This study was registered on 3 March 2025 with the PROSPERO database of systematic reviews (CRD420251003303) and reported following the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) guidelines (14). During the conduct of the review, amendments were made to the original protocol and formally updated in the PROSPERO registration record. The amendments were: (1) Search Strategy: Due to institutional database access limitations, the search was confined to PubMed, Web of Science, Cochrane Library, and Embase. (2) Outcome Scope: To ensure a focused and feasible synthesis, the protocol was amended to explicitly define psychological outcomes as the primary focus of this review, while the original version included both psychological and physiological outcomes. This manuscript reports the findings based on the final, amended protocol.

We searched PubMed, Web of Science, Cochrane Library and Embase from database inception to 4 March 2025 and updated our search on December 2nd, 2025. Our search strategy integrated the terms “psychological intervention,” “critical,” and “clinical trial,” employing a combination of Medical Subject Headings (MeSH) and keywords to conduct the search. We conducted a concurrent search for references pertaining to pertinent systematic reviews and clinical guidelines. Comprehensive retrieval strategies for all databases are detailed in Supplementary 1.

We have selected the full-text studies based on the following criteria: (1) types of studies: we included RCT. We did not limit our studies based on the duration of follow-up. We excluded short reports, research letters, conference abstracts, and studies that had not been published in their entirety in peer-reviewed scientific journals. (2) Types of participants: we included who were adult patients (aged > 18 years) admitted to the ICU for at least 24 h, or their family members. In our study, a family member is defined as the individual within the familial unit who is most actively engaged in the treatment and care decision-making processes for the patient, irrespective of the presence of a consanguineous relationship with the patient (15). (3) Types of interventions: we included studies comparing psychological interventions (either alone or in combination with other treatments) with any comparator intervention. Psychological interventions were defined as all types of counseling, psychoeducation, social support, or therapy that are based on psychological principles and aimed at improving general well-being (16). We also included interventions that were explicitly defined as “psychological interventions” by the study authors (17). (4) outcome measures: the primary outcome measures comprised anxiety, depression, and sleep quality, while the secondary outcome measures encompassed PTSD and quality of life. For all outcome indicators of interest, we extracted the relevant result data from all available follow-up time points. We imposed no restrictions on the publication language. Articles from all databases were imported into EndNote X9 for organization. Following the removal of duplicates, an initial screening was conducted by reviewing the titles and abstracts. Subsequently, a more comprehensive screening was performed through read full-text. The process was conducted independently by two researchers. In instances of disagreement, consensus was achieved through consultation with a third researcher. Furthermore, we excluded any articles for which full texts were not accessible.

Two researchers independently employed standardized forms to extract data separately. Any disagreements were resolved through consultation with a third researcher. The extracted data encompassed: (1) fundamental information about the literature, including the first author, country, and publication year; (2) research design details, such as the type of research design, inclusion and exclusion criteria for participants, the number of participants, and follow-up duration; (3) participant demographics, including age and gender; (4) specifics of the psychological intervention, such as the key components of both the psychological and control interventions, the provider of the intervention, as well as the dosage, frequency, and duration of the intervention; and (5) outcome data such as anxiety, depression, sleep quality, PTSD and quality of life.

For all outcomes of interest, data were extracted from all available follow-up time points. Based on the approach by Ho et al. (17), the data were categorized according to the following time intervals: pre-intervention (i.e., baseline); immediately post-intervention (i.e., at the end of treatment or within < 2 months after the intervention); short-term sustainability (from ≥ 2 months to < 6 months post-intervention); medium-term sustainability (from ≥ 6 months to < 12 months post-intervention); and long-term sustainability (≥ 12 months post-intervention).

Two reviewers independently assessed the methodological quality of individual studies using the Cochrane Risk of Bias Tool 2.0 (18). This tool comprises five domains of bias: bias arising from the randomization process, bias due to deviations from intended interventions, bias due to missing outcome data, bias in measurement of the outcome, and bias in selection of the reported result. Each domain was classified as being at “low,” “some concerns,” or “high” risk of bias. Two authors independently assessed the risk of bias, and any discrepancies in the quality assessment were resolved through consultation with a third author.

All statistical analyses were performed using RevMan (Review Manager) Version 5.4.1. A two-tailed p-value < 0.05 was considered statistically significant for overall effects. For continuous outcomes, the treatment effect was expressed as the Standardized Mean Difference (SMD) with 95% CI. The magnitude of the effect size, expressed as the SMD, was interpreted according to Cohen’s proposed criteria, whereby an absolute SMD value of approximately 0.2 is considered a small effect, approximately 0.5 a medium effect, and approximately 0.8 a large effect (19). Statistical heterogeneity among the included studies was assessed using the Cochran’s Q test (χ² test) and quantified using the I² statistic. Given the anticipated clinical and methodological diversity, a random-effects model was used as the primary analytical approach. A fixed-effect model was additionally applied for comparison only in cases of negligible heterogeneity (I² < 30%). To explore potential sources of substantial heterogeneity (I² ≥ 50%), pre-specified subgroup analyses were conducted based on: type of psychological intervention, ICU clinical setting, assessment tool, provider expertise, mode of delivery and follow-up duration. Between-subgroup differences were tested using the standard χ² test. Sensitivity analyses were conducted to assess the robustness of the pooled results by: Sequentially removing each individual study to examine its impact on the overall effect size. All studies with a “high-risk” quality assessment were excluded for conducting the sensitivity analysis. Comparing results between studies with a low overall risk of bias and those with a high risk of bias. If a sufficient number of studies (n ≥ 10) were included in a meta-analysis, potential publication bias would be assessed visually using a funnel plot and statistically using Egger’s regression test.

A total of 4, 226 records were initially obtained. Following the removal of 1, 211 duplicate records and the exclusion of 2, 791 records deemed irrelevant based on a review of their titles and abstracts, 224 studies were identified as meeting the criteria for full-text screening. Of these, 199 studies failed to satisfy the inclusion eligibility standards. Consequently, 25 RCTs were ultimately included in the analysis (6–10, 20–39). Figure 1 shows the PRISMA flow diagram of study selection.

A total of 25 studies included 3,849 participants. Among them, 22 studies included 3,070 ICU patients (6–10, 20–31, 34, 36–39), and 5 studies included 779 family members of ICU patients (6, 32, 33, 35, 37). Table 1 provides a summary of the characteristics of the studies included and Table 2 provides a detailed comparison of the characteristics of the psychological interventions.

Among the 22 studies on patients, 4 studies were from China (26, 29–31), 3 studies were from the United States of America(USA) (7, 20, 21), 3 studies were from Iran (8, 9, 28), 2 studies were from the United Kingdom (23, 34), 2 studies were from Korea (24, 39), 2 studies were from Turkey (27, 38), 1 study were from Belgium (25), 1 study was from Canada (10), 1 study was from the Republic of Cyprus (22), 1 study was from French (6), 1 study was from Denmark (37) and 1 study was from 6 European countries (36). The participants of 7 studies came from Cardiac Care Unit (CCU) (8, 9, 24, 25, 27, 31, 38), 7 studies came from multiple different ICUs (6, 7, 20, 21, 23, 26, 36), 1 study came from Burn Intensive Care Unit (BICU) (28), 1 study came from Surgical Intensive Care Unit(SICU) (29, 39), 1 study came from MSICU (37), and 6 studies did not specify from which ICU they came (10, 22, 30, 34). Psychological interventions vary significantly in their therapeutic components, including behavioral therapy (9, 20, 28), CBT (8, 31), ICU diaries (6, 10, 34, 36, 37), mindfulness and meditation (7, 21, 24–26, 39), psychoeducation (10), and psychological interventions delivered with non-psychological co-interventions (22, 23, 27, 29, 30, 38). The modes of delivery included face-to-face sessions (6, 8–10, 20, 22, 23, 27–30, 34, 36–38), virtual reality (VR) (24–26, 31, 39), and remote delivery (via telephone or apps) (7, 21). The duration and frequency of the psychological interventions also differed considerably.

Among the five studies conducted on the family members of ICU patients, 2 studies were from USA (32, 33), 1 study was from French (6), 1 study was from Denmark (37) and 1 study was from 2 European countries (35). The participants of 2 studies came from multiple different ICUs (6, 35), 1 study came from CCU (32), 1 study came from MSICU (37), and 1 study did not specify from which ICU they came (33). The psychological intervention related to family members can be classified into three categories based on the intervention mechanism: including behavioral therapy (32), CBT (33) and ICU diaries (6, 35, 37).

Eighteen studies were classified as low-risk, while seven were classified as high-risk; the assessment results of the two authors were highly consistent. (1) Bias in the randomization process: 3 studies were rated as high risk (20, 25, 30). Among them, one study (20) did not clearly specify whether allocation concealment was implemented before recruitment, one study (25) provided insufficient details regarding the randomization process, and another study (30) lacked a detailed description of the randomization method and allocation concealment mechanism. (2) Bias in deviations from intended interventions: 1 study (37) was classified as a high-risk study due to the presence of cross-contamination between groups. (3) Bias in missing outcome data: 1 study (33) was rated as high-risk because it did not specify whether the missing data had been appropriately handled. (4) Bias in measurement of the outcome: the risk of bias was judged as low for all included studies. Although the primary outcomes were assessed using patient-reported measures, which are inherently subjective, all studies employed well-validated instruments (e.g., hospital anxiety and depression scale (HADS) for anxiety/depression, impact of events scale-revised (IES-R) for PTSD). The use of these standardized tools, with established reliability and validity, minimized the potential for measurement bias. (5) Bias in the selection of the reported result: all reported results were predefined in the methods section, and most of them have provided published research protocols; therefore, all 25 studies were identified as low risk. (6) other bias: 2 studies (32, 35) were rated as high risk due to an excessively small sample size.

In addition to quantitative findings, qualitative insights from several studies are noteworthy. Rousseaux et al. (25) found that, compared with the virtual reality hypnosis (VRH) group, the anxiety levels of patients in the simple hypnosis group were significantly higher (p = 0.007). However, no significant differences in anxiety levels were observed between the other intervention groups (virtual reality group, VRH group) and the control group. Data from Wang et al. (26) showed that acute respiratory distress syndrome (ARDS) survivors receiving the Virtual Reality-Integrated Psychological Intervention (VR-IPI) exhibited consistently superior improvement trends in anxiety, depression, and PTSD symptoms at all follow-ups (3, 6, 9, 12 months) compared to those receiving Traditional Psychological Counseling (TPC). Nielsen et al. ‘s (37) study showed that diaries authored by relatives significantly reduced the relatives’ own symptoms of PTSD, but had no significant effect on the patients’ PTSD symptoms, or on the anxiety, depression, or quality of life of either patients or relatives. Ozdemir and Yilmaz’s (38) study demonstrated that structured psychosocial nursing interventions, combining cognitive behavioral therapy with sensory modulation (e.g., eye masks, earplugs), significantly reduce anxiety levels and improve postoperative sleep quality in patients undergoing open heart surgery.

Due to the number of included studies for each outcome being less than ten, the statistical power of Egger’s regression test was insufficient. Therefore, formal statistical testing was not performed, and the assessment relied primarily on visual inspection of funnel plot symmetry. The funnel plot showing the immediate effects of psychological intervention on patients’ anxiety, depression, sleep quality, PTSD, and quality of life is presented in Supplementary 45–49. (1) Depression: the funnel plot exhibited asymmetry, with a gap in the bottom-right quadrant. This suggests the presence of publication bias, indicating that small-scale studies with negative results may have been omitted, potentially leading to an overestimation of the intervention effect. (2) Sleep quality: significant asymmetry and a bottom gap were observed, indicating substantial publication bias and/or heterogeneity, which may undermine result stability. (3) PTSD: this funnel plot shows asymmetry, with only 2 studies on the left side. This indicates publication bias, meaning that studies with positive results may have been omitted, which could lead to an underestimation of the estimated intervention effect. (4) The funnel plots for anxiety and quality of life showed relatively good symmetry, indicating a lower risk of publication bias for these outcomes.

Sensitivity analysis was performed to determine the impact of each study on the pooled SMDs. The overall effect sizes did not change significantly when one study was excluded at a time, which indicated the strong stability of the results.

Our study evaluated the efficacy of psychological interventions on psychological distress, sleep and quality of life among ICU patients and their family members. A total of 25 RCTs involving 3, 849 participants were included. The main findings are summarized as follows: (1) For ICU patients, psychological interventions show potential in alleviating anxiety symptoms, an effect sustained from the immediate post-intervention period through short-term follow-up. For depression, a significant reduction was observed immediately post-intervention, but this improvement was not sustained at short-term follow-up. However, no statistically significant improvements were observed for sleep quality, PTSD, or quality of life at either the immediate or short-term follow-up assessments. (2) For family members of ICU patients, psychological interventions did not show statistically significant effects on anxiety, depression, PTSD, or quality of life, either immediately post-intervention or at short-term follow-up.

This study indicates that psychological interventions in the ICU have the potential to alleviate patients’ immediate anxiety and reduce anxiety levels in the short term. This is consistent with several previous studies, supporting the positive role of psychological intervention in the ICU (40, 41). However, the meta-analyses showed very high statistical heterogeneity, greatly limiting the reliability of applying the “overall effect” to clinical practice. We performed several subgroup analyses, revealing that follow-up duration, ICU environment, mode of delivery, type of psychological intervention, assessment tool, and provider’s expertise could significantly influence the effect size. For instance, CBT had the strongest impact, while mindfulness/meditation was not statistically significant. Face-to-face and VR interventions were effective, unlike remote interventions like telemedicine. Psychological interventions conducted in a homogeneous critical care setting (e.g., exclusively in CCUs or BICUs) were associated with a significantly greater reduction in patient anxiety compared to those implemented in mixed ICU environments. This finding may be explained by the more defined patient profiles and reduced clinical variability in homogeneous units, facilitating tailored and effective intervention delivery. It is important to underscore that these subgroup analyses are primarily exploratory and are largely derived from a limited number of studies.

Notwithstanding the heterogeneity and limitation inherent in the available evidence, the findings of this review nevertheless offer important implications for planning psychological services within the ICU. Firstly, regarding intervention selection and feasibility, the analysis suggests that even brief, manualized interventions led by nurses (such as ICU diaries or psychoeducation) may yield immediate benefits (10). Such interventions, with their low reliance on specialized psychological resources and high practicality, are well-suited for integration into the foundational psychological care within the ICU. Secondly, the study indicates that the intervention’s impact might diminish over time. This phenomenon may stem from a shift in the sources of patient anxiety after ICU discharge—moving from acute physiological stress and environmental fear during the critical illness phase to chronic worries about functional recovery, sequelae, and the future during convalescence (42). This underscores the urgent need for developing stepped or continuous psychological support protocols to ensure long-term maintenance of treatment effects (43).

This study indicates that psychological interventions may provide a slight to moderate reduction in the immediate depressive symptoms experienced by ICU patients. However, the finding is tempered by substantial heterogeneity (I² = 86%) and a lack of sustained benefit in the short term. Subgroup analysis identified statistically significant variations attributable to factors including follow-up duration, type of intervention, mode of delivery, ICU setting, assessment instruments, and the professional expertise of the providers. Although the results of these subgroup studies are exploratory and are limited by the number and scale of the included trials, they provide crucial guidance for clinical decision-making. For instance, the environment of the ICU as a modulating factor indicates that the psychological intervention may need to be tailored according to the specific stressors and rehabilitation trajectories of different patient groups. The marked reduction in the observed effect during the short-term follow-up, coupled with the duration of follow-up as a critical influencing factor, underscores a significant challenge confronting the clinical field: the inadequacy of relying exclusively on intervention measures provided by the ICU to achieve long-term psychological health recovery in ICU survivors (44). Symptoms presenting shortly after ICU admission may diminish over time, whereas those manifesting during extended follow-up are likely to persist (45). Consequently, clinical practice must transition from offering isolated interventions to developing comprehensive care pathways. Recent systematic reviews and meta-analyses have also emphasized the necessity of the PICS follow-up system (46).

This study indicates that psychological interventions do not have a statistically significant effect on improving sleep quality among ICU patients (p = 0.19). Nonetheless, the notably high heterogeneity observed across the studies (I² = 97%) implies that the effect of the intervention may be influenced by a variety of factors. The sensitivity analysis indicated that the exclusion of two studies, both categorized under CBT (Ghelasi et al. (8); Li et al. (31)), resulted in a substantial reduction in overall heterogeneity, decreasing from 97 to 59%. However, the heterogeneity remained relatively elevated. This preliminary observation suggests that these two particular CBT studies are the primary contributors to the observed inconsistency. The subgroup analysis indicates that the type of intervention may play a crucial role in modulating the observed effects. Specifically, the CBT subgroup, as examined in two studies, demonstrated significant and substantial improvements in sleep. Nonetheless, this subgroup exhibited extremely high heterogeneity (I² = 97%), and the limited number of studies precludes the ability to draw definitive conclusions.

This study indicated that the overall effect of psychological intervention measures on alleviating post-traumatic stress disorder among patients in the intensive care unit did not achieve statistical significance, both in terms of immediate effects (p = 0.54) and follow-up effects (p = 0.56). Upon exclusion of the study conducted by Liang et al. (29) from the sensitivity analysis, the heterogeneity index (I²) was found to be 0%. Although the combined effect size exhibited a slight alteration, it did not achieve statistical significance (p = 0.22). This finding suggests that the overall result, indicating ineffectiveness, is robust and not solely influenced by individual studies. Although subgroup analyses suggest that the assessment tool, follow-up duration, and ICU type may be important moderating factors, these findings are primarily based on a single study (29). They are statistically underpowered and carry the risk of multiple comparisons. Therefore, they can only serve as hypotheses for future research and should not inform clinical decision-making.

The results of our meta-analysis indicated that psychological interventions did not lead to a statistically significant enhancement in patients’ quality of life, either immediately post-intervention (p = 0.14) or during the short-term follow-up period (p = 0.57). The sensitivity analysis indicated that the exclusion of the study conducted by Li et al. (31) resulted in the complete elimination of heterogeneity (I² = 0%); however, the overall effect remained statistically insignificant. This finding suggests that the overall conclusion is robust and not unduly influenced by any single study. This indicates that, while the psychological interventions are deemed safe, their impact on the comprehensive and multifaceted construct of quality of life remains constrained. Quality of life is a multifaceted construct that encompasses elements from the physical, psychological, and social domains (47).

Despite some preliminary evidence suggesting potential benefits of specific interventions in selected outcomes, the aggregated results of this meta-analysis did not show statistically significant effects. The meta-analysis examining symptoms of anxiety, depression, and post-traumatic stress disorder revealed no statistically significant differences between the intervention and control groups at both the immediate post-intervention and short-term follow-up assessments. However, considerable and at times substantial heterogeneity was observed in the aggregated estimates; for example, the I² statistic for the immediate effect of anxiety reached as high as 79%, while that for PTSD was 65%. The sensitivity analysis suggests that the observed heterogeneity in post-traumatic stress disorder outcomes is predominantly attributable to the study conducted by Nielsen et al. (37). The reason might be related to the cross-contamination between the intervention group and the control group in the study. Concerning the quality of life, the limited availability of data precludes the possibility of conducting a quantitative synthesis. Furthermore, the individual studies that are accessible do not indicate any significant differences between the groups (32, 33, 37).

The considerable heterogeneity observed may be attributable to the varying baseline vulnerabilities and pre-existing mental health risk factors among family members, which the included trials may not have sufficiently accounted for. Family members of ICU patients experience psychological stress not just from the sudden ICU admission, but also from the ‘cumulative burden’ or ‘double blow’ effect due to existing stressors piling up. For instance, the family members’ personal history of severe infections (such as COVID-19) can exacerbate the psychological health problems of the ICU patients’ families (48). Moreover, sociodemographic and role-related factors, including gender—wherein women generally report elevated levels of distress—occupational stress, and the substantial caregiving responsibilities assumed during or prior to an ICU stay, can markedly elevate the baseline psychological risk (49). In conclusion, the present findings, which are statistically insignificant, should not be construed as definitive evidence that psychological support is ineffective for all family members of patients in the ICU.

This study possesses several strengths: the reporting strictly adhered to the PRISMA guidelines, a comprehensive search strategy was implemented to minimize omissions, and both patients and their family members—two crucial populations—were examined. However, the study also presents limitations: (1) The number of included studies was limited, particularly concerning outcomes related to family members and certain secondary outcomes, thereby constraining the certainty of the conclusions. (2) Substantial heterogeneity was observed in several pooled analyses. Although we performed sensitivity analyses and extensive subgroup analyses, some potential sources of heterogeneity—such as specific intervention intensity and detailed patient acuity—could not be fully explored due to inconsistent reporting in the original studies. (3) Exploratory results of subgroup analysis: the number of studies included in most subgroup analyses is limited, and the statistical test power may be insufficient. These results are meant to generate hypotheses, not conclusions, and should be interpreted with caution. (4) Additionally, the funnel plot analysis revealed asymmetry in depression, sleep quality, and PTSD outcomes, suggesting the possibility of publication bias.