This is a single-center, 3-arm, parallel-group randomized pilot trial conducted in surgical ICUs at Houston Methodist Hospital in Houston, Texas. Participants will be randomized in a 1:1:1 ratio to ReCognitionVR 2.0 cognitive stimulation games, passive VR control exposure, or standard-of-care nurse-led reality orientation. All participants will continue to receive usual postoperative and ICU care.

The protocol is PRO00041199, modified version 2 dated April 13, 2026. The protocol lists US Food and Drug Administration investigational device exemption G260027 and classifies the study as a nonsignificant risk device study under 21 CFR 812.3(m). The trial is a behavioral and device-enabled feasibility study. It does not involve an investigational drug, biologic, dietary supplement, or collection of biological specimens.

The study is designed as a pilot and feasibility trial rather than an efficacy trial. The analytic emphasis is on estimating recruitment, retention, outcome completion, safety, tolerability, usability, acceptance, adherence, and variability in cognitive and delirium-related measurements. Reporting will be informed by SPIRIT (Standard Protocol Items: Recommendations for Interventional Trials) guidelines (Checklist 1), the CONSORT (Consolidated Standards of Reporting Trials) extension for pilot and feasibility trials (Multimedia Appendix 1), and CONSORT-EHEALTH (Consolidated Standards of Reporting Trials of Electronic and Mobile Health Applications and Online Telehealth) guidance for digital health interventions.

Figure 1 shows the study schema, and Table 1 summarizes the study-scheduled activities. Eligible participants are adults aged 65 years or older who will be undergoing elective or emergency major surgery with expected postoperative ICU admission for approximately 3 days. Participants must have mild-to-moderate cognitive impairment, defined by a Montreal Cognitive Assessment (MOCA) score of 15 to 24 [34]. This range identifies patients at increased risk of delirium who may still be able to interact with a low-movement cognitive intervention.

Key exclusion criteria include preexisting intubation and invasive mechanical ventilation, severe mental illness, admission for drug overdose, severe vertigo, history of severe postoperative nausea and vomiting, hearing or visual impairment that would preclude VR use, severe dementia defined by a MoCA score below 15, active delirium at screening, and active suicidal ideation or behavior. Participants who require prolonged postoperative intubation for more than 24 hours will be considered screen failures after surgery.

Potential participants will be identified through surgeon clinics, preoperative workflows, and hospital admission lists. Screening occurs 1 to 30 days before surgery, with final eligibility confirmed before enrollment. Written informed consent will be obtained from the participant or a legally authorized representative (LAR) before study-specific procedures are performed.

Because the intervention is preventive, patients with active delirium at screening will be excluded. Suicidality screening will include medical record review and the Columbia-Suicide Severity Rating Scale Screen [35]. Patients with evidence of active suicidal ideation or behavior will not be enrolled and will be referred for clinical evaluation according to institutional policy.

The study population includes older adults with mild-to-moderate cognitive impairment; therefore, decisional capacity will be assessed when a LAR is not available. Capacity assessment will address the participant’s ability to understand the study purpose and procedures, appreciate personal implications of participation, reason about alternatives and consequences, and communicate a voluntary and consistent choice.

For participants who lack full decisional capacity, written informed consent will be obtained from a LAR before enrollment. Assent will be sought from participants who retain sufficient ability to understand the study at a basic level. Study personnel will use simplified language to explain the study’s purpose, procedures, the voluntary nature of participation, and the right to stop. Behavioral dissent, verbal refusal, distress, or resistance will be respected regardless of LAR consent. Assent or dissent will be documented in the study record.

A meeting review after enrollment began identified the need for standardized communication regarding research MoCA results. Research MoCA scores will not be used to make a clinical diagnosis. If a participant requests their result and the institutional review board permits disclosure, the result will be accompanied by written language explaining that preoperative anxiety, illness, and the testing context can affect performance, and that follow-up with primary care or neurology may be appropriate if concerns persist after recovery.

After eligibility confirmation and baseline assessment, participants will be randomized in a 1:1:1 ratio using simple computer-based randomization implemented through the REDCap randomization module [36]. Simple randomization was selected because the trial is a small pilot study in which complex stratification could increase operational burden and create sparsely populated strata. Participants and intervention staff cannot be blinded to VR exposure. Participants will be blinded to the study hypothesis. Outcome assessors administering cognitive and delirium measures will remain blinded to group assignment. Data analysts will be blinded to allocation to the extent feasible, recognizing that usability data apply only to participants exposed to VR.

Figure 2 shows the iterative development model from the prototype to ReCognitionVR 2.0, representing a major redevelopment of an earlier ReCognitionVR prototype evaluated in preliminary feasibility studies [32,33]. The upgrade was guided by human-centered design principles, prior bedside feasibility experience, staff feedback, engineering review, and ICU workflow constraints. The goal was to reduce operational complexity, improve intervention fidelity, and increase suitability for older hospitalized and critically ill patients.

Key refinements included a simplified launch interface, reduced setup burden, improved controller usability, decreased dependence on real-time headset casting, and enhanced device-generated performance capture. In this context, casting refers to wirelessly mirroring the VR headset display to an external device, such as a tablet or bedside monitor, so staff can observe what the participant sees during the session. Although casting can support staff observation and troubleshooting, it may be affected by variability across hospital networks, device pairing challenges, audio or visual lag, and additional workflow demands. Therefore, ReCognitionVR 2.0 prioritized direct in-app data capture and dashboard-based monitoring to reduce reliance on continuous live mirroring during bedside implementation.

A central methodological advance is the automated capture of game performance data. The platform records game difficulty, head movement, controller movement, and movement of virtual objects during each game, from which performance-indicative metrics may be derived.

The engineering review documented that headset and controller position and rotation data can be captured at approximately 60 Hz, with time-stamped game events such as balloon spawn, first visibility, hover, user pop, and auto-pop. These data reduce reliance on manual recording and permit exploratory derivation of engagement, attention, latency, inattention, completion efficiency, and performance-stability metrics.

The meeting-based usability review also identified future software development priorities that are not protocol-changing interventions for the current trial. These include more explicit virtual-nurse instructions for easy, medium, hard, and survival modes; fixed forward-facing interaction zones; reduced extraneous cognitive load; and potential future task modules targeting selective attention, inhibition, working memory, rule switching, processing speed, and error monitoring. These items will inform future development of ReCognitionVR 3.0 and a subsequent R01-stage protocol.

The primary outcome is the feasibility of the trial and intervention. Feasibility will be assessed through recruitment rate, outcome data completion, safety and tolerability, usability, acceptance, and adherence. Recruitment will be summarized by the numbers approached, agreeing, and consenting, as well as by monthly accrual. Outcome data completion will be quantified using 2 prespecified denominators. The primary completion metric will be the number of completed protocol-scheduled assessments divided by the number of assessments that became due from randomization through the end of follow-up. An assessment will be considered due if its scheduled time point occurs before the earliest of hospital discharge, death, withdrawal of consent, or study discontinuation; assessments that are not due because follow-up has ended will be excluded from the denominator. Assessments that are due but not completed because of clinical instability, participant refusal, competing clinical care, logistical barriers, or inability to complete the assessment will remain in the primary denominator and will be categorized by reason for noncompletion. As a secondary descriptive metric, evaluable-assessment completion will be calculated among due assessments for which the participant met prespecified assessment-specific eligibility criteria (eg, adequate arousal and ability to respond for cognitive testing, and VR exposure for usability and acceptance measures). Safety and tolerability will be summarized using AE rates, session interruptions, and short-interval changes in vital signs. Table 2 explains the overall outcome and measurement strategy, and Table 3 summarizes the feasibility benchmarks for progression planning.

Usability will be assessed using the System Usability Scale among VR-exposed participants [44]. Acceptance will be assessed using a 4-item Behavioral Intention Scale at hospital discharge [45,46]. Adherence will be defined as completion of at least 2 VR sessions in accordance with the assigned VR protocol, with dose modification when clinically indicated.

Secondary outcomes estimate preliminary effects on attention and executive function. The Months Backward Test will be assessed using both a pass/fail threshold and the number of correct months recited before the first error [47]. The NIH Toolbox Flanker Inhibitory Control and Attention Test will provide measures of accuracy and reaction time [48].

Device-generated performance metrics will be treated as exploratory digital end points as described in Table 4. Prespecified and candidate metrics by difficulty level include session duration, sensitivity to signal presence, reaction-time summaries, idle time as a potential marker of inattention, commitment percentage, head-gaze-derived attention stability, and engagement trajectories across epochs (beginning, middle, and end of the session).

The exploratory clinical outcome is the incidence of postoperative delirium. Delirium will be assessed using the CAM-ICU after arousal is confirmed with the RASS [38,43]. Delirium incidence is defined as at least 1 CAM-ICU-positive episode from postoperative days 1 through 7 or until hospital discharge, whichever occurs first.

Screening assessments include informed consent or LAR consent, assent when applicable, medical and surgical history, the MoCA, the RASS, the CAM-ICU, physical examination a review of concomitant medications, and suicidality screening. On postoperative day 1, participants will undergo baseline cognitive assessment using the Months Backward Test and the NIH Toolbox Flanker Inhibitory Control and Attention Test, then be randomized and, if clinically eligible, receive the first assigned intervention. VR sessions and safety assessments will occur on postoperative days 1 through 3. Delirium assessments will continue through postoperative days 1 to 7 and at hospital discharge. Usability and acceptance measures will be collected at discharge (Table 1).

Study data will be entered into REDCap using standardized case report forms [36]. Data elements include recruitment and consent metrics, eligibility criteria, intervention delivery, adherence, vital signs, AEs, cognitive assessments, delirium assessments, usability surveys, acceptance measures, and device-generated VR performance metrics. Identifiable data will be stored separately from study data using unique participant identifiers. Data quality will be supported by staff training, standard operating procedures, source data verification, REDCap query resolution, protocol deviation tracking, and periodic review as part of the study quality-control process. Blinded outcome assessors will receive standardized training in cognitive and delirium assessments. Case report forms and intervention logs will be reviewed for completeness, consistency, and timeliness.

The planned sample size is 90 participants, with 30 participants per arm. This sample size was selected to estimate feasibility metrics with clinically useful precision, generate preliminary estimates of variability and effect size for cognitive outcomes, evaluate safety and tolerability signals, and inform design assumptions for a future efficacy trial. No formal hypothesis-testing power calculation was performed because the trial is designed primarily to estimate feasibility.

Analyses will follow the intention-to-treat principle and primarily focus on feasibility and the estimation of intervention effects rather than on formal hypothesis testing. Feasibility outcomes will be summarized descriptively using counts, percentages, means with SDs, medians with IQRs for continuous variables, along with 95% CIs, as appropriate. Feasibility benchmarks will be interpreted as progression criteria rather than formal hypothesis tests.

For cognitive outcomes, between-group differences at follow-up will be estimated using effect sizes and 95% CIs. Linear mixed-effects models may be used to account for repeated measurements and within-participant correlation, including a random intercept for participant. Models will adjust for baseline cognitive performance and, where sample size permits, a limited set of clinically relevant covariates to avoid overfitting.

Device-generated performance metrics will be summarized descriptively and explored as indicators of intervention engagement, fidelity, and tolerance. Where appropriate, these metrics may also be examined in relation to clinical outcomes in exploratory analyses.

For delirium incidence, odds ratios and 95% CIs will be estimated using logistic regression if event counts permit. Candidate covariates include age, baseline cognitive function, sex, race or ethnicity, surgery type, anesthesia duration, comorbidity burden, and other clinically relevant delirium risk factors. If event counts are sparse, penalized or exact logistic regression methods may be considered, or analyses may be limited to descriptive comparisons.

Missing data will be described by study arm and outcome domain. Sensitivity analyses may examine worst-case assumptions for binary outcomes and model-based approaches for continuous outcomes (eg, mixed-effects models using maximum likelihood), which are valid under missing-at-random assumptions.

A protocol deviation involving the omission of the instructional video for one participant in a VR intervention will be flagged in the analytic dataset. If this participant’s VR performance metrics differ materially from those of other participants, sensitivity analyses will assess the impact of including versus excluding these data in exploratory digital performance analyses, subject to IRB and analytic review.

The study will be conducted under Houston Methodist Institutional Review Board (IRB) approval and in accordance with Good Clinical Practice, the Declaration of Helsinki, institutional human-subjects protection policies, and applicable regulatory requirements. The modified protocol is version 2, dated April 13, 2026. The protocol identifies the study as a nonsignificant risk device study under 21 CFR 812.3(m) with FDA IDE G260027. Deidentified data sharing will comply with institutional policy, IRB requirements, and National Institutes of Health data-sharing expectations.

Results will be disseminated through peer-reviewed publication and scientific presentation. Authorship will follow the International Committee of Medical Journal Editors criteria.

This protocol describes a randomized pilot trial of ReCognitionVR 2.0, an immersive VR-based cognitive stimulation intervention designed for older surgical ICU patients with mild-to-moderate cognitive impairment. The study addresses a major implementation gap in delirium prevention: the need for scalable, engaging, standardized, and monitorable nonpharmacologic interventions that can be delivered within high-acuity postoperative care [15,49].

The principal innovation is the integration of orientation, environmental modulation, low-movement interactive tasks, adaptive difficulty, and device-generated performance capture into a single bedside workflow. The transition from the original prototype to ReCognitionVR 2.0 is methodologically important because it reflects iterative refinement informed by feasibility data, workflow barriers, and human factors considerations. The revised platform reduces setup complexity, limits reliance on network-dependent casting, simplifies staff-facing operation, and provides objective performance data for future end point development.

The protocol treats human-factors outcomes as central feasibility outcomes. Usability, acceptance, adherence, session interruption, task completion, physiologic tolerance, and staff workflow compatibility are essential when introducing technology into critical care. An intervention that is cognitively plausible but operationally fragile is unlikely to be scalable. By measuring these domains prospectively, the trial will clarify whether immersive VR cognitive stimulation can be implemented with sufficient fidelity in a vulnerable postoperative ICU population.

The passive VR control arm strengthens the design by distinguishing active cognitive stimulation from nonspecific effects of headset use, novelty, immersion, staff attention, and time in a virtual environment. This distinction is important for future interpretation of efficacy. If cognitive or delirium-related signals emerge, comparison with passive VR exposure will help determine whether the signal is attributable to active task engagement rather than VR exposure alone.

The safety framework is intentionally conservative. Older postoperative ICU patients with cognitive impairment are susceptible to postoperative pain, dizziness, nausea, visual discomfort, hemodynamic instability, fatigue, delirium, and clinical interruptions [50-56]. The protocol mitigates these risks by requiring a calm and alert state before each session, conducting all sessions in bed, restricting tasks to low-amplitude upper-extremity movement, monitoring vital signs before, during, and after sessions, allowing dose modification, and suspending VR exposure if delirium develops. Early experience further underscores the importance of readiness checks, particularly oxygen-connection verification and confirmation of clinical stability before headset placement.

The study has limitations. It is a single-center study and not powered to test clinical efficacy. Participants and intervention staff cannot be blinded to VR exposure. Clinical heterogeneity across surgical populations may influence cognitive recovery and the risk of delirium. Simple randomization may lead to baseline imbalances across arms, although this is acceptable for a feasibility study and will be considered in interpretation. Several VR-derived measures, including head gaze–based attention stability and engagement trajectories, are being explored and require empirical validation before they can be used as efficacy end points. Finally, delirium incidence is included as an exploratory outcome and should be interpreted as hypothesis-generating.

This study is positioned to generate essential information for the next phase of investigation. It will determine whether older surgical ICU patients with cognitive impairment can be recruited, randomized, exposed to immersive VR, monitored safely, and assessed with sufficient completeness. It will also provide estimates of intervention tolerance, device usability, adherence, variability in cognitive outcomes, feasibility of delirium ascertainment, and operational barriers. Meeting-based feedback has already identified future software development priorities, including more autonomous, avatar-guided instructions, additional cognitive domains, and enhanced digital phenotyping. These data will inform optimization of ReCognitionVR 3.0 and design of a future adequately powered multicenter trial.

This randomized pilot trial will evaluate the feasibility, usability, safety, acceptance, adherence, and preliminary cognitive and delirium-related outcomes of ReCognitionVR 2.0 in older surgical ICU patients with mild-to-moderate cognitive impairment. By combining immersive cognitive stimulation, bedside orientation, conservative safety monitoring, and device-generated performance metrics, the trial will provide critical methodological data for developing scalable VR-enabled delirium-prevention strategies in critical care.