Follow-up focused on psychological intervention initiated after intensive care unit in adult patients and informal caregivers: a systematic review and meta-analysis

Studies

We included randomized controlled trials that assessed the effects of follow-up after ICU discharge on mental health outcomes among adult patients and informal caregivers. We analyzed papers including published and unpublished articles, abstracts of conferences, and condolence letters. We excluded studies with cluster randomized or quasi-randomized trials, cohort studies, case-control studies, and case series. Furthermore, while including studies for this SR/MA, we did not apply restrictions pertaining to language, country, observation period, or publication year.

In May 2021, we searched the following databases: MEDLINE (PubMed), the Cochrane Central Register of Controlled Trials (Cochrane Library), EMBASE (Dialog), the Cumulative Index to Nursing and Allied Health Literature (CINAHL) (accessed via EBSCO), and APA PsycInfo (Ovid). In May 2021, we searched for ongoing and unpublished trials in trial registers such as ClinicalTrials.gov and the World Health Organization International Clinical Trials Platform Search Portal (WHO ICTRP), respectively. Details of these searches have been listed in the protocol (Yoshihiro et al., 2021). We conducted a ‘snowball’ search to identify studies that used reference lists of publications eligible for full-text review (including international guidelines) (National Institute for Health and Care Excellence, 2009; Nolan et al., 2021) and used Google Scholar to identify and screen those studies. We reconducted these searches in May 2022. Additionally, we contacted the authors of the original studies for unpublished or additional data.

Population

We included trials with adult patients (age ≥18 years) admitted to ICUs and their informal caregivers; these trials were randomized during both ICU and hospital discharge. We included studies involving informal caregivers regardless of whether the admitted patient survived. We excluded studies involving patients and their caregivers who were younger than 18 years, did not provide consent for participation, or showed cognitive impairment. Furthermore, studies involving patients or caregivers who had experienced myocardial infarction or were in their perioperative period were excluded. In this article, we have referred to our target population of “critically-ill adult patients” as “patients.” if not necessary.

Interventions

We defined intervention as a service or program initiated after ICU discharge (within one month after hospital discharge), including multidisciplinary interventions, follow-up clinics, and other programs. In the included studies, we recognized counseling such as cognitive-behavioral therapy, that interventions target mental health conditions. In addition, we included psychological intervention performed as needed after monitoring. We incorporated all intervention periods by all professionals. In the included studies, nurses and physicians intervening in therapies had been trained for each study.

We excluded studies involving interventions in the ICU that were comprised of participant-led initiatives like ICU diaries and ICU records, interventions that provided general information pertaining to post-intensive care syndrome using web tools or video materials, or that compared enhanced physical rehabilitation with usual care. We did not predefine the details of the psychological interventions because we wanted to verify interventions that improved psychological outcomes other than physical and diary interventions.

Outcomes

We included trials with defined clinical outcomes, such as symptoms of depression and PTSD, and all adverse events were considered primary outcomes among patients and caregivers (Marra et al., 2018). Additional outcomes among patients included anxiety, health-related quality of life (HR-QoL), pain, readmission, and long-term mortality; additional outcomes among caregivers included anxiety and HR-QoL. We followed core outcome sets (Angus & Carlet, 2003; Major et al., 2016; Needham et al., 2017). We selected outcomes for mental health as primary outcomes. We defined depression, PTSD, and anxiety as the prevalence rate of significant symptoms based on definitions by the included studies’ authors, measured between three months and one year after randomization or ICU discharge. We defined adverse events using the incidence proportion of all adverse events set by the original authors during the follow-up period of included studies. We defined HR-QoL using a mental component summary of the Medical Health Survey Short-Form 36 (SF-36), measured between three months and one year after randomization or ICU discharge. SF-36 was used for self-reported evaluation scales for the evaluation of HR-QoL (Angus & Carlet, 2003; Needham et al., 2017). If the outcome of HR-QoL was measured by other self-reported evaluation scales in included studies, we assessed whether the scales could be synthesized with SF-36. We defined pain using self-reported evaluation scales for pain set by the original authors, measured between three months and one year after randomization or ICU discharge. We defined readmission as the proportion of readmission (at least once) during the follow-up period of the included studies. For long-term mortality, we collected the reported mortality at the longest timepoint available in the study, which ranged between 3 and 12 months after randomization.

Selection process

Three reviewers (SY, YK, and KS) independently screened the titles and abstracts of records during the initial screening. We assessed records—included in the initial screening—for eligibility based on the inclusion criteria by reading the full texts. We resolved disagreements between two reviewers via discussion with a third reviewer (TS) to achieve consensus. We combined machine learning classifiers during the selection process (Marshall et al., 2018).

Data collection process

Three reviewers (SY, YK, and KS) independently extracted data from the included studies using a standardized data collection form. We pre-checked the form by using 10 randomly selected studies. We extracted the following characteristics:

Methods: Study design, study follow-up period, and study country;

Participants: Country, setting, mental condition (depression, PTSD, and anxiety), sample size, age, relationship of informal caregivers with patients, and attrition;

Interventions: type, intervention about the psychological problem, providers, media, initiation, duration, and frequency;

Outcomes: primary and additional outcomes specified and collected, and the timepoints reported.

Study risk-of-bias assessment

Two to three reviewers (SY, YK, and KS) independently classified the risk of bias as “low”, indicating “some concerns”, or “high” based on the Risk-of-Bias 2.0 (Sterne et al., 2019). We resolved disagreements between two reviewers via discussion with the third reviewer (TS) to achieve consensus. As participants could not be blinded to the intervention owing to its nature, we assessed the overall risk-of-bias using four domains, which excluded the estimation of measurement-of-outcome.

Effect measures

We analyzed the dichotomous variables by calculating risk ratios (RR) with 95% confidence intervals (CIs). We analyzed the continuous variables using standard mean differences (SMD) with 95% CI.

Synthesis methods

We synthesized the collected variables (except for adverse events) using the random-effects method; data for patients and informal caregivers were synthesized separately. We used the Review Manager software (RevMan 5.4.2) for quantitative synthesis.

Dealing with missing data

We used available data published and inquired to authors. We performed (modified) intention-to-treat data for all dichotomous data as much as possible. For continuous data, we did not impute missing data and performed a meta-analysis of the available data in the original studies and the converted data from available data based on the method in the Cochrane handbook (Higgins et al., 2020).

Assessment of heterogeneity

We assessed heterogeneity by visual inspection of the forest plot and I² statistics (I² values of 0% to 40%: might not be important; 30% to 60%: may represent moderate heterogeneity; 50% to 90%: may represent substantial heterogeneity; 75% to 100%: considerable heterogeneity). We performed Cochrane Chi² test(Q-test) for I² statistic and defined P values less than 0.10 as statistically significant.

Sensitivity analysis and subgroup analysis

We conducted the sensitivity analysis and subgroup analysis for the primary outcomes where sufficient data were available. We conducted sensitivity analysis of patients using studies measured by the Depression subscale of the Hospital Anxiety and Depression Scale(HADS-D) score for depression, studies measured by the Impact of Event Scale-Revised (IES-R) score for PTSD, and exclusion of imputed data. We conducted the sub-group analyses by timing for initiation of follow-up(in-hospital, out-hospital, or in- and out-hospital). For analysis for informal caregivers, we conducted sensitivity analysis using studies measured by IES-R scores for PTSD. We divided the ICU survivors and non-survivors in the sub-group analyses for informal caregivers.

Reporting bias assessment

We identified the number of studies that had not been published on ClinicalTrials.gov and WHO ICTRP. We assessed outcome reporting bias by comparing the outcomes defined in trial protocols with the outcomes reported in the publications. We assessed the publication bias of outcomes by visual inspection of the funnel plots.

Certainty assessment

Two reviewers (SY and TU) evaluated the certainty of evidence based on the Grading of Recommendations Assessment, Development and Evaluation (GRADE) approach (Hultcrantz et al., 2017). We resolved disagreements between two reviewers via discussion with the third reviewer (KY) to achieve consensus. We generated a table to summarize the findings of the seven outcomes (except for long-term mortality) using GRADE Pro GDT (https://gradepro.org) based on the Cochrane Handbook (Higgins et al., 2020). We selected the following outcomes for patients: (1) depression, (2) PTSD, (3) all adverse events, (4) anxiety, (5) HR-QoL, (6) pain, and (7) readmission. We selected the following outcomes for informal caregivers: (1) depression, (2) PTSD, (3) all adverse events, (4) anxiety, and (5) HR-QoL.

Difference between protocol and review

We did not conduct Egger’s test as we synthesized data from fewer than 10 studies. We could not conduct planned sensitivity and sub-group analyses for PTSD and adverse events among patients and depression and adverse events among informal caregivers. We added a sub-group analysis for the endpoints of the measured outcomes, dividing them into 6 months and 12 months.

Depression

As shown in Fig. 2 and Table 2, ICU follow-ups resulted in little to no differences in the prevalence rate of depressive symptoms among patients (RR 0.89, 95% CI [0.59, 1.34]; I² = 1%; four studies, 758 patients; low-certainty evidence) (Abdelhamid et al., 2021; Jensen et al., 2016; Jones et al., 2003; Schmidt et al., 2016; Vlake et al., 2021); we detected slight heterogeneity. Planned sensitivity analyses of studies using the Depression subscale scores of the Hospital Anxiety and Depression Scale (HADS-D) yielded similar findings (RR 0.90, 95% CI [0.50–1.63]). Planned sensitivity analysis that excluded the imputed data showed a similar trend (RR 1.08, 95% CI [0.55–2.09]). Sub-group analysis for the timing of follow-up initiation showed a similar trend in the group of initiation from both ICU discharge and hospital discharge. In the sub-group analysis, there was no difference in the endpoint to measure depressive symptoms between 6 months and 12 months. Details of the analysis are provided in Fig S2.

Post-traumatic stress disorder

ICU follow-ups resulted in little to no differences in the prevalence rate of PTSD symptoms among patients (RR 0.84, 95% CI [0.55–1.30]; I² = 53%; four studies, 732 patients; low-certainty evidence) (Jensen et al., 2016; Jones et al., 2003; Schmidt et al., 2016; Vlake et al., 2021); we detected moderate heterogeneity (Fig. 2 and Table 2). Planned sensitivity analysis of studies using the Impact of Event Scale- Revised scores (IES-R) yielded similar results (RR 0.51, 95% CI [0.08–3.23]). The planned sensitivity analysis that excluded the imputed data generated similar findings (RR 1.06, 95% CI [0.75–1.50]; Fig. S2). Sub-group analysis for the endpoint to measure PTSD symptoms showed a similar trend in the endpoint to measure depressive symptoms between 6 months and 12 months. Details of the analysis are provided in Fig. S2.

Adverse events

Although evidence indicates considerable uncertainty, ICU follow-ups resulted in little to no differences in the occurrence of adverse events (Vlake et al., 2021) (Table 2). Two studies included adverse events as outcome measures (Bloom et al., 2019; Vlake et al., 2021). One published article (Bloom et al., 2019) did not report the results pertaining to adverse events, and we could not obtain information about adverse events from its authors. This study defined adverse events as the need for intervention to prevent events such as mortality, prolonged hospitalization, acquisition of disability, congenital anomalies, and birth defects. Another study (Vlake et al., 2021) defined adverse events as incidents of cybersickness, delirium, or the use of haloperidol. Considering the clinical heterogeneity in studies, we included all types of adverse events except for cybersickness.

Anxiety

ICU follow-ups resulted in little to no differences in the prevalence rate of anxiety symptoms among patients (RR 1.04, 95% CI [0.68–1.60]; I² = 0%; two studies, 488 patients; low certainty of evidence) (Jensen et al., 2016; Jones et al., 2003); no significant heterogeneity was detected (Table 2 and Fig. S3).

Health-related quality of life

ICU follow-ups resulted in little to no differences in the HR-QoL scores among patients (SMD 0.05, 95% CI [−0.08–0.18]; I² = 0%; seven studies, 905 patients; low-certainty evidence) (Abdelhamid et al., 2021; Cox et al., 2018; Cox et al., 2019; Cuthbertson et al., 2009; Jensen et al., 2016; Schmidt et al., 2016; Vlake et al., 2021); no significant heterogeneity was detected (Table 2 and Fig. S3). Of the seven studies, four measured the HR-QoL using the Mental Component Summary (MCS) of the Short-Form-36 (SF-36) (Abdelhamid et al., 2021; Cuthbertson et al., 2009; Jensen et al., 2016; Schmidt et al., 2016), one study used the MCS of the SF-12 (Vlake et al., 2021), and two studies used the EuroQoL Visual Analogue Scale (EQ-VAS) (Cox et al., 2018; Cox et al., 2019). The analysis of studies using the MCS of the SF-36 and the SF-12 yielded similar findings (SMD 0.04, 95% CI [−0.11–0.19]).

Pain

ICU follow-ups resulted in little to no differences in the pain scores among patients (SMD −0.08, 95% CI [−0.32, 0.17]; I² = 0%; three studies, 258 patients; low-certainty evidence) (Abdelhamid et al., 2021; Schmidt et al., 2016; Vlake et al., 2021); no significant heterogeneity was detected (Table 2 and Fig. S3). One study (Schmidt et al., 2016) measured pain intensity using the Graded Chronic Pain Scale; one study (Abdelhamid et al., 2021) used the pain comportment of the SF-36. For one study (Vlake et al., 2021), we obtained data for the pain comportment of the SF-12 which was converted to the VAS 100 scale via author inquiry.

Readmission

ICU follow-ups resulted in little to no significant in the proportion of patients readmitted to the hospital during follow-up periods (RR 0.95, 95% CI [0.77–1.16]; I² = 18%; seven studies, 1,016 patients; low certainty evidence) (Abdelhamid et al., 2021; Bloom et al., 2019; Cox et al., 2018; Cox et al., 2019; Daly et al., 2005; Jensen et al., 2016; McWilliams, Benington & Atkinson, 2016); no significant heterogeneity was detected (Table 2 and Fig. S3).

Long term mortality

ICU follow-ups resulted in little to no differences in long-term mortality among patients (RR 0.95, 95% CI [0.74–1.21]; I² = 0%; nine studies, 1,608 patients) (Abdelhamid et al., 2021; Cox et al., 2018; Cuthbertson et al., 2009; Jensen et al., 2016; Jones et al., 2003; Kredentser et al., 2018; Schmidt et al., 2016; Valsøet al., 2020; Vlake et al., 2021) (Fig. S3); no significant heterogeneity was detected.

Depression

Although the evidence indicated considerable uncertainty, ICU follow-ups increased the prevalence rate of depressive symptoms—measured using the HADS-D—among informal caregivers (RR 1.58 95% CI [1.01–2.46]; one study, 188 caregivers; very low-certainty evidence) (Kentish-Barnes et al., 2017) (Table 3). However, the other two studies (Bohart et al., 2019; Cox et al., 2018) did not report the proportion of informal caregivers with depressive symptoms, but instead provided their HADS-D scores. The point estimate of HADS-D score was higher in the ICU follow-up groups than control; thus, no inconsistencies were observed.

Post-traumatic stress disorder

Although the evidence indicated considerable uncertainty, ICU follow-ups increased the prevalence rate of PTSD symptoms—measured using the IES-R—among informal caregivers (RR 1.36, 95% CI [0.91–2.03]; I² = 19%; two studies, 303 caregivers; very low certainty of evidence) (Bohart et al., 2019; Kentish-Barnes et al., 2017) (Fig. 3 and Table 3); we detected slight heterogeneity. Planned sensitivity analysis of studies using the IES-R showed that ICU follow-ups significantly increased the proportion of patients with PTSD(RR 1.51, 95% CI [1.09–2.09]) (Fig. S4). One study (Cox et al., 2018) measured the IES-R scores and not the proportion of informal caregivers with PTSD; the point estimate of the IES-R scores was higher for the ICU follow-up group. In a sub-analysis, we found that only caregivers with non-survivors developed PTSD owing to ICU follow-ups (Fig. S4). In another sub-analysis, there was no difference in the endpoint to measure PTSD symptoms between 6 and 12 months.

Adverse events

Eligible studies with informal caregivers did not define any adverse events (Table 3).

Anxiety

Although the evidence indicated considerable uncertainty, ICU follow-ups increased the prevalence rate of anxiety symptoms, measured using the Anxiety subscale of the HADS (HADS-A), among informal caregivers (RR 1.17, 95% CI 0.83 to 1.63; two studies, 272 caregivers; very low-certainty evidence) (Jones et al., 2004; Kentish-Barnes et al., 2017) (Table 3 and Fig. S5); no significant heterogeneity was detected (I² = 0%). One study (Cox et al., 2018) measured the HADS-A scores and not the proportion of caregivers with anxiety; the point estimate of the HADS-A scores was higher for the ICU follow-up group.

Health-related quality of life

Although the evidence indicated considerable uncertainty, ICU follow-ups had little to no effect on the HR-QoL measured using the MCS of the SF-36 among informal caregivers (MD −0.70, 95% CI [−4.51, 3.11]; I² = 0%; two studies, 133 caregivers; very low certainty of evidence); no significant heterogeneity was detected (Ågren et al., 2019; Bohart et al., 2019) (Table 3 and Fig. S5).