Protocol for a multicentre, randomised controlled trial of computerised cognitive rehabilitation in survivors of critical illness: the RETURN-III study

Abstract

Introduction

Cognitive impairment after critical illness is a common occurrence that has a profound impact on millions of people worldwide each year. Currently, no interventions after critical illness have been found to promote cognitive recovery. Cognitive training has shown promise in other populations (eg, community-dwelling adults), predominantly in improving the trained cognitive domain (eg, memory, reasoning or executive function). In pilot work, in-person cognitive training with intensive care unit (ICU) survivors was found to not only improve the trained domain but also may reduce disabilities in daily function. Computerised cognitive rehabilitation (CCR) is a more scalable option, which we have demonstrated the feasibility of use among ICU survivors. Whether CCR is efficacious in ICU survivors remains unknown.

Methods and analysis

The Returning to Everyday Tasks Utilising Rehabilitation Networks-III (RETURN-III) study is a multicentre, randomised, double-blinded, parallel group controlled trial testing the hypothesis that CCR versus active control will result in less cognitive impairment after critical illness. Adult patients who recently experienced acute respiratory failure or shock in medical and surgical ICUs at two hospitals in Nashville (TN, USA) will be enrolled. Participants will be randomised within 12 weeks of hospital discharge to a 12-week home-based programme of CCR (or control), with the goal of five 30 min sessions per week. In-person assessments will be performed at baseline, 3 months and 12 months after randomisation. The primary outcome is global cognitive function 3 months after randomisation using the Repeatable Battery for the Assessment of Neuropsychological Status (RBANS) score. Secondary outcomes will include 12-month RBANS global cognitive function, 3-month and 12-month subjective cognition by the Cognitive Self-Report Questionnaire, daily function related to cognition by the Functional Activities Questionnaire and processing speed/executive function by Trail Making Test part B. With an anticipated enrolment of 160 patients, the study will have at least 80% power to detect a 7-point or higher difference in RBANS cognition scores between the two groups (<0.5 SD based on age-adjusted population norms) with a type 1 error rate of 5%.

Ethics and dissemination

The RETURN-III study was approved by the institutional review boards at each participating site. The results will be submitted for publication in a peer-reviewed journal and presented at one or more scientific conferences following completion of data collection.

Trial registration number

NCT04353804.

STRENGTHS AND LIMITATIONS.

• Broad inclusion criteria of adult patients of varying ages and baseline health status admitted in medical/surgical intensive care units (ICUs) at two sites increase generalisability of the results (strength) although the necessity of being able to communicate in English and to have WiFi at home might limit generalisability to non-English speakers and people of lower socio-economic status.

• The double-blind design with the use of an active-control comparator allows for the assessment of the efficacy of the intervention while controlling for non-specific or placebo effects.

• Cognitive testing is performed with robust, multi-domain, objective, in-person assessments that are sensitive to small changes after critical illness.

• The intensive 12-week programme, in-person assessments and known high-rates of competing healthcare needs among ICU survivors introduces potential for attrition.

• The inclusion criteria do not include a cut-off for baseline (post critical illness) cognitive score, which might allow for enrolment of patients without cognitive impairment but also avoids excluding patients whose ‘normal’ scores still reflect a drop from their pre-ICU status (strength).

Introduction

Millions of adults survive an admission to an intensive care unit (ICU) each year, a number that has been increasing as acute medical care and life-sustaining therapies continue to improve.^1–3 Furthermore, while ICU care was previously more focused on survival, there has been a welcomed push over the past two decades to assess patient-centred long-term outcomes and quality of life following discharge deemed meaningful by the patient.^4–6 Cognitive function is core among these patient-centred outcomes, with qualitative studies suggesting that maintaining cognition after medical therapy or acute illness is a priority beyond survival alone.^{7 8}

Cognitive decline occurs in a third or more of ICU survivors over the months to years following their incident illness.^9–11 Cognitive impairment after critical illness commonly affects specific domains including memory, attention, processing speed and executive function.^{9 12–14} It is also associated with worse quality of life, greater disability in the instrumental activities of daily living (IADLs) and decreased return to employment.^15–17

Our group and others have shown that delirium during critical illness is one of the most robust and consistent risk factors for cognitive impairment after hospitalisation.^{9 18–20} Avoiding and mitigating delirium is currently the primary inpatient approach for managing long-term cognitive outcomes. This approach focuses on non-pharmacologic bundles of care that promote early mobility and light sedation to reduce days with delirium.^21–23 However, once a patient admitted to the ICU leaves the hospital, currently no therapies have been proven to reduce the burden of cognitive impairment.²⁴

Among the candidate interventions to improve ICU survivorship is cognitive rehabilitation. Cognitive rehabilitation uses principles of neuroplasticity to target specific cognitive domains (eg, executive function, memory, processing speed and visuospatial skills). Data on its efficacy have been inconsistent despite decades of research, largely resulting from immense heterogeneity in treatment population, intervention type and outcomes assessed (eg, trained task, closely related tasks and distantly related tasks).^{25 26} Nevertheless, large-scale trials have shown its potential in select populations. In 2002, the ACTIVE study demonstrated that in-person cognitive training with community-dwelling older adults can improve performance in the trained cognitive domains (eg, reasoning, memory and processing speed),²⁷ which was not only durable but led to fewer disabilities in IADLs up to 10 years later.^{28 29} Our group initially demonstrated that a similar approach with intensive in-person cognitive/physical rehabilitation is efficacious for improving executive cognitive performance and reducing disabilities in a small randomised trial of ICU survivors.³⁰ However, the scalability of such interventions is rather limited.

For this reason, computerised cognitive rehabilitation (CCR) has been proposed to reduce barriers and increase access. CCR can be performed at home using a tablet, mobile phone or computer. It has been demonstrated to improve cognitive performance in community-dwelling adults.^{31 32} In particular, evidence suggests benefit in cognitive domains of processing speed, attention, executive function and memory.^{33 34} Studies have also demonstrated an effect on brain structure, which suggests a benefit beyond task familiarity or practice effect.³⁵ In a pilot study to the currently presented trial, our group showed the feasibility of deploying CCR to ICU survivors in a non-randomised fashion.³⁶

While cognitive rehabilitation has shown some efficacy in older adults and is feasible in ICU survivors, it remains uncertain whether it will improve cognition and daily function in adults with recent critical illness who carry a high burden of healthcare utilisation, comorbidities and readmissions.^{37 38} We therefore are conducting a randomised trial to test the hypothesis that a 12-week programme of CCR in ICU survivors will reduce the severity of cognitive impairment compared with active control.

Objective 1

To test the hypothesis that a CCR intervention in medical and surgical ICU survivors will reduce the severity of cognitive impairment at 3 and 12 months post-randomisation compared with ICU survivors using non-specific control computer games.

Objective 2

To test the hypothesis that a CCR intervention in medical and surgical ICU survivors will improve key measures of integrative function (eg, IADLs) at 3 and 12 months post-randomisation compared with ICU survivors using non-specific control computer games.

Methods and analysis

This manuscript describes key elements of the study protocol and statistical analysis plan. This manuscript was prepared in accordance with the Standard Protocol Items for Randomised Trials (SPIRIT) statement.³⁹

Trial design and setting

The Returning to Everyday Tasks Utilising Rehabilitation Networks-III (RETURN-III) study is a multicentre, randomised, double-blinded, parallel group controlled trial testing the hypothesis that CCR versus active control (ie, non-specific control computer games) will reduce the severity of cognitive impairment after critical illness. This study recruits participants from a quaternary care academic medical centre (Vanderbilt University Medical Center (VUMC)) and a Veterans Affairs (VA) Hospital (Tennessee Valley Healthcare System – Nashville VA), both in Nashville, Tennessee, USA. Baseline and follow-up assessments will take place at various locations including clinics, research workspace and patient homes. Intervention/control programmes will take place at participants’ place of living and on their own daily schedule.

Patient and public involvement

While we did not have a formal patient committee or input specifically for this trial, our group frequently uses feedback and gained experiences from several of our ICU survivor support groups to inform research interventions, outcomes and dissemination of findings.

Eligibility

Eligible patients are adults (aged ≥18 years) with a recent ICU admission for acute respiratory failure or shock (ie, high-risk population for cognitive impairment)^{9 40 41} and who are no longer requiring ICU level of care. Acute respiratory failure and shock are determined by the study personnel based on chart review of respiratory support (high flow nasal cannula, non-invasive ventilation, invasive mechanical ventilation, extracorporeal membrane oxygenation (ECMO)), receipt of vasoactive medications or mechanical circulatory support (ie, intra-aortic balloon pump and ECMO). Patients are not considered to be in respiratory failure or shock if they receive these treatments for brief (<12 hours) postoperative management without other evidence of clinically relevant respiratory failure or shock. Critical care-trained physicians (MLR, TJM, MFM and MBP) either determine the presence of respiratory failure or shock or are available for consultation when clinical uncertainty arises.

Patients who have pre-existing severe cognitive impairment based on chart documentation or through surrogate report via the validated Informant Questionnaire on Cognitive Decline in the Elderly (IQCODE),⁴² are unwilling to commit to the intervention, are under consideration for hospice or who have personal situations or medical conditions that limit their ability to participate in CCR are excluded. Full inclusion/exclusion criteria are shown in Box 1.

Box 1. Eligibility criteria.

Inclusion criteria

• Adults (aged ≥18 years) with a recent ICU stay (medical or surgical) requiring treatment for respiratory failure and/or shock (ie, high-risk population)

• No longer requiring ICU level of care

Exclusion criteria

• History of pre-existing severe cognitive impairment (documentation in medical record or score of IQCODE ≥3.8

• Unwilling to commit to participation in the intervention

• Under consideration for hospice

• Primary residence over 100 miles from the enrolling site if the patient is unwilling to return to the enrolling site for follow-up

• Homeless without secondary contact available

• Severe substance abuse or neuropsychiatric disorder of a severity that prevents independent living

• Active suicidal ideation

• Any past or present behaviour that may be deemed a safety risk for follow-up

• Blind, deaf or unable to understand/communicate in English

• Required ICU level of care <24 hours

• Not capable of completing computer-based training*

• Co-enrollment in another study, determined on a case-by-case basis

• Incarcerated

Screening, recruitment and consent

Screening and initial recruitment will be performed during the incident ICU hospitalisation through screening of ICU census list via electronic health records. Once identified as eligible, the study personnel will approach patients during their admission to determine interest in the study or send electronic health record invitations for patients already discharged. If interested, patients will be followed up by phone, text or additional electronic health record message no later than 12 weeks after discharge to obtain written informed consent (online supplemental file 1). Extensions to this timeline are allowed for recurrent hospitalisations and will be reported in the final manuscript. Consent will be obtained by participant only without allowing surrogate consent, as participants unable to provide consent universally fall into an exclusion criterion (see online supplemental file 2). Recruitment will also occur at the VUMC Recovery Clinic and the VUMC Lung Transplant Clinic, both outpatient clinics that frequently see eligible adults with recent critical illness. The study timeline and participant flow are summarised in figure 1.

Figure 1. Flow diagram of study activities including screening, enrolment, randomisation, intervention/active control programme and follow-up assessments. Created in BioRender. CSRQ, Cognitive Self-Report Questionnaire; FAQ, Functional Activities Questionnaire; RBANS, Repeatable Battery for the Assessment of Neuropsychological Status; TMT Part B, Trail Making est Part B.

Intervention

The intervention arm will receive a 12-week programme of CCR (BrainHQ by Posit Science). The CCR programme will contain eight different games that each target a specific cognitive domain, including sensory processing speed, attention, memory, visual–spatial skills and executive function (see online supplemental appendix for a full list and descriptions). All exercises adapt difficulty based on participant performance, providing a patient-specific ‘dose’ of rehabilitation meant to induce brain plasticity by creating a discrepancy between cognitive ability and demand.³⁵ Participants will be asked to complete 30 min of games for 5 days a week during the 12-week intervention period. Games will be presented during sessions in a random order, and participants must achieve a certain score prior to moving on to the next game.

Comparator

The active control group will receive a 12-week programme that utilises eight commercially available computer games (eg, Sudoku, Word Search II and Bricks Breaking Hex; see online supplemental appendix for a full list and descriptions). This will allow for face validity of cognitive intervention to maintain blinding of participants. These active control games will not target specific domains and will not have an adaptive algorithm-driven difficulty, which is part of what makes them acceptable for active control. Participants will be asked to complete 30 min of games for 5 days a week during the 12-week intervention period. Games will be presented to participants randomly, and participants must play at least 5 min before moving on to the next game.

Randomisation and blinding

Following informed consent documentation, patients will be randomised in a 1:1 ratio using permuted block sizes and stratified by days of delirium from their recent ICU admission (0 or 1 day, 2 or 3 days and ≥4 days), education level (<12 or ≥12 years of education) and age (<60 or ≥60 years). These variables were chosen for their strong association as risk factors for cognitive impairment after critical illness.^{9 18 43 44} The randomisation scheme was created by an external statistician and uploaded into the REDCap randomisation module.⁴⁵

All participants can access their CCR or active-control programme via a loaned or personal tablet. Treatment assignments will be known only to the enrolling study personnel who will have no role in the follow-up assessment of outcomes. Thus, the patients and follow-up study personnel will be blinded (ie, double-blinded). Unblinding is not allowed, as there is no foreseeable benefit to the participant’s health. Blinding of participants will not be actively assessed, and participants in the active control arm might recognise the games (eg, Sudoku). However, they would not know if these games have been adjusted or intensified for cognitive training, thus maintaining an adequate level of blinding.

Compliance and withdrawal

Using the CCR platform, the study team will remotely monitor the number of levels played per week per participant to track compliance. Participants who do not play any levels for 1 week will be sent a text reminder. In the event of recurrent acute illness or life events that prevent intervention/control adherence, up to two additional weeks will be allowed in their training schedule. Open-label use of concurrent brain health therapies by the participant or prescribed by a clinician during the trial period will be allowed. Clinician-administered cognitive rehabilitation during the intervention period will be tracked study-wide as part of self-reported follow-up questions at the 3- month assessment.

Study participants may stop their intervention/control programme or completely withdraw at any time point with or without stated reason. Participants who play zero games for a continuous 4-week period will be administratively removed from active study. Participants who choose to stop intervention/control without fully withdrawing from the study or who are administratively removed from active study will be offered the option of completing the subsequent 3-month assessments.

Primary outcome

The primary outcome is cognitive impairment as measured by the global cognitive score on the Repeatable Battery for the Assessment of Neuropsychological Status (RBANS)⁴⁶ 3 months after randomisation.

RBANS is a thorough and robust individually administered test designed to comprehensively assess multiple domains of neuropsychological status in adults. It is typically administered in 45 min and is composed of 12 subtests (list learning, story memory, figure copy, line orientation, digit span, coding, picture naming, semantic fluency, list recall, list recognition, story recall and figure recall), generating scores in five cognitive domains (immediate and delayed memory, attention, language and visuospatial/constructional abilities), which combine to form a Global Index Score. The RBANS Global Index Score has age-adjusted population norms with mean±SD of 100±15 (for ages 20–89) for global cognition and has been validated in several populations with correlation to functional outcomes.^{47 48}

Secondary outcomes

Secondary outcomes are as follows:

• Disability in IADLs measured by the Functional Activities Questionnaire (FAQ) score 3 and 12 months after randomisation (key secondary outcomes)

• Cognition as measured by the RBANS global cognition score 12 months after randomisation

• Trail Making Test (TMT) Part B score 3 and 12 months after randomisation

• Cognitive Self-Report Questionnaire (CSRQ) score 3 and 12 months after randomisation

The FAQ⁴⁹ is a 10-item questionnaire regarding IADLs, for example, cooking, medication management and balancing finances, which are considered necessary for independent living. Each question is scored on a 4-point ordinal scale (total scores range from 0 to 30) with higher scores indicating more disabilities.⁵⁰ TMT Part B is a common neuropsychological test that provides information on visual search, scanning, speed of processing, mental flexibility and executive function by requiring the participant to draw lines sequentially through 25 encircled, alternating numbers and letters.⁵¹ Scores represent time in seconds to complete the test with a maximum score of 300. CSRQ contains 24 scored and one unscored statement on cognition and mood during the preceding 2 weeks. It is answered on a 5-point Likert scale (total score, 0–120), with higher scores indicating worse cognitive symptoms.³¹ Subscales include cognition, hearing, and social, the latter two meant to overlap with cognitive impact on speech-processing and social engagement, respectively (table 1).

Table 1. Outcomes.

Outcome	Assessment	Description
Primary	RBANS global score at 3 months	Composite score of 5 cognitive domains
Secondary	Disabilities in IADLs by Functional Activities Questionnaire at 3 and 12 months*	10-item questionnaire related to cognitively demanding daily tasks
	RBANS global score at 12 months	Composite score of 5 cognitive domains
	Trail Making Test Part B at 3 and 12 months	Evaluation of processing speed and executive function
	Cognitive Self Report Questionnaire at 3 and 12 months	25 statements answered on a 5-point Likert scale regarding self-perceived cognition during preceding 2 weeks

^*Represents key secondary outcomes of integrative function. All time points are in relation to randomisation.

IADLs, instrumental activities of daily living; RBANS, Repeatable Battery for the Assessment of Neuropsychological Status.

Data collection

Patient characteristics and baseline data will be collected by electronic health record chart review and during the baseline assessment by surveys, questionnaires and cognitive testing (figure 1). Variables from the baseline assessment will be assessed again at 3 and 12 months, along with collection of any new health status since the prior assessment (eg, hospitalisation or surgery).

Trial risks

The risk of loss of privacy, which occurs in all research studies, will be minimised by storing all data on a secure, password-protected, online platform (REDCap).⁴⁵ Otherwise, the risk of CCR is minimal. Emotional frustration due to game difficulty or technical challenges is possible; this risk will be mitigated by providing participants with access to study personnel for troubleshooting of software.

Sample size and power analysis for primary and key secondary outcomes

To test the hypothesis that CCR will result in less decline in global cognition 3 months after randomisation compared with control as measured by the RBANS, a total of 160 ICU survivors will be recruited and randomised. This original sample size incorporated an estimated 10% study attrition and loss to follow-up, thus giving at least 144 participants available for follow-up assessments. This would provide us with at least 80% power to detect a 7-point difference or greater (less than one half SD) in RBANS global cognition score, which is clinically meaningful.⁵² Due to heightened attrition rates, an additional power calculation is also added for the complete-case analysis as described in that section.

At a two-sided significance level of 5% and assuming a normal distribution for FAQ (mean 0.06±0.24), the estimated 144 participants available for assessments would provide us with at least 80% power to detect a change of ≥0.1 units in the FAQ scale, between the intervention and control groups. The sample size computations were performed, assuming normal distribution for these outcomes using two-sample t-test and a type I error rate of 5%.

Statistical principles

Participant flow diagram of this trial will be reported as specified by the Consolidated Standards of Reporting Trials (CONSORT) 2025 guidelines.⁵³ The flow diagram will display all stages of the trial including screening, exclusions, enrolment, randomisation, death, study withdrawal and follow-up. Demographic information, clinical characteristics at baseline and study outcomes will be presented for the intervention and control groups. Median and IQR will be used for describing continuous variables, and frequency (%) will be used for describing categorical variables.

No hypothesis testing for the differences in baseline characteristics between the treatment groups will be conducted as recommended by the guidelines for reporting parallel group randomised clinical trials.⁵³ Statistical significance level will be set at 5% for all outcomes. All effect estimates will be reported with 95% CIs. No multiple comparison adjustments will be made for secondary outcomes following the standard practice of analysing multiple prospective clinical trial outcomes.⁵⁴

Missing data will be imputed using clinical imputation rules when applicable and model-based imputation otherwise. When conducting model-based imputation, predictive mean matching will be used within the Hmisc and rms packages with 20 imputed datasets. Variables included in the analysis model will be included in the model-based imputation model. Rubin’s pooling rule will be used. Since separate models are considered for both time points, missingness will be considered intermittent and missing at random. Any additional component of the imputation process will be reported in detail.

Primary analysis

Analysis population will be based on the modified intent-to-treat principle for all outcomes. The analysis cohort will include participants who were randomised and received at least a single day of intervention or control.

Primary and secondary outcomes will be analysed using a multivariable regression model, which will include the treatment group and the baseline measure of the corresponding outcome (eg, baseline RBANS global cognition score for the RBANS global cognition model).

Proportional odds logistic regression (POLR) will be used for modelling skewed continuous outcomes. Proportional odds assumption will be checked graphically using multiple cut-offs of the outcome, that is, binary logistic regression models will be fit using different cut-offs of the outcome, and the log ORs (estimated coefficients) will be plotted against cut-offs to check proportionality assumptions. ORs and 95% CIs will be reported for POLR models.

Simple linear regression will be used for continuous outcomes that are normally distributed and satisfy the model assumptions, which will be checked graphically using residuals versus fitted values plots and quantile–quantile plots. Regression coefficients along with 95% CIs will be reported for linear regression.

Restricted cubic splines with three knots will be incorporated into the multivariable regression models (when appropriate) to model non-linear relationships between the outcome and the continuous covariates. Robust standard errors will be used to account for heteroscedasticity of variances due to clustering within sites. R V.4.5.2 (or above) will be used for all analysis.

Sensitivity analysis

To evaluate the robustness of our results to imputation of missing outcomes, sensitivity analysis will be conducted for the primary and secondary outcomes including only participants with at least partial assessments (defined as either partial or full completion of outcomes specified in this statistical analysis plan) at that time point (complete case analysis). At the time of this protocol paper, attrition will be higher than initially expected, approaching 30%–40%. If that occurs, our complete-case analysis will still include 96–112 participants with primary outcomes, giving that analysis 80% power to detect a difference of 8.0–8.6 points in RBANS between groups.

Trial monitoring

Given the minimal risk nature of this intervention and moderate size study, there is no Data Safety and Monitoring Board or interim analysis.

Protocol amendments and recruitment expansion

The details described above reflect RETURN-III study protocol version 16, dated 22 December 2025. We chose the timing of this protocol publication to adequately reflect the protocol that most closely aligns with eventual trial results. In this section, we describe amendments that impact the primary trial manuscript, and we do not include several amendments that are related to nested sub-studies unless there is potential to affect the protocol or variables described in this manuscript.

The initial protocol, dated 14 February 2020, was approved by the Department of Veterans Affairs Medical Centre Tennessee Valley Healthcare System’s IRB. Trial initiation was postponed because of the COVID-19 pandemic. Trial enrolment began in October 2022 using the initial protocol, which enrolled and randomised patients during their hospitalisation for critical illness. After noting a high attrition rate with enrolment very early in critical illness, we submitted a protocol dated 1 June 2023, in which initial recruitment would still occur during hospitalisation, but we would re-contact patients for consent and baseline scheduling within 12 weeks of hospital discharge to allow some physical and medical recovery (figure 1).

We submitted a protocol (VUMC version 1.0) dated 6 June 2023 to the IRB of VUMC to expand recruitment to this site. Recruitment initially began at VUMC on 27 November 2023 in the ICU Recovery Clinic, a once weekly outpatient clinic for post-ICU follow-up care.⁵⁵ We subsequently expanded recruitment 1 February 2024 to inpatients who had recently survived critical illness during that same hospitalisation, following exact procedures as the Nashville VA. VUMC Protocol Version 11 dated 21 January 2025 allowed for recruitment from the VUMC lung transplant clinic.

Ethics and dissemination

This trial has been reviewed and approved by the institutional review board (IRB) at each participating site: IRB of VUMC and IRB of VA Tennessee Valley Healthcare System. We plan to disseminate the findings by publication following completion of primary outcome data collection. The complete trial protocol and statistical analysis plan can be accessed via email to the investigators. Sharing of a de-identified dataset will be allowed pending request and project review.

Trial enrolment began 3 October 2022 and is anticipated to end April 2026, with 12-month outcome data collection through May 2027. Data analysis, manuscript preparation and submission will occur during the subsequent months with anticipated publication by end of 2027.