NCDs severity may moderate the effects of nonpharmacological interventions. However, missing MMSE data prevented a more precise analysis of this relationship. Future studies should further investigate this relationship.

As stated above, attendance variability may have affected our results since well-being appeared to be associated with the number of sessions attended. However, because the interpretation of this covariate was not prespecified, this finding should be considered exploratory. Moreover, attendance was not randomly assigned, and participants who attended more sessions may also have differed from those with lower attendance in terms of motivation, health status, and fatigue. Therefore, this exploratory association between session attendance and improvements in well-being should be interpreted cautiously and cannot be considered evidence of a causal relationship. Nevertheless, the improvement in well-being remained significant after adjustment for session attendance, suggesting that the observed benefits cannot be explained solely by differences in exposure to study sessions. Taken together, these findings suggest that differences in attendance may have contributed to the observed improvement in well-being but are unlikely to account for it entirely. Future studies should further investigate factors influencing adherence and their impact on intervention outcomes.

A further limitation concerns the assessment of cognitive impairment. Missing MMSE data prevented a precise analysis of the relationship between NCD severity and intervention outcomes. These missing data were mainly due to the assessment being considered infeasible by the clinical staff, particularly for participants with advanced cognitive impairment. Consequently, NCD severity could not be systematically accounted for in our analyses. Furthermore, in line with the pragmatic design of the study and the principle of data minimization required during the ethical approval process, detailed clinical information such as comorbidities and medication use was not collected. As a result, adjustment for these potential confounders was not possible, and residual confounding cannot be excluded. Future studies should consider alternative measures of NCD severity, such as the Clinical Dementia Rating, along with more comprehensive clinical characterization of participants.

Personal preferences may also moderate the effects of VR interventions [44]. However, the design and sample size of the present study do not allow us to explore this question. While individualized VR interventions might prove more effective [44], they are also more demanding in terms of staff time. We therefore believe that both individualized and group-based approaches may be complementary, offering nursing homes different options depending on their available resources.

Methodological limitations should be considered. Despite a larger sample than many previous studies, statistical power was limited, particularly due to missing data (eg, MMSE, IWBE, and SOBRI) and participants’ absenteeism. Attendance was influenced by various factors, including weather conditions, medical appointments, fatigue, lack of motivation, or logistical constraints, reflecting real-world challenges in nursing homes. This limitation in statistical power, combined with an imbalance across BPSD categories, did not allow us to examine potential differential effects of VR on specific NPI subitems, even though it would have been of interest. This could be an interesting avenue for future research.

Additionally, the choice of comparison group could be viewed as a limitation. First, the control group received nondigital cognitive stimulation, and it is possible that a generational effect may make older adults naturally more inclined toward traditional activities than VR. Second, this control group effectively enhanced well-being and social interaction. While this may have made it harder to detect the benefits of VR, it offers a meaningful comparison with an accessible, widely used intervention. Comparing the intervention to an active rather than passive control helps isolate its specific therapeutic effects. Ideally, including a passive control group would have allowed us to distinguish these effects from natural fluctuations in well-being and the impact of study participation. However, the active control group still provides valuable insights by assessing the intervention’s added value beyond general engagement effects and existing alternatives.

A further limitation is that the social interactions were assessed based on only 4 minutes of observation during the 25-minute interaction period. This relatively brief observation window may have limited the ability to capture the full range and variability of participants’ responses. Additionally, the absence of video recordings prevented subsequent detailed behavioral analyses and may have reduced the precision of the observational measures.

Another limitation lies in the difficulty of ensuring that participants remained focused on the VR content during the intervention. It would be valuable for future studies to include such a measure, for example, by using eye-tracking VR headsets.

Nonpharmacological VR interventions for BPSD are still evolving, and many knowledge gaps remain in the literature [45]. Our study contributes to this field by providing data from a randomized controlled trial in multiple nursing homes using an active control group and existing institutional resources (health care professionals as facilitators, available rooms, and standard equipment). This demonstrates the feasibility (and safety) of implementing such an intervention in nursing homes. With 53 participants, our sample exceeds most prior studies. A recent review found that, of 19 VR studies, 89.5% (n=17) had ≤40 participants, 26.3% (n=5) were randomized controlled trials, 5.3% (n=1) took place in nursing homes, and 36.8% (n=7) involved health care professionals as facilitators [17]. Our study, therefore, provides valuable insights using a rigorous methodology while maintaining ecological validity in real-world nursing home settings.

Beyond its clinical findings, this study also highlights the methodological challenges of conducting pragmatic intervention trials in nursing home residents with NCDs. Missing clinical data, variable attendance, difficulties in administering standardized cognitive assessments, and the need to minimize participant burden reflect the realities of conducting research in routine nursing home care rather than issues specific to this study. These challenges should therefore be considered not only limitations but also important considerations for the design of future pragmatic trials. Achieving methodological rigor in this population requires balancing internal validity with ecological validity to ensure that interventions remain both scientifically evaluable and feasible in routine clinical practice. Future studies should consider assessment tools applicable across the full spectrum of NCD severity (eg, the Clinical Dementia Rating), standardize the timing of postintervention assessments whenever possible, systematically document intervention exposure and participant engagement, and collect key clinical characteristics, including comorbidities and medication use, while balancing scientific needs with ethical and regulatory requirements regarding data minimization. Beyond evaluating VR itself, this study provides methodological insights that may help improve the design and conduct of future pragmatic trials in nursing home settings.

This study highlights the potential benefits of VR group interventions for individuals with BPSD, particularly in enhancing immediate well-being with minimal associated risks. However, rather than demonstrating superiority over conventional cognitive stimulation, our findings suggest that immersive VR represents a feasible and safe alternative that warrants further investigation through adequately powered pragmatic trials.

These findings suggest that immersive VR should not necessarily be viewed as a replacement for existing nonpharmacological approaches, but rather as an additional therapeutic option whose value may depend on individual characteristics and implementation strategies. Larger pragmatic trials are now needed to identify which residents are most likely to benefit and under which conditions VR can provide clinically meaningful advantages in routine nursing home care.