Before conducting screening, recruitment, and data collection procedures, we obtained approval from the Institutional Review Board (IRB) at University Hospitals Cleveland Medical Center (IRB00001691), in accordance with the ethical standards of the Declaration of Helsinki. Participants provided written informed consent before data collection, understanding that all study data would be deidentified and stored securely in an encrypted REDCap (Research Electronic Data Capture; Vanderbilt University) database prior to analysis. We offered a US $25 gift card as compensation for participants who completed all study procedures.

Initially, we planned to conduct this study as a parallel group randomized trial; however, we encountered usability issues with the REFRAME intervention that prevented us from implementing it at the start of our recruitment period. Therefore, we modified our design to a pilot before-and-after (nonrandomized) study with a 1:1 group allocation ratio. We enrolled the first 20 participants in our control condition (usual care [UC]), followed by participants in the intervention condition (REFRAME). We oversampled the intervention group, aiming to complete all 3 intervention modules with 20 participants.

We defined UC as the communication and decisional support routinely provided to surrogate decision-makers of ICU patients. UC practices varied among clinical teams, ranging from brief interactions to extensive discussions. We recognize that research procedures, such as obtaining informed consent, may have heightened participants’ awareness of their emotional needs, but these processes were not considered a structured intervention.

REFRAME is a tablet-based app designed to support emotion regulation in ICU surrogates. The intervention consisted of 3 sequential modules corresponding to stages in the Extended Process Model of Emotion Regulation, with the user interface design grounded in Cognitive Load Theory [24,35]. Apart from incorporated assessment components (eg, surveys), all educational and instructional components were audio guided with corresponding closed captioning. For minor disruptions (<15 minutes) or attentional lapses, participants were allowed to repeat previous content from their desired location within the module. We required participants to restart the intended module after major interruptions (>15 minutes). Each module required 10-15 minutes to complete. To minimize the risk of contamination and minimize staff burden, bedside clinicians were not involved in delivering or reinforcing REFRAME and received no training in intervention principles.

We designed module 1 to evaluate the users’ beliefs about emotion regulation [36], establish a foundational knowledge base related to participants’ awareness of physiological and psychological symptoms of anxiety and depression, introduce the concept of reappraisal, and provide an example of how to apply it within the context of the surrogate role (Figure S1 in Multimedia Appendix 1). Module 2 further expands upon this knowledge by introducing users to different reappraisal strategies (ie, reconstruing and repurposing) with context-relevant examples (Figures S2 and S3 in Multimedia Appendix 1), guiding participants through an evaluation of the various applied examples, and establishing an implementation intention to use their preferred strategies when they become aware of symptoms discussed in module 1 [27,37,38]. Finally, module 3 integrates the principles of reappraisal to evaluate a hypothetical tracheostomy and feeding tube decision-making scenario from multiple perspectives: self-interested, projected interests, best interest, and substituted interest (Figure S4 in Multimedia Appendix 1) [39].

We enrolled adults (aged ≥18 years) recognized by the ICU team as the legally authorized representative for a decisionally incapacitated adult who had been residing in the ICU for at least 48 hours and was not expected to be discharged in the following 24 hours. We excluded participants who could not read or speak English because the study instruments and the study intervention were available only in English. We recruited all participants from the cardiothoracic, medical, neurological, and surgical ICUs at a tertiary medical center in northeast Ohio.

We determined the sample size for this pilot trial using recommendations for estimating variance and minimizing overall trial size in pilot and main trials. For small- to medium-effect sizes (δ=0.2-0.5), Whitehead et al [40] recommend a pilot trial sample size of 20 participants per arm to estimate variance with reasonable precision while avoiding unnecessary participant recruitment. Similarly, a large group of investigators recently reported small- to medium-effect sizes (Cohen d=0.24-0.39) of a cognitive reappraisal intervention on negative emotions across a multinational sample (N=21,644) [30]. Thus, we sought to recruit 20 participants per study arm to evaluate REFRAME’s implementation.

We used recommended items from the Phenotypes and eXposures Social Determinants of Health Toolkit to measure the surrogate’s age (years), gender identity, racial identity, marital status, household income, education, and employment status; we also used previous team-generated items to measure the surrogate decision-maker’s relationship to the patient, advance care planning experience, and decision-making experience [48].

We measured surrogates’ baseline perceived stress levels with the 10-item Perceived Stress Scale (PSS) [49]. Participants rated each item on a 5-point Likert scale ranging from “never” (0) to “very often” (4). We calculated total scores by summing item responses, with higher scores reflecting greater perceived stress. To align the PSS timeframe with the surrogate’s acute ICU experience and other study instruments, we modified its reflection period from “over the past month” to “in the last week.” This modification is supported by evidence of the PSS’s psychometric robustness over shorter timeframes [50], its strong correlations with baseline anxiety (r=0.68; P<.001) and depressive symptom severity (r=0.66; P<.001), and its excellent internal consistency (α=0.90) in our sample.

We recruited participants from the adult cardiothoracic, medical, neuroscience, and surgical ICUs at University Hospitals Cleveland Medical Center in Cleveland, Ohio. We followed a standardized, IRB-approved protocol to ensure consistency. Each day, a research assistant (RA) would screen the electronic medical records of adult ICU patients to verify their age (≥18 years) and ICU length of stay (≥48 hours). For these patients, the RA would confer with the patient’s bedside provider (ie, nurse, nurse practitioner, physician, or physician assistant) to confirm discharge status (≥24 hours) and evaluate the patient’s decision-making capacity using an IRB-designated assessment: (1) Is the patient alert and able to communicate? (2) Is the patient comfortable enough to communicate? and (3) Is the patient medically stable enough for an informed consent discussion? The patient was deemed incapacitated if any of the assessment parameters were not met. In rare cases of uncertainty from the provider or RA, the RA would consult the principal investigator (GP) for a designation. Upon confirming the patient criteria, the RA would consult the provider to identify and introduce the patient’s surrogate. Then, the RA would confirm the surrogate’s age (≥18 years), verify their ability to read and understand English, introduce the study, and ascertain their interest in participating. For interested surrogates, the RA would initiate an informed consent discussion within the patient’s room or a preferred location near the ICU.

The RA would confirm and notify the surrogate of their group allocation upon obtaining written informed consent. Next, they would administer the sociodemographic questionnaire, the psychological distress instruments, and other instruments not referenced in this study. This interview (T0) required about 20 minutes. The RA conducted 3 follow-up interviews (T1-T3) 24-48 hours after each preceding interview. For those in the UC group, the RA would administer the psychological distress instruments in person or by telephone; these interviews lasted 5-10 minutes. For those in the REFRAME condition, the psychological distress, acceptability, and appropriateness assessments were embedded within the interventional modules, with each module requiring 10-15 minutes for completion. The REFRAME modules were not scheduled relative to specific clinical milestones. Because this was a feasibility pilot, our priority was to evaluate implementation procedures and maintain comparable assessment timepoints between study arms, rather than synchronizing intervention delivery with specific clinical milestones.

We incorporated fidelity-enhancing strategies across 4 of the 5 domains outlined by the National Institutes of Health Behavior Change Consortium framework [51,52]. We did not measure the enactment of treatment-related skills because the primary focus of this study was on evaluating REFRAME’s implementation in surrogates.

Before our primary analyses, we examined descriptive statistics to assess variable distributions, identify potential outliers and influential cases, and evaluate assumptions for our quantitative analyses. We assessed patterns of missing data with descriptive summaries, Pearson chi-square tests, independent-sample t tests (2-tailed), and logistic regression.

Between August 2022 and June 2023, we screened 2587 patients for eligibility, with the majority being ineligible due to a pending discharge status (989/2587, 38.2%), an ICU length of stay of less than 48 hours (583/2587, 22.5%), or possessing capacity for decision-making (387/2587, 15.0%). Of the 95 eligible surrogates (3.7%), a total of 46 (48.4%) declined participation. During the 10-month recruitment period, we enrolled 49 surrogates (4.8 per month). The first 20 participants were allocated to UC, and the following 29 to REFRAME. Notably, 1 participant assigned to the REFRAME condition withdrew from the study after providing informed consent but before baseline data collection, resulting in a final analytic sample of 48 surrogates.

As shown in Figure 1, we experienced greater attrition in the REFRAME group (n=9) than in the UC group (n=3), with withdrawal reasons including patient discharge or death and the participant either being nonresponsive, feeling overwhelmed, or physically ill. Despite this, study group (χ²₁=1.8; P=.18) and all other proposed covariates were not associated with study attrition, suggesting our data were at least missing at random.

Table 1 compares surrogate baseline characteristics by study group. Overall, our sample’s average age was 58.6 (SD 13.2) years, and it was predominantly comprised of women (36/48, 75%) and White individuals (41/48, 85.4%). Most participants were married (33/48, 68.8%), employed full-time or part-time (27/48, 56.2%), nonbaccalaureate educated (26/48, 54.2%), and living in a household with an annual income of less than US $100,000 (25/48, 52.1%). Most surrogates had not discussed the patient’s preferences regarding cardiopulmonary resuscitation (32/48, 66.7%), mechanical ventilation (30/48, 62.5%), artificial nutrition (40/48, 83.3%), or dialysis (33/48, 68.8%), despite more than half reporting that they had discussed end-of-life care (25/48, 52.1%). More than 62.5% (30/48) of surrogates reported moderate to severe anxiety symptoms, and one-third (16/48, 33.3%) endorsed moderate to severe depressive symptoms. While the baseline characteristics of participants were mostly balanced across groups, 75% (15/20) of surrogates in the UC group were the patient’s spouse, compared with 39% (11/28) in the REFRAME group (Pearson χ²_1,48=6.0; P=.01). Furthermore, 95% (19/20) of surrogates in the UC group lacked experience in making health care decisions for the patient, a larger proportion than surrogates in the REFRAME group (19/28, 67.9%) (Fisher exact test, P=.03).

Table 3 displays within- and between-group changes in psychological distress severity from baseline (T0) to approximately 1 week postbaseline (T3). At T0, anxiety and depressive symptoms were higher in the REFRAME group (t₃₄=−1.69; P=.10 and t₃₄=−1.33; P=.19), and at T3, psychological distress severity decreased in both groups. However, anxiety symptoms decreased by an average of 2.2 points more in the REFRAME group (95% CI −2.08 to 6.53), and depressive symptoms decreased by an average of 4.01 points more in the REFRAME group (95% CI −0.01 to 8.02). These differences favoring the REFRAME group correspond to a small effect for anxiety symptoms (d=0.35, 95% CI −0.31 to 1.01) and a medium effect for depressive symptoms (d=0.68, 95% CI −0.001 to 1.35).

The findings from our sensitivity analysis align with those from our primary analysis (Tables S2 and S3 in Multimedia Appendix 2). Anxiety symptoms decreased in both groups from T0 to T3, with the REFRAME group reporting a larger decrease (b=−2.14, 95% CI −6.20 to 1.92). Of our prespecified covariates, perceived stress (b=1.38, 95% CI 0.99-1.78), identifying as a woman (b=−4.82, 95% CI −8.74 to −0.89), and having prior decision-making experience (b=−6.26, 95% CI −11.17 to −1.35) were positively associated with anxiety severity. Similarly, depressive symptoms decreased in both groups from T0 to T3, with the REFRAME group reporting a larger decrease (b=−4.06, 95% CI −7.91 to −0.21). Unlike the anxiety model, perceived stress was the only covariate associated with depression severity at P<.05 (b=1.36, 95% CI 0.93-1.79). Figure 2 depicts the model-estimated marginal means for changes in anxiety and depressive symptoms from T0 to T3, adjusted for all prespecified covariates, providing convergent evidence for the robustness of the observed patterns in our primary analysis.

Our findings should be considered with several limitations in mind. First, the small, homogeneous sample limits how well the results apply to more diverse populations, such as non-White men. Second, the nonexperimental design introduces selection bias because surrogates who declined participation due to distress may have benefited most from REFRAME, and observed improvements could partly result from regression to the mean, as baseline distress was higher in the REFRAME group. Additionally, participants’ awareness of their group assignment might have influenced their decision to consent, potentially inflating acceptability ratings. Third, while REFRAME aimed to reduce acute psychological distress, the short follow-up restricts the ability to assess long-term or delayed effects on decision-making and outcomes. Fourth, higher attrition in the intervention group could indicate scalability challenges in larger REFRAME evaluations. Fifth, we did not set predefined feasibility benchmarks for enrollment or consent rates, which limits formal comparisons across trials; however, the enrollment rate suggests that a future, fully powered randomized trial of REFRAME is feasible. Lastly, reliance on self-report instruments introduces the possibility of social desirability or recall biases, and the PROMIS measures might reflect immediate ICU-related distress rather than chronic psychiatric symptoms, which should be considered when interpreting short-term changes in distress in this context.

We provide preliminary evidence that REFRAME can be successfully implemented and is acceptable to ICU surrogates during the acute phase of critical illness, with qualitative feedback supporting its value in normalizing their experiences and providing concrete strategies for managing distress. We also observed improvements in depressive symptoms that favored surrogates in the REFRAME group, although the nonrandomized design and small sample size prevent definitive conclusions about the intervention’s effects. Nonetheless, our findings highlight the practicality of delivering brief, on-demand emotional support to surrogates by tablet-based tools during this high-stress period. A larger, randomized trial with longer follow-up is needed to replicate our findings, evaluate its impact on surrogate decision-making, and examine its downstream influences on long-term psychological well-being. With continued refinement and more rigorous testing, interventions like REFRAME have the potential to improve psychological well-being in surrogates during critical moments of a patient’s ICU stay and to inform future strategies for supporting those making high-stakes medical decisions for a loved one with critical illness.