Critical care transition programs on readmission or death: A systematic review and meta‐analysis

Deterioration after ICU discharge may lead to readmission or even death. Interventions (eg, critical care transition programs) have been developed to improve the clinical handover between the ICU and the ward. We conducted a systematic review with meta‐analysis and trial sequential analysis (TSA) according to Cochrane Handbook and Grading of recommendations, assessment, development and evaluations (GRADE) methodology to assess the impact of these interventions on readmission and death (PROSPERO, no CRD42019121746).


| INTRODUC TI ON
Leaving the intensive care unit (ICU) may not be the end of critical illness for all patients. Despite recovery and no longer need of lifesustaining therapies, deterioration may occur after ICU discharge leading to unplanned readmission or even death. Readmission to the ICU is associated with unfavorable results, such as longer hospital stay, higher mortality, 1 and increased healthcare costs. In Sweden, unplanned readmission within 72 hours after ICU discharge is considered an adverse event. 2 On the contrary discharge at the earliest appropriate time reduces unnecessary use of expensive health care and improves availability of beds for other critically ill requiring ICU admission. 3 The decision when to discharge is a demanding challenge for intensivists, influenced by individual and teamwork factors in addition to organizational issues. 4 Different interventions, for example, ICU liaison nurse (ICU LN), handover forms, outreach services, has been developed in order to improve the clinical handover between ICU and general ward at discharge. 5 At the moment, the benefits of use of ICU transition programs are not clear. Two previous systematic reviews have not been able to find a reduced mortality using ICU transition programs. 5, 6 Niven and colleagues suggested in their meta-analysis a reduced risk of ICU readmission with transition programs including an outreach team or ICU LN. 6 Van Sluisveld and colleagues found that interventions including ICU LN and handover forms resulted in improved continuity of care and in reduced adverse events but no evidence of a reduction in ICU readmission. 5 Therefore, we aimed to systematically review randomized and non-randomized studies with historical control groups to examine whether critical care transition programs reduce ICU readmission or in-hospital mortality in adult ICU discharged patients, in comparison with usual care.

| Protocol and guidance
This systematic review with meta-analysis and trial sequential analysis (TSA) was conducted using established methods recommended by the Cochrane Handbook for Systematic Review of interventions 7 and the Grading of recommendations, assessment, development and evaluation (GRADE) working group. 8 The protocol was registered in the International prospective register of systematic reviews (PROSPERO, no CRD42019121746) prior the review and deviations from the planned methods are presented in Table S1 Deviations. We reported the findings according to the preferred reporting items for systematic reviews and meta-analyses (PRISMA). 9

| Ethical issues
There is no funding for this study. This review does not require ethical approval or informed consent since there will be no direct contact with individual patients.

| Search strategy and data sources
The search strategy was developed by investigator (J.Ö.), who then  Table S2 Search.

| Data extraction and bias assessment
Two independent investigators (J.Ö. and J.G.) screened the titles and abstracts of reports and reviewed each potentially eligible study in full text. Eligible studies that met any of the following exclusion criteria were not included in the review: no control population, ICU readmission not reported, post-ICU mortality not reported. Data were extracted in duplicate using a standardized form and created tables for the characteristics and outcomes. If studies did not report important outcomes of interest for inclusion or present information required to calculate the statistics in extractable format, authors were e-mailed (two attempts) for additional data. All pre-planned ICU readmission timeframes could not be assessed due to lack of data. When more than one follow-up time was provided for the outcome data, the longest follow-up during the same hospitalization was used.

Editorial Comment
Careful clinician handover of care of patients to the next responsible care group is an integral part of daily care of patients which is presumed to contribute positively to patient safety. In this interesting systematic review of the patient transition from ICU to regular hospital ward, no firm evidence was observed to confirm patient benefit from reports of critical care transition programs, that is, structured clinical handover between the ICU and the ward.
The same investigators independently used the ROBINS-I 10 for methodological assessment of observational studies in domains: bias due to; confounding, in selection of participants into the study, in classification of interventions, due to deviations from intended intervention, due to missing data, in measurement of outcomes, and in selection of the reported result. The authors were not contacted to clarify unclear ratings. Disagreements between the two investigators were resolved by discussion with the third author (T.M.).

| Data synthesis
All pooling analyses were done using random-effect model because of clinical and methodological heterogeneity. For dichotomous outcomes, we calculated and reported relative effects as risk ratios (RR, the ratio of the risk of an unfavorable outcome among ICU discharged patients in a transition program vs usual care) with 95% CI.
A value of P < .05 was considered statistically significant. We assessed heterogeneity using the chi-squared test (threshold P = .10), and quantified using the I 2 statistic, which describes the proportion of total variance across trials that is attributed to heterogeneity.
According to the methods described in the Cochrane Handbook for Systematic Reviews of Interventions, 7 we interpreted the amount of heterogeneity from I 2 values as low 0%-40%, moderate 30%-60%, and substantial 50%-90%. We planned to analyze the following modifiers of effect: (a) Type of intervention, (b) case mix, and (c) if goals-of-care were determined before ICU discharge. We were able to conduct subgroup analysis (threshold P = .10) concerning type of intervention, but not the other domains due to insufficient data. For detection of publication bias, we used visual evaluation of funnel plot symmetry. We planned sensitivity analyses by performing metaanalyses on primary outcome based on study design and risk of bias.
We applied TSA (α = 0.05 and β = 0.20) since it reduces the risk of type-I error in a cumulative meta-analysis and may provide information on how many more patients need to be included. 12 Information size was calculated as diversity-adjusted information size, suggested by the relative risk reduction (RRR) of the intervention in the included studies. We calculated the event proportion in the control group as an unweighted mean of the proportion with the outcome in the control groups of all the included studies. Sensitivity analyses were performed assuming RRRs of 25% and 10%. Copenhagen) 13 to conduct conventional meta-analyses and trial sequential analysis, respectively.

| Grading of recommendations, assessment, development, and evaluation (GRADE)
We summarized the quality of evidence for primary outcomes applying GRADE-levels 8  possible risk of bias, inconsistency, indirectness, imprecision, and publication bias using GRADEpro GDT software. 14

| RE SULTS
The final yield was 15 studies investigating interventions facilitating the transfer from the ICU to the general ward, including 14 beforeafter and one matched cohort study. Studies were excluded from the quantitative analysis as they did not report the number of patients discharged to the ward following their initial ICU admission and/ or events (Table 1, summary of exposure by primary outcome), and the corresponding authors were either not responding or unable to provide the requisite data. [15][16][17][18] Two corresponding authors supplied originally not reported data on in-hospital mortality. 19,20 The

TA B L E 2 (Continued) (Continues)
Health Evaluation (APACHE) II score ranged between 12 23 (median) and 29 24 (Pre-RRT, mean). When comparing the cohorts within studies, three studies demonstrated a significant better health condition in the intervention group before ICU discharge. 19,21,24 In two of them the intervention group were older, 19,20 and in one male to a greater extent. 19 Patients discharged outside the hospital or dead in the ICU could not be subjected to a follow-up intervention. The studies were judged to have overall serious (n = 14) or critical (n = 1) risk of bias  Table S3 ROBINS-I.

| Readmission
Eleven studies that involved 20 475 ICU discharged patients were included in the present meta-analysis. [19][20][21][22][23][24][25][26][27][28][29] An intervention facilitating the transfer from the ICU to the general ward seems Sensitivity analyses of RRRs (25% and 10%) were consistent with primary analysis ( Figure S1). Our application of GRADE methodology led us to conclude that the accumulated evidence for this outcome is of very low quality (downgraded four levels) due to bias in the included studies, inconsistency in effect, difference in interventions, wide confidence intervals, and a required information size not met (TSA) ( Table 3, summary of findings and GRADE).

| In-hospital mortality
The pooled results showed no difference in in-hospital mortality among discharged patients in a critical care transition program as

2008-2012
Before-and-after retrospective. Note: Continuous data given as mean (SD or 95% CI) or median (interquartile range) as provided by the study authors.
Abbreviations: CCOS, critical care outreach service; CG, control group; ICU LN, intensive care unit liaison nurse; IG, intervention group; MET, medical emergency team; N/A, not available; RRT, rapid response team. a Restricted to patients discharged to the ward and receiving a scheduled CCOS visit within 48 h. Historical control group.

| Subgroup analysis
In the subgroup analysis according to type of intervention, CCOS and RRT were associated with a reduced risk or readmission and inhospital mortality (only CCOS) after ICU discharge (Figures 3 and 5).
The test for subgroup differences according to type of intervention indicated that there was no statistically significant subgroup effect (readmission: P = .16, mortality: P = .12).

| Secondary outcomes, sensitivity analysis, and publication bias
The reporting of secondary outcomes was sparse and varied, preventing conduct of a meta-analysis. However, there were no significant differences between groups. Three studies reported on LOS ICU after readmission and none of these results achieved statistical significance. 17,26,29 The nurse-led CCOR reduced LOS ICU for readmitted patients with, in mean 5-3 days 17 and in median 4-3 days, 26 respectively. Conversely, Garcea et al reported increased LOS ICU, from in mean 6.2-8.3 days following outreach. 29 Three studies reported LOS in hospital and none of these results achieved statistical significance. 20,26,29 One study reported a non-significant decrease in both ICU mortality, 18%-16%, and in-hospital mortality, 35%-26%, after outreach in readmitted patients. 26 One study investigating the RRT reported no difference between groups with a new DNR order after ICU discharge. 20 Duration of mechanical ventilation and renal replacement therapy after readmission were reported in one study. 29 After the introduction of an outreach team patients readmitted to the ICU were ventilated non-significantly longer (5.5 days vs 3.8 days) and there was no difference in renal support days. As F I G U R E 3 Forest plot examining the risk of ICU readmission among patients within a transition program vs usual care. Horizontal lines represent 95% confidence intervals (CI). CCOS, critical care outreach service; I 2 , heterogeneity; ICU LN, intensive care unit liaison nurse; MET, medical emergency team; RRT, rapid response team; TP, transition program all studies were non-randomized and judged at least serious risk of bias, no sensitivity analysis was conducted. The funnel plots used to evaluate the publication bias did not demonstrate any obvious asymmetry (Figures S3 and S4). Using the GRADE methodology we are led to conclude that the quality of evidence is very low.

| D ISCUSS I ON
The modest reduced risk of ICU readmission in our review is con- Sluisveld and colleagues systematically reviewed interventions with the aim to improve the ICU discharge to general ward. 5 Inclusion criteria regarding outcome reported were less strict than ours-they accepted any outcome measure addressing the quality or safety of the discharge. Consequently, five types of interventions and four types of outcomes were identified. Effective interventions included the ICU LN and handover forms in improving the continuity of care and reducing preventable adverse effects, however, they found no improvement in ICU readmission or mortality.

F I G U R E 4
Trial sequential analysis (TSA) for a relative risk reduction of readmission of 22% with a critical care transition program after ICU discharge in 11 studies with 20 475 patients. A required diversity-adjusted information size of 31 339 patients was calculated based on a control event proportion of 7.0%, a relative risk reduction of readmission of 22% suggested by all studies, α = 0.05 two-sided and β = 0.20 (power 80%), and diversity D2 = 76%. The cumulative Z-curve crosses the traditional boundary (P = .05) but not the trial sequential monitoring boundary indicating lack of firm evidence for a beneficial effect of 22% relative risk reduction. There is insufficient information to reject or detect an intervention effect of 22% RRR of readmission as the required information size is not yet reached Moreover, one of the included studies in our review that could not be quantitatively analyzed-the study by Stelfox and colleagues, included over 30 000 patients (which is approximately 50% more the patients in our analysis), and no association to ICU readmission or mortality was found. 18 Possible explanations for findings in our review must be considered. First of all, the interventions may have been aimed at the wrong population-they were implemented as a transition program for the general ICU population, while they may be effective only for discharged patients with factors that reflect a greater severity and complexity of illness and are associated with increased risk of readmission or death. 30 And secondly, even if the interventions were aimed at patients with increased risk, the studies retrieved and analyzed data from the overall ICU discharged population which may have obscured an actual effect in a "high-risk population". Third, the outcome measured might be wrong. Although ICU readmission and mortality are common, practical and objective metrics they may not fully reflect an effectiveness of a critical care transition program in terms of facilitating the discharge. In this meta-analysis the ICU LN and a redesigned discharge process did not seem to affect the risk of readmission or death. They were, however, associated with more efficient discharges 23,27 and improved survival for patients requiring readmission. 26 And indeed, the ICU readmission rate as a quality marker of a follow-up intervention may be flawed as patients deteriorating in the wards and requiring readmission are expected to be noticed.
There was no statistically significant subgroup effect analyzing type of intervention as defined in the studies. While risk reduction in both readmission and death was mainly attributed to the studies investigating CCOS, it would be unwise to draw conclusions as an unexplained high heterogeneity within the subgroup remained and these interventions vary in tasks performed, staffing, and coverage. 28 Within the RRT subgroup the risk of readmission was reduced, but not death. In one study the in-hospital mortality actually increased. 19 These numbers were not originally reported but obtained after contact with the corresponding author. The mechanism by which the mortality increased is unclear.
Although we adopted the appropriate measures, such as a protocol registration prior to statistical analyses and following TA B L E 3 Summary of findings established methods recommended by the Cochrane Handbook for Systematic Reviews, meta-analyses are only as good as the data reported in each study. Our findings are limited by the before-after design and that not all studies could be quantitatively analyzed. In four studies we could not ascertain either the number of patients discharged, readmitted, or dead necessary for a quantitative analysis. [15][16][17][18] An additional potential limitation is the great clinical variability or heterogeneity found among the studies identified, complicating both the synthesis and interpretation of results. This variability affects both the transition programs being evaluated in the intervention and most likely the usual care (eg, pre-existing usual ICU discharge process) in the control populations. Finally, the exclusion of studies only reporting one of the study outcomes introduces a potential study selection bias affecting the strength of the meta-analytic estimate. Nonetheless, the funnel plot of the outcomes showed no apparent publication bias and in the absence of randomized controlled studies, they provide the best available evidence. This review, including over 20 000 ICU discharged patients around the world in the last 20 years, challenges the effectiveness of critical care transition programs. We found no clear benefit in terms of reducing risk of readmission or death after ICU discharge. This conclusion should be interpreted with caution, since the present review was underpowered (according to TSA) and the included observational studies were considered to be of at least serious risk of bias.
Either further robust methodology trials are required to develop F I G U R E 5 Forest plot examining the risk of in-hospital mortality after ICU among patients within a Transition Program vs usual care. Horizontal lines represent 95% confidence intervals (CI). CCOS, critical care outreach service; I 2 , heterogeneity; ICU LN, intensive care unit liaison nurse; MET, medical emergency team; RRT, rapid response team; TP, transition program effective interventions, or possibly, these outcomes as a measure of effectiveness of critical care transition programs have had their day.

CO N FLI C T O F I NTE R E S T
None of the authors declare any conflict of interest.

Jonas Österlind
https://orcid.org/0000-0001-7439-2711 Tomi Myrberg https://orcid.org/0000-0002-8802-2321 F I G U R E 6 Trial sequential analysis (TSA) for a relative risk reduction of in-hospital mortality of 18% with a critical care transition program after ICU discharge in 11 studies with 20 423 patients. A required diversity-adjusted information size of 66 920 patients was calculated based on a control event proportion of 11.3%, a relative risk reduction of readmission of 18% suggested by all studies, α = 0.05 two-sided and β = 0.20 (power 80%), and diversity D2 = 90%. The cumulative Z-curve do not cross the traditional boundary (P = .05) nor the trial sequential monitoring boundary or futility boundary suggesting a lack of firm evidence and more studies needed