The Effects of Digital Health Interventions for Family Members in Intensive Care Units: Systematic Review and Meta-Analysis of Randomized Controlled Trials

Introduction

Patients admitted to the intensive care units (ICUs) frequently experience varying degrees of organ or system failure, which can be life-threatening [1]. Their families may endure enormous burdens and suffer from psychological disorders like anxiety, depression, and posttraumatic stress disorder (PTSD). The Society of Critical Care Medicine’s 2024 Clinical Practice Guidelines on Family-Centered Care for Adult ICUs underscore the need for delivering patient-centered and family-centered care in the ICU to uphold clinical excellence [2]. Nonetheless, various psychological challenges for family members, including anxiety, depression, acute stress disorder, PTSD, and complicated grief, are induced by the uncertainty of disease trajectories, the complexity of medical decisions, and the urgent demand for time. The term “postintensive care syndrome family” (PICS-F) encompasses these mental health disorders [3,4]. In addition to undermining family functioning and integrity, these problems may hinder the family’s ability to facilitate the patient’s recovery in the ICU [5,6].

A retrospective study [6] showed that the prevalence of psychological disorders related to PICS-F among family members of ICU patients within 6 months post admission was 15.1%, with spouses, as primary caregivers, exhibiting a heightened risk of psychological disorders. Furthermore, 1 research indicated that 10% to 69% of family members of ICU patients experienced moderate to severe psychological distress, including symptoms of anxiety, depression, and PTSD, potentially persisting for up to a year [7,8]. Family members endure not only emotional distress, such as anxiety and depression, but also are influenced by several biopsychosocial factors, including sleep deprivation, diminished quality of life (QoL) and mental health, and insufficient social support. These factors make it difficult for them to provide adequate family support to facilitate the patient’s recovery [5]. Common contributors to PICS-F among family members include female sex, relatively younger age, younger patient age, lower educational level, unmarried status, inadequate communication skills of ICU medical personnel, and the degree of family participation in decision-making [3,9]. Several strategies, such as ICU diaries [10], information support, and peer support, have been implemented to mitigate PICS-F but often focus on the patient’s hospitalization problems [11,12], and due to the family dynamics, geography, family resources, and family values, the effect of related intervention on PTSD, anxiety, and depression among family members is unclear [13]. Traditional educational tools are rarely implemented in the ICU environment due to high service costs, limited resources, and a lack of personalization [14,15]. With the rapid development of internet technology, mobile health care solutions have become popular, providing innovative psychological empowerment interventions [16].

Digital health technology (DHT) was initially introduced by the World Health Organization in the Global Strategy for Digital Health (2020-2024) and is defined as a comprehensive technical practice that facilitates clinical care, health-related education for patients and professionals, and public health through the extensive usage of electronic information and communication technologies [17,18]. Compared with conventional interventions, digital health interventions (DHIs) amalgamate computing platforms, connectivity, software, networked devices, and sensors—including smartphone apps, wearable devices, web-based communication platforms, videoconferencing, chatbots, artificial intelligence, and virtual reality (VR) technologies—leveraging technological advancements to transcend temporal and spatial limitations [19,20]. DHIs provide family members with greater accessibility, affordability, and convenience in receiving timely support and information, which is critical in the ICU context. In addition, DHIs help reduce health care spending [21,22]. DHIs have been shown to improve the psychological outcomes and QoL of caregivers, such as depressive symptoms, perceived stress, anxiety, and stress in other high-stress situations, such as caregivers of patients with cancer [23], dementia [24,25], and mental illness [15]. This study highlights the potential value of its application to family members of ICU patients. However, other studies have suggested that DHIs have little effect on alleviating the anxiety and depression of these family members [26]. Therefore, there are ambiguities in the existing research results on whether DHIs can effectively alleviate the psychological distress symptoms of family members of ICU patients [26].

After searching relevant databases, no meta-analysis of these intervention outcomes has been found. Randomized controlled trials (RCTs) continue to serve as the gold standard for delivering the most rigorous scientific findings [27]. Therefore, a systematic review and meta-analysis of RCTs focusing on the effects of DHIs on ICU family members is essential. This meta-analysis will assess the effects of DHIs on alleviating psychological symptoms, improving QoL, and enhancing communication quality among ICU family members, thereby providing critical evidence for the formulation of a comprehensive and effective psychological support program.

Methods

Research Design

This systematic review and meta-analysis were planned, conducted, and reported following the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-analyses) [28] (Multimedia Appendix 1). A systematic literature search, screening, study selection, data extraction, and risk-of-bias assessment were conducted according to the Cochrane methodology [29].

Search Strategy

The search strategy was reported in accordance with the PRISMA-S (Preferred Reporting Items for Systematic Reviews and Meta-analyses literature search extension) checklist (Multimedia Appendix 2). In total, 2 investigators (Chuchu Z and WJ) independently systematically searched 9 databases, including CNKI [China Knowledge Resource Integrated Database], Wanfang, VIP (Weipu Database), SinoMed, Cochrane Library, PubMed, Embase, CINAHL, and Web of Science, from inception to October 11, 2025. The search terms included “digital health,” “intensive care units,” “family,” and “randomized controlled trial,” which were pilot-tested and validated by independent researchers. No published search filters for study design were used. To identify additional studies, we manually searched the reference lists of included articles and relevant websites, including PROSPERO clinical trial registries. Before the final data analysis, the database searches were updated. Gray literature, such as conference abstracts and academic dissertations, was excluded. The whole search strategy is available in Multimedia Appendix 3.

Eligibility Criteria

Studies were selected based on the following inclusion criteria: (1) population: adult family members (≥18 y) of adult ICU patients (a family member could be a spouse, partner, friend, or relative [blood-related or not]); (2) interventions: all DHIs, which may include one or more methods, such as telephone, video, websites, or mobile apps, wearable devices, and others; (3) control: a comparison group that received either standard care, routine care or a blank control; (4) outcomes: the primary outcomes associated with the psychological health of family members included anxiety, depression, PTSD, and other related outcomes like QoL; and (5) study type: RCTs.

Studies were excluded if they met one of the following criteria: (1) articles were excluded if the DHI was used for monitoring or tracking purposes only; (2) duplicate publications, or (3) full text was not available.

Study Selection

The search results were imported into NoteExpress (Aegean Software) for review. After removing duplicates, 2 investigators (Chuchu Z and RY) independently screened the titles and abstracts, then reviewed the full texts of the remaining records. Any disagreements were resolved by discussion with the third investigator (WS). The κ coefficients were used to specify the consistency of independent work. The degree of consistency was identified as slight, fair, moderate, substantial, and almost perfect when the κ coefficients were calculated as 0.00-0.20, 0.21-0.40, 0.41-0.60, 0.61-0.80, and 0.81-1.00, respectively [30].

Data Extraction

Two authors (Chuchu Z and JY) independently extracted data from each of the included studies using a structured data extraction form, including (1) study: author, year, country, and design; (2) participants: patients’ diagnosis, sample size, female, age, relationship with patients, occupation, marital status, education level, and financial burden; (3) control; (4) experimental: intervention measurement, intervention delivery (categorized as mobile app, internet-based, and others [eg, telephone calls, VR, and emails]), intervention duration, and data collection time points; (5) outcomes and measurements; and (6) results. If necessary, the corresponding authors were contacted by email and asked for the missing information. A third reviewer (XG) verified the extracted data.

Study Risk of Bias Assessment

The Cochrane’s risk-of-bias tool for randomized trials, version 2 (RoB 2), was used by 2 independent reviewers (JZ and ZH) to assess [31,32]. The RoB 2 tool consists of 5 domains: randomization process, deviations from intended intervention, missing outcome data, measurement of the outcome, and selection of the reported result. The risk of bias was assessed as low, some concerns, or high in the RoB 2 based on responses to signal questions in each domain. The κ coefficient was calculated for each item to identify the degree of consistency.

Quality of Evidence

The Grading of Recommendations, Assessment, Development, and Evaluation (GRADE) approach was used to assess the certainty of the body of evidence [33]. According to the approach, the evidence quality of RCTs was initially graded as high and downgraded to moderate, low, or very low when any limitations were identified in risk of bias, inconsistency, directness, imprecision, or publication bias. However, the evidence could be upgraded by a significant effect and dose-response gradient. Furthermore, 2 authors (Chaoyi Z and YS) independently assessed the quality of evidence using the GRADE handbook, and the interrater agreement was evaluated using the κ coefficient. Disagreements were resolved through discussion, or help was sought from a third author (YZ).

Statistical Analysis

For continuous outcomes, the Hartung-Knapp-Sidik-Jonkman random-effects method [34] was used to obtain standardized mean differences (SMDs) with 95% CIs when studies used different psychometric scales. In studies with follow-up lasting more than 1 month, data from the shortest follow-up period were used. A meta-analysis was undertaken if more than 2 studies reported outcomes in the same format; otherwise, the studies’ findings were described narratively [35]. We applied random-effects models because of suspected heterogeneity across studies to provide a relatively conservative estimate of the combined results [36]. Moreover, the prediction intervals (PI) were reported to indicate heterogeneity in the same metric as SMD and to reflect the variation in actual treatment effects expected in future studies, which aids clinical decision-making [37]. A leave-one-out sensitivity analysis was performed to assess the robustness of the synthesized results. Subgroup analysis was conducted to explore the potential sources of heterogeneity. These analyses were based on patient primary diagnosis (medical vs surgical), relationship with patients (spouse vs children), and type of DHT (education-based vs communication-based vs mental health–related). Finally, if sufficient studies were available for the outcome (n>10), a funnel plot was generated to assess potential small-study effects [38]. The Egger test was performed to further quantify funnel plot asymmetry, with a P value <.05 considered statistically significant [39]. A 2-tailed P<.05 was considered statistically significant. All analyses and visualizations were conducted using the meta package in R software (version 4.5.1; R Core Team).

Ethical Considerations

This study and paper used published data, so no ethics approval was needed.

Results

Study Selection

The PRISMA flow diagram in Figure 1 summarizes the decision pathway for final study inclusion. In total, 2 authors (Chuchu Z and WJ) independently conducted the study selection process, and the κ coefficients for title or abstract screening and full-text reading were 0.82 (P<.001) and 0.84 (P<.001), respectively, indicating almost perfect consistency in these steps. We identified 2624 records through database searches and 2 records through citation searching. After title and abstract screening, we excluded 1944 records and identified 55 records for full-text evaluation. After the full-text screening, we excluded 38 records that did not meet the inclusion criteria and included 17 studies [14,16,26,40-53].

‎

Study Characteristics

Overview

The characteristics of all the included studies are summarized in Table 1. Among the 17 RCTs, 2 were cluster RCTs [41,43], and 4 were pilot RCTs [26,45,47,53]. Studies were conducted in the United States (n=8) [26,41,42,45,47,51-53], China (n=3) [16,40,50], the Netherlands (n=1) [43], Norway (n=1) [48], Germany (n=1) [44], Austria (n=1) [14], Iran (n=1) [46], and Korea (n=1) [49], which were published from 2017 to 2025.

Participants

The sample sizes ranged from 25 to 275, yielding a total of 1864 participants (intervention: 942, 50.5% and control: 922, 49.5%). Regarding patient types, only 1 study [40] did not report; 11 studies [14,26,41-48,52] focused on medical patients; and 5 studies [16,49-51,53] focused on surgical patients. Furthermore, 7 studies [41,42,44,49,50,52,53] recruited not only family members but also ICU patients or ICU health care staff, whereas the remaining 10 studies recruited only family members. In addition, 14 studies reported the mean ages of family members, ranging from 38.68 (SD 11.18) years to 57.40 (SD 12.90) years. Except for the study by Yuan et al [50], all studies reported that 46.3% to 85.7% of participants were female. Concerning the relationship to patients, 7 studies [16,40,43,45,46,48,50] reported that the main participants were children, and 10 studies reported that they were spouses. Moreover, 9 studies [14,16,26,41,42,46,51-53] reported participants’ occupation, indicating most of them were employed; 10 studies [14,26,43,45,47-51,53] did not report marital status; the other 7 did, and the majority of participants were married. Regarding educational level, 5 studies [45,49-52] did not report it; 8 reported that most participants had nontertiary education; and 4 reported that most participants had university education or higher. Financial burden was not reported in 13 studies; 2 studies [41,42] reported no or low burden, and 2 studies [16,40] reported moderate or high burden (Table 1).

Intervention

In the included studies, various types of DHT were used. In total, 9 studies [16,26,40,41,45,49-53] used mobile apps, 5 studies [14,42,44,46,47] used internet-based technology, and 3 used other technologies (1 was a video tool [49], 1 was a digital assessment tool [48], and 1 was VR technology [43]). Furthermore, 10 studies [41-46,49,51-53] reported the specific duration of the intervention, and the remaining studies indicated that family members could use digital intervention tools during the ICU or within months after discharge. Regarding data collection timing, in addition to pre- and postintervention assessments, 6 studies [14,16,26,43,45,51] collected data at 1 month post intervention; 1 study [45] at 2 months post enrollment; 10 studies [14,26,41-43,47,48,51-53] at 3 months post enrollment or discharge; 4 studies [26,42,43,51] at 6 months post discharge; and 2 studies [14,51] at 12 months post admission. No adverse outcomes or events were reported across studies (Table 2).

Control

Control groups varied in different studies; 12 studies [16,40-43,45-50,52] adapted usual or routine care, including regular education on the ICU environment, basic intensive care processes, the use of monitoring devices, ICU rules, spontaneous conversations with the ICU nurses and meetings with the physicians, effective communication with critically ill patients, and basic principles of patient care. In addition, 4 studies [14,26,51,53] used a standard app or website that provided only basic information, and 1 study [44] was a blank control (Table 2).

Outcomes and Assessment Tools

Table 2 summarizes the outcomes and measurement tools of each study. All the assessment scales used in the outcome measurement were validated. In total, 5 scales were used to assess the anxiety and depression levels of ICU patients’ families. Of the total, 9 of the included studies [14,16,26,42,43,45,48,49,53] adopted the Hospital Anxiety and Depression Scale to assess anxiety and depression. Moreover, 2 studies [41,52] assessed anxiety using the Generalized Anxiety Disorder 7 scale and depression using the Patient Health Questionnaire-9 Scale. Another study [40] assessed anxiety and depression using the Depression Anxiety Stress Scale-Chinese. One study [51] used the Patient-Reported Outcomes Measurement Information System questionnaire. One study [50] used the Self-Rating Anxiety Scale to assess anxiety. One study [44] used the Brief Symptom Inventory-18. The measurement tools used for PTSD varied among studies. In total, 3 studies [41,42,52] measured the Post-Traumatic Stress Symptom Inventory, 3 studies [44,45,53] used the PTSD Checklist for DSM-5 (Diagnostic and Statistical Manual of Mental Disorders [Fifth Edition]), and 5 studies [14,16,26,43,48] used the Impact of Event Scale-Revised. In addition, 5 studies measured QoL with 3 scales. 1 study [44] measured the Health Questionnaire of the EQ-5D-5L; 2 studies [45,48] measured the 12-Item Short-Form Health Survey, and 2 studies [16,43] measured the 36-Item Short-Form Health Survey.

Risk of Bias

Two authors (JZ and ZH) used the Cochrane Risk of Bias Assessment Tool version 2 independently to assess the methodological quality of the included studies, and the weighted κ coefficients of domains ranged between 0.55 (P=.002) and 1.00 (P<.001), indicating substantial to almost perfect interrater agreement of the independent work in this part. The results are presented in Figure 2 [14,16,26,40-53]. Specifically, all included studies were reported as randomized, but selective bias was evident because 3 studies [45,48,53] did not report allocation concealment. Due to the nature of DHIs, blinding of participants and investigators was not possible; only 1 study [14] used a double-blind design for the intervention because the structure was the same across both websites. However, almost all studies reported using intention-to-treat analysis to estimate the effect of intervention assignment, thereby reducing bias. A total of 5 studies [40,42,45,50,53] reported deviations from the intended intervention arising from the trial context, and are therefore rated as high-bias. In addition, 3 studies [26,43,52] reported incomplete data, which increased the risk of attrition bias. Furthermore, 12 studies were blinded to outcome assessors, and the other 5 studies [26,44,48,49,53] addressed bias in outcome measurement. Moreover, 2 studies [14,45] provided insufficient information to permit judgment and might have led to overestimated effects and introduced selection bias into the reported results. Finally, 4 studies were assessed as “low-risk bias,” 10 as “some concerns,” and 3 as “high-risk bias.” However, in accordance with the consensus agreement of all the authors, the studies with a high risk of overall bias were ultimately reserved.

‎

Quality of Evidence

The GRADE approach [33] was used to assess the overall certainty of the evidence. The κ coefficient in this part was 0.88 (P<.001), indicating almost perfect interrater consistency. Among the 7 outcomes analyzed, the certainty of evidence was downgraded to moderate for QoL and quality of communication (QOC). The certainty for anxiety, depression, PTSD, and self-efficacy was downgraded to low. Stress was downgraded to very low certainty of evidence (Table 3).

Effect of DHIs

Psychological Outcomes

In total, 12 studies assessed the impact of DHIs on anxiety among ICU family members. The pooled results showed that DHIs did not lead to a statistically significant reduction in anxiety levels (SMD –0.34, 95% CI –0.68 to 0.00; P=.05). The heterogeneity among studies was substantial (I²=80.19%; τ=0.49; τ²=0.24), and the certainty of the evidence was rated as low. The 95% PI of the effect of DHIs on anxiety was –1.46 to 0.79, which indicates that we can be 95% confident that the true effect size from a future individual study would fall within this range (Figure 3 [16,26,40-43,45,48-50,52,53]). Similarly, 12 studies investigated the impact of DHIs on depression among ICU family members. The results revealed no significant effect (SMD –0.26, 95% CI –0.52 to 0.01; P=.06), with moderate heterogeneity (I²=67.80%; τ=0.37; τ²=0.13). The certainty of this evidence was also rated as low. A wide 95% PI (–1.11 to 0.60) suggests that future studies could show effects ranging from a large reduction to a modest increase in depression levels (Figure 4 [14,16,26,40-43,45,48,49,52,53]). In total, 11 studies evaluated the effects of DHIs on PTSD among ICU family members. The meta-analysis demonstrated no significant difference in PTSD scores between groups (SMD –0.21, 95% CI –0.49 to 0.06; P=.11), with substantial heterogeneity (I²=69.98%; τ=0.34; τ²=0.12). The certainty of this evidence was also rated as low. The 95% PI (–1.03 to 0.60) indicates that the actual effect of DHIs on PTSD could vary widely, indicating potential for both beneficial and harmful effects (Figure 5 [14,16,26,41-45,48,52,53]). Furthermore, regarding stress and self-efficacy, which were evaluated in only 2 studies each, we discussed them qualitatively. 2 studies reported different results regarding stress: 1 study [40] found no effect of DHI on stress. In contrast, another [46] reported a significant effect of DHI in reducing stress among ICU family members. Neither study [45,48] on self-efficacy found a significant benefit from DHIs. The certainty of the 2 pieces of evidence was rated as very low and low.

‎

‎

‎

QoL

In total, 5 studies assessed the impact of DHIs on QoL among ICU family members. A random-effect meta-analysis indicated no statistically significant improvement in QoL (SMD 0.09, 95% CI –0.10 to 0.28; P=.36). Despite the absence of heterogeneity (I²=0.00%; τ=0.14; τ²=0.02), the actual effect of DHIs on QoL in the future falls within the broader PI range (–0.41 to 0.60), suggesting that future outcome could vary significantly across different populations and settings (Figure 6 [16,43-45,48]). The certainty of this evidence was rated as moderate.

‎

QOC

In total, 4 studies evaluated the effect of DHIs on the QOC among ICU family members. The pooled analysis showed no statistically significant difference between groups (SMD 0.14; 95% CI –0.03 to 0.31; P=.10; Figure 7 [41,42,47,52]). Despite the absence of heterogeneity (I²=0.00%; τ=0.08; τ²=0.01), the actual effect of DHIs on QoL in the future falls within the broader PI range (–0.26 to 0.55), suggesting that future outcomes could vary significantly across different populations and settings. The certainty of this evidence was rated as moderate.

‎

Subgroup Analysis

We conducted subgroup analyses for the three outcome indicators of anxiety, depression, and PTSD based on potential covariates, including patients’ primary diagnosis, relationship with patients, and type of DHT (Table S1 and Figures S1-S3 in Multimedia Appendix 4). Regarding anxiety, the results of subgroup analyses showed that there was no significant difference (P>.05) in the patients’ primary diagnosis (χ²₂=2.60; P=.27), relationship with patients (χ²₁=0.23; P=.63), and type of DHT (χ²₂=1.43; P=.49), suggesting that these factors may not be related to the change of intervention effect. Within the subgroup, family members of medical patients showed a slight improvement trend (SMD –0.14, 95% CI –0.26 to –0.01; P=.04). From the perspective of heterogeneity sources, the patients’ primary diagnosis and the type of DHT may be potential sources of heterogeneity. Regarding depression, subgroup analysis showed significant differences between subgroups with different patients’ primary diagnosis (χ²₂=29.17; P<.01). Among the surgical patients’ family members (3 studies), depression scores in the intervention group were significantly reduced (SMD –0.94, 95% CI –1.64 to –0.24, t₂=–5.80; P=.03). In studies focusing on medical patients, no significant effect of the intervention on family members depressive symptoms was observed (SMD –0.04, 95% CI –0.14 to 0.05, t₇=–1.11; P=.30). There was no statistical significance (P>.05) in the subgroup differences in relationship with patients (χ²₁=0.38; P=.54) and type of DHT (χ²₂=3.11; P=.21). Among the subgroups of type of DHT, the effect size of the education-based DHT group (4 studies) had a more minor statistical significance (SMD –0.11, 95% CI –0.21 to –0.01, t₃=–3.47; P=.04). Patients’ primary diagnosis and type of DHT may be significant sources of heterogeneity. Regarding PTSD, the differences between subgroups in the patients’ primary diagnosis (χ²₁=3.29; P=.07), relationship with patients (χ²₁=1.30; P=.25), and type of DHT (χ²₂=2.75; P=.25) did not reach statistical significance (P>.05). However, looking at the sources of heterogeneity, these factors may be the primary drivers. Due to the limited number of studies in some subgroups, these results should be interpreted with caution.

Sensitivity Analysis

For anxiety, it had no impact on the statistical significance of the results after removing the study conducted by Xiong et al [16] (SMD –0.18, 95% CI –0.38 to 0.01; P=.06; Figure S4 in Multimedia Appendix 4), which decreased heterogeneity to 27.36%. Regarding depression, removing the study by Xiong et al [16] (SMD –0.17, 95% CI –0.38 to 0.05; P=.11; Figure S5 in Multimedia Appendix 4) decreased heterogeneity to 36.09%, but it did not alter the statistical significance of the results. For PTSD, sensitivity analyses demonstrated that removing the study by Xiong et al [16] (SMD –0.08, 95% CI –0.21 to 0.05; Figure S6 in Multimedia Appendix 4) resolved heterogeneity, although it did not alter the statistical significance of the results. For QoL, statistical significance was observed only after removing the study by Drop et al [43] (SMD 0.19, 95% CI 0.01 to –0.37; P=.04; Figure S7 in Multimedia Appendix 4). Regarding QOC, the effect of DHIs remained stable across studies, indicating robust findings.

Bias Assessment

Visual inspection of the funnel plots for anxiety, depression, and PTSD showed approximate symmetry (Figures S1-S3 in Multimedia Appendix 5). Egger tests did not provide statistically significant evidence of small-study effects for anxiety (P=.24), depression (P=.20), and PTSD (P=.44; Table S1 in Multimedia Appendix 5). However, we note that small-study effects can arise from various sources beyond publication bias, such as heterogeneity or methodological differences. For other outcomes, there were insufficient numbers of systematic reviews (n<10) to create funnel plots.

Discussion

Principal Findings

In total, 17 studies were included, revealing that compared with control groups, DHIs may not confer a significant benefit in reducing anxiety, depression, PTSD, or stress among ICU family members. Moreover, they also did not notably improve self-efficacy, QoL, or communication quality. Given the low certainty of the evidence for the proposed association, the substantial study heterogeneity, and the overall risk of bias in the included studies, the findings should be interpreted with caution. Subgroup analyses indicated that patients’ primary diagnosis may influence the intervention’s effect, with the benefits being more pronounced for families of surgical patients. Patients’ primary diagnosis, relationship with patients, and type of DHT may be a significant source of heterogeneity. Given the PI value, adverse effects of the intervention cannot be ruled out in future studies, and more RCTs are needed to provide robust results.

The observed effect of DHIs on anxiety aligns with previous research concerning improvements in anxiety among caregivers of patients with cancer [23]. The ICU environment is characterized by high stress, restricted access, asymmetric information, and uncertain prognosis, which can significantly increase the psychological burden of family members [54,55]. DHI compensates for the absence of in-person support by using mobile apps, network platforms, and virtual visits [56]. It provides timely updates on diseases, health education, and emotional support for family members, thereby reducing uncertainty and equipping them with knowledge and skills that, in turn, alleviate anxiety symptoms [57]. However, the clinical significance of this finding is tempered by its statistical marginality (P=.05), a low certainty of the evidence, and risk of bias in some studies, which significantly reduces the credibility of this potential effect. It is critical to underscore that substantial heterogeneity (I²=80.19%) yields a wide 95% PI (–1.46 to 0.79), suggesting that the distribution of effects on anxiety is unpredictable. This finding should only be considered a preliminary exploratory conclusion and requires further validation through additional research. No significant effects were observed on depression, PTSD, stress, self-efficacy, QoL, and QOC. It is important to note that the certainty of the evidence for these outcomes is low, with an associated risk of bias, so the reasons for the lack of significant observed effects should be considered from multiple perspectives. From a clinical view, the onset of PTSD is often delayed and complex, typically manifesting months to a year post discharge, making it challenging for short-term interventions and follow-ups to capture long-term effects [14,44]. Additionally, stress, self-efficacy, QOC, and QoL are inherently multidimensional indicators influenced by confounding factors, including economic conditions and social support, beyond psychological state [45,58,59]. From a methodological perspective, existing studies have limitations, including risk of bias, low certainty of the evidence, and imprecision—particularly given that only 2 studies were included for stress and self-efficacy. Current evidence is insufficient to identify the potential benefits or harms of DHIs for these outcomes, and subsequent research is very likely to alter these estimates.

The observed significant heterogeneity across outcomes prompted extensive subgroup analyses. While these analyses pointed to several potential moderators—such as patients’ primary diagnosis and type of DHT for depression—their interpretability is limited. For example, the notable reduction in depression among surgical patients (SMD –0.94, 95% CI –1.64 to –0.24), derived from only 3 studies, may reflect a type Ⅰ error or chance variation rather than an actual differential effect. Similarly, although traditional factors such as family members’ sex and relationship with patients did not reach statistical significance in our analyses, existing literature consistently indicates that female caregivers, especially spouses, often endure greater caregiving stress and more persistent psychological distress [14,60,61]. The overall heterogeneity remained largely unexplained by subgrouping, further underscoring the robustness of these comparisons. Thus, these analyses should not be viewed as confirming specific moderators but rather as highlighting priority variables for future, adequately powered studies.

The limited effectiveness of DHIs may also be attributed to the design of the interventions. Most existing interventions are information-centric, providing unidirectional education rather than incorporating structured psychotherapy, skills training, or facilitating two-way communication [42,62]. Such a format appears insufficient for translating into actual skill enhancement, family interaction, or overall well-being improvement [43]. Furthermore, some studies reported positive subjective feedback from families despite nonsignificant quantitative outcomes, indicating that current measurement tools may lack sensitivity to nuanced experiential gains [40,43,44].

Moving forward, future research should use mixed methods designs to capture subjective experience and identify needs of high-risk subgroups; tailor interventions according to patients’ diagnosis, cultural context, and characteristics of caregiver; develop more interactive and personalized DHIs formats, such as artificial intelligence–based scenario simulations, two-way communication platforms, and longitudinal support programs; extend follow-up duration and use multimodal outcome measures that combine patients and family members reported experiences with clinically relevant indicators; and explore hybrid online-offline interventions to support continuity of care across different phases of the ICU trajectory.

Strengths and Limitations

This systematic review has several strengths. First, to our knowledge, this is the first meta-analysis to synthesize and appraise the evidence about the effects of DHIs on psychological and related outcomes among family members in the ICU. Second, we strictly followed PRISMA guidelines to strengthen transparency and clarity of reporting. Third, we searched across 9 databases from inception to October 11, 2025, reflecting the most recent advancements in the field of DHI. Moreover, we used the Cochrane RoB 2 and GRADE methodology for quality assessment. Finally, subgroup analyses, sensitivity analyses, and bias assessment were conducted to validate the stability of the results and explore potential sources of heterogeneity.

However, there were several limitations to this review. First, the considerable heterogeneity among studies, as evidenced by the wide PIs for our primary outcomes, indicates uncertainty about the intervention effect. Discrepancies in participants’ characteristics and intervention protocols could lead to inconsistencies and bias in the study findings. Second, the interventions included in the review ranged widely, from simple information websites and SMS text messages to VR- and app-based psychotherapy, and intervention duration varied. Pooling across these fundamentally different modalities may dilute or obscure which components work. Third, while blinding interventionists poses difficulties in DHT care settings, the absence of blinding for study participants and interventionists in most of the included studies introduced risks of bias, resulting in a methodological error. Fourth, the search approach has certain limitations as it exclusively incorporated papers published in English or Chinese, rendering findings from articles in other languages unknown and constraining the global generalizability of the results. Fifth, while no significant small-study effects were detected, the ability of these tests may be constrained by the limited number of included studies. Hence, we cannot conclusively dismiss potential publication bias. Sixth, the majority of the selected trials were conducted in developed countries. This bias may result from the restricted access to technology and the low practicality of technology-based interventions in low-income countries. Seventh, the results regarding the effect of DHIs on psychological outcomes, QoL, and communication quality are inconclusive because of the limited number of studies and the wide PIs. Negligible effects are likely from poor statistical power, not a genuine absence of an intervention effect.

Implications of Nursing Practice

Given the current limitations and low certainty of the evidence, our findings have limited direct implications for clinical practice but provide valuable insights for future research. ICU nurses, as the medical personnel with the most frequent interactions with patients and their families, are ideally positioned to lead DHIs [23]. They can facilitate online meetings involving physicians, patients, and families to dismantle information barriers, guided by shared decision-making theory. More importantly, nurses can leverage digital platforms to enable personalized communication. This nurse-driven, tailored digital communication strategy not only provides adequate informational support to mitigate their negative emotions but also significantly improves communication quality and overall QoL [63]. The subgroup analyses suggest that DHIs’ effectiveness may not be universal but can be enhanced by tailoring interventions to specific family or patient characteristics. In the future, using technologies such as videoconferencing, virtual simulations, and advanced language models, we can deliver more comprehensive and personalized psychological support interventions led by nurses via the internet and mobile devices. Simultaneously, it is necessary to advocate large-sample, multicenter, high-quality, and reproducible RCTs to provide more robust data to support DHI’s findings on the psychological effects and QoL of ICU family members in China.

Conclusions

In conclusion, this study provides the first comprehensive synthesis of RCT evidence on DHIs for ICU family members, revealing significant uncertainty in the estimated effects. Despite its methodological rigor and use of PIs, the conclusions of this review are constrained by the intrinsically heterogeneous and low certainty of the existing evidence. Unlike previous reviews, this work integrates more recent studies and focuses explicitly on the family members of ICU patients. Future research should prioritize the development of theory-informed, personalized, and interactive DHIs. Incorporating mixed methods designs and targeting high-risk subgroups would be particularly valuable. Although direct clinical recommendations are limited, DHIs hold potential for enabling more proactive and individualized communication strategies for families of ICU patients.