Communication tools for end-of-life decision-making in the intensive care unit: a systematic review and meta-analysis

Background For many patients admitted to the intensive care unit (ICU), preferences for end-of-life care are unknown, and clinicians and substitute decision-makers are required to make decisions about the goals of care on their behalf. We conducted a systematic review to determine the effect of structured communication tools for end-of-life decision-making, compared to usual care, upon the number of documented goals of care discussions, documented code status, and decisions to withdraw life-sustaining treatments, in adult patients admitted to the ICU. Methods We searched multiple databases including MEDLINE, Embase, CINAHL, ERIC, and Cochrane from database inception until July 2014. Two reviewers independently screened articles, assessed eligibility, verified data extraction, and assessed risk of bias using the tool described by the Cochrane Collaboration and the Newcastle Ottawa Scale. Pooled estimates of effect (relative risk, standardized mean difference, or mean difference), were calculated where sufficient data existed. GRADE was used to evaluate the overall quality of evidence for each outcome. Results We screened 5785 abstracts and reviewed the full text of 424 articles, finding 168 eligible articles, including 19 studies in the ICU setting. The use of communication tools increased documentation of goals-of-care discussions (RR 3.47, 95 % CI 1.55, 7.75, p = 0.020, very low-quality evidence), but did not have an effect on code status documentation (RR 1.03, 95 % CI 0.96, 1.10, p = 0.540, low-quality evidence) or decisions to withdraw or withhold life-sustaining treatments (RR 0.98, 95 % CI 0.89, 1.08, p = 0.70, low-quality evidence). The use of such tools was associated with a decrease in multiple measures of health care resource utilization, including duration of mechanical ventilation (MD −1.9 days, 95 % CI −3.26, −0.54, p = 0.006, very low-quality evidence), length of ICU stay (MD −1.11 days, 95 % CI −2.18, −0.03, p = 0.04, very low-quality evidence), and health care costs (SMD −0.32, 95 % CI −0.5, −0.15, p < 0.001, very low-quality evidence). Conclusions Structured communication tools may improve documentation of EOL decision making and may result in lower resource use. The supporting evidence is low to very low in quality. Further high-quality randomized studies of simple communication interventions are needed to determine whether structured, rather than ad hoc, approaches to end-of-life decision-making improve patient-level, family-level, and system-level outcomes. Trial registration PROSPERO CRD42014012913 Electronic supplementary material The online version of this article (doi:10.1186/s13054-016-1264-y) contains supplementary material, which is available to authorized users.


Rationale
With its advanced technology, the intensive care unit (ICU) can provide life-saving medical treatment to the sickest of patients; however, that same technology can also prolong the dying process for patients who are unlikely to survive. Furthermore, many people who are approaching the end of life (EoL), either due to advancing age or progressive disease, would opt for fewer invasive and aggressive treatments in favor of a more palliative or comfort-based approach if asked, but such preferences are often undocumented in the medical record [1]. As a result, ICU clinicians and the substitute decision-makers (SDMs) often engage in goals-of-care discussions to interpret the patient's known values and preferences in the context of their illness, and to decide which ICU treatments would be in keeping with their wishes [2]. Given the medical and moral complexity of such discussions, and the need to conduct them under acute, often stressful conditions, many communication tools (including decision aids, structured meeting plans, and educational interventions) have been developed in order to assist SDMs and clinicians with EoL decision-making.
However, there remains uncertainty as to whether the use of structured communication tools for EoL decisionmaking is superior to usual care. Therefore, we conducted a systematic review of the medical literature to determine the impact of communication tools for EoL decision-making in the ICU on the following outcomes: the number and quality of EoL discussions between SDMs and healthcare providers (HCPs); the documentation of code status; and decisions to withdraw or withhold life-sustaining treatments.

Protocol and registration
The protocol for this review is available in the PROS-PERO registry (http://www.crd.york.ac.uk/PROSPERO/ display_record.asp?ID=CRD42014012913).

Eligibility criteria
We included randomized controlled trials (RCTs) or prospective observational studies with a control group (including pre-post studies in which participants acted as their own control) published as articles in peerreviewed journals, restricted to the English language. To be eligible for this review, studies must have included patients over the age of 18 years, and evaluated a communication tool to assist in EoL decision-making in comparison to a control group.
For our study, we defined a communication tool as any intervention designed to directly assist individual patients and SDMs in decision-making, or their clinicians to better facilitate the EoL decision-making process. This included traditional decision aids in any format (paper, video, computer, etc.), and other structured approaches to assisting decision-making, including organized meeting plans, consultation with services for the purpose of assisting with decision-making (e.g., ethics or palliative care), and educational interventions on EoL care options. Interventions designed solely for informationsharing (e.g., breaking bad news, providing emotional support) were excluded, because although such interventions may affect decisions at the EoL, it is not their explicit purpose to do so (Table 1). Communication tools for ICU settings are distinct from those in the ambulatory or non-ICU inpatient setting, as they are directed towards patients and SDMs, and not towards patients, who are generally too ill to participate in EoL decisionmaking; furthermore, communication tools in the ICU setting are usually directed towards decisions about current acute care, rather than advance care planning to prepare for future acute illness. For this reason, we report findings from eligible studies that were conducted in the ICU. Studies conducted in the ambulatory setting and in the inpatient non-ICU setting, and studies of educational interventions for improving clinicians' competencies in EoL communication and decision-making will be analyzed and reported separately.
For this paper, which focuses on the ICU setting, our primary outcomes were: (1) proportion of patients with documented goals of care discussions; (2) proportion of patients with documented code status; and (3) proportion of patients with new decisions to withdraw or withhold life-sustaining treatments, as we considered these to be the most immediate patient-level outcomes affected by goals-of-care discussions. Our secondary outcomes were (1) patient or family satisfaction with EoL care; 2) patient or family knowledge about EoL care, including knowledge about palliative care and intensive care; (3) quality of communication between the patient/SDM and HCPs; (4) health care resource utilization (including duration of mechanical ventilation, ICU length of stay, hospital length of stay, and financial cost of care); and (5) the acceptability of the intervention. Secondary outcomes 1-3 were selected to help understand the mechanisms by which communication tools affect EoL decisions, while secondary outcome 4 was chosen to understand the resource implications of using these tools, and, for study participants who received the intervention, secondary outcome 5 was chosen to help understand whether the use of such tools, irrespective of other benefits, would be acceptable to patients/SDMs.

Information sources and search strategy
We searched the following databases from database inception until the present: MEDLINE (1946 through July 2014); Embase (1980 through July 2014); CINAHL (1982 through July 2014); Cochrane Database of Clinical Controlled Trials (2005 through July 2014); and ERIC (1966 through July 2014). Our search terms included: "communication," "decision-making," "end-of-life," "cardiopulmonary resuscitation" (complete electronic search strategies for each database can be found in Additional file 1: Appendix 1). We also hand-searched the reference lists of eligible articles to identify further articles for screening.

Study selection
Retrieved titles and abstracts were screened independently and in duplicate by two reviewers (HC, SO) for potential eligibility using standardized, piloted screening forms. The full text of all articles that passed initial screening by either reviewer were then assessed independently and in duplicate using standardized, piloted eligibility forms. Disagreement about study eligibility was resolved by consulting with a third reviewer (JY). Reviewers were not blinded as to article authors, journal, or results, when screening for eligibility. Kappa statistics were calculated to assess the inter-rater reliability of the screening and eligibility phases [3]. Studies were then divided based on study type into outpatient, inpatient, or intensive care unit settings; or educational interventions for clinicians.

Data collection process and data items
Study data were collected using standardized, piloted online forms by the two reviewers (HC and SO). Study authors were contacted to obtain missing data for our primary or secondary outcomes. We collected data on publication information, study dates and population characteristics, study interventions, our primary and secondary outcome measurements, and the study methods required to assess the risk of bias in individual studies.

Risk of bias in individual studies
For RCTs, we assessed the risk of bias in individual studies using the Cochrane risk-of-bias tool with regard to random sequence generation, allocation concealment, blinding of participants and personnel, incomplete outcome data, selective reporting, and where appropriate, cluster design [4]. Each domain was assessed independently by both reviewers and reported as having high, low, or uncertain risk of bias. Studies were considered to have low risk of bias if assessed as having low risk of bias in all domains; to have uncertain risk of bias if assessed as having uncertain risk of bias in at least one domain and no domains at high risk of bias; and to have high risk of bias if there was high risk of bias in any domain. For studies with uncertain risk of bias, we attempted to contact study authors to clarify the relevant issue(s), and revised the overall study risk-of-bias accordingly. Disagreement between reviewers about risk of bias was resolved by consulting with a third reviewer (JY). For observational cohort and case-control studies, we used the Newcastle-Ottawa scale to assess risk of bias, using seven stars for our cutoff for good vs. poor studies [5]. For uncontrolled before-after studies, the National Institutes of Health rating system was applied [6].

Synthesis of results
We used Revman 5.3 software to conduct our analyses [7]. For each outcome, similar studies were pooled, with priority given to randomized trials i.e., data were sought from RCTs first and non-randomized studies were only used in the absence of data from RCTs. However, given the small number of RCTs found, we also reported the results of observational studies in separate analyses for comparison. Pooled outcomes (standardized mean difference (SMD) or mean difference (MD) for continuous variables, relative risk (RR) for dichotomous variables) and 95 % confidence intervals (95 % CI) were calculated using a random-effects model. For cluster-randomized RCTs, intraclass correlation coefficients were used to adjust the effective sample size to adjust for clustering effects within groups [8]. Where standardized deviations were missing, or represented by interquartile ranges, estimates were generated using the calculations described in the Cochrane handbook [9].

Investigation of heterogeneity and subgroup analyses
Clinical heterogeneity was assessed by reviewers investigating study populations, interventions, and comparisons. If the studies were considered to be of sufficient similarity for data pooling, heterogeneity was quantified for each outcome of interest and using the I 2 statistic, with values greater than 50 % indicating substantial heterogeneity [9]. We conducted one exploratory subgroup analysis, comparing the results of the communication tools between patients who survived to discharge from ICU (ICU survivors) and those who died in the ICU (ICU nonsurvivors). Several of the studies only reported outcomes for ICU non-survivors. We chose to report these results separately, as the subgroup of ICU non-survivors may represent a population more likely to have withdrawal of life support (due either to pre-existing advance directives, severity of illness, or medical comorbidities), and would thus be inappropriate to pool with studies that included both ICU survivors and non-survivors. This subgroup is also of interest as it may correlate with a group of patients for whom ICU resources constitute 'non-beneficial care' [10].

Publication bias
Publication bias was assessed using visual inspection of funnel plots generated in Revman 5.3, where sufficient numbers of studies existed to permit interpretation [11].

Rating of quality of evidence
We used the Grading of Recommendations Assessment, Development, and Evaluation (GRADE) approach to assess the quality of evidence for each outcome [12]. Outcomes for which the majority of evidence was derived from RCTs was considered to initially be of high quality, while those from which the majority of evidence was from observational studies started at low quality, with both types rated up or down after considering the risk of bias across studies (e.g., publication bias), potential biases and their direction within each study, and the imprecision, inconsistency, and indirectness of the evidence. GRADE summary of findings tables were generated using the online GradePRO software [13].

Study selection
Initial database searches retrieved 5727 articles, with 58 articles found using reference screening. After exclusion of duplicate references and conference abstracts, title and abstract screening resulted in 424 articles eligible for full text review (κ = 0.648; 95 % CI 0.601, 0.695). A total of 168 articles were eligible for our systematic review after full-text review and additional manual reference screening, of which 19 studies were conducted in the ICU setting ( Fig. 1).

Study characteristics Study settings and populations
Study publication dates ranged from 1995 to 2014. Most were conducted in the USA or Canada (n = 16, 84 %), with three conducted in France (16 %). Three studies were conducted in medical ICUs, two in surgical ICUs, and fourteen in mixed medical-surgical or non-specified ICU type. Seven studies specifically focused on patients with longer ICU or hospital stays [14][15][16][17][18][19][20]. Two studies were conducted in patients considered likely to die [21,22]. One study specifically evaluated interventions in patients for whom there were conflicts over goals of care [23] (Table 2).

Risk of bias within studies
Five of the included studies were RCTs, of which three were considered to have an overall low risk of bias [14,21,24], one to have uncertain risk of bias [23], and one to have high risk of bias [22]. The fourteen observational studies were prospective cohort studies, with four having Newcastle-Ottawa scores considered to be good [16,17,25,26], and the remainder of cohort studies having a poor quality rating (Tables 3 and 4).
One RCT (108 patients) [14] reported financial costs amongst the subgroup of ICU non-survivors, finding no differences in health care costs (US$167,350 vs. US$164,670, p = 0.92). Two observational studies (124 patients) [17,29] also found no differences in this subgroup (SMD −0.76, 95 % CI −1.57, 0.04, p = 0.06) 5. Patient and family acceptability of the intervention: only two RCTs reported measures of the  [23,24]. One observational study of a video decision-support tool also reported that a majority of patients were comfortable with the intervention, found it helpful, and would recommended it to others [30] Discussion Given the centrality of EoL decision-making to the care provided in the ICU, it is surprising that our review found so few studies evaluating the use of structured communication tools to assist SDMs and clinicians with this process. Only very low-quality evidence (from one RCT with a high risk of bias and four observational studies) was found that evaluated whether such tools increase the number of documented goals-of-care discussions, indicating that future studies are very likely to alter our estimates of effect. We found low-quality evidence that the use of structured communication tools does not increase the number of patients with documented code status or DNR status, or decisions to withdraw or withhold treatment. Equally surprising is that the use of communication tools, in comparison to usual care, had minimal to no effect in the studies we found that reported on these outcomes. This may be because the studies lacked sufficient power to find a true difference, the tools themselves were ineffective, the tools have variable efficacy without a significant class effect for their use, or because there is minimal room for measurable improvements beyond standard care once the patient is in the ICU. If the latter is true, it may be more effective to direct structured interventions earlier in the course of a patient's care, before they are in the ICU, when there is a greater opportunity for such interventions to help patients and families create documented care plans, and improve the concordance between the care they receive, and the care they wish to receive. The one area where significant improvements with the use of structured communication tools was demonstrated was in health care resource utilization, although the quality of evidence was again very low. The existing evidence suggests that the use of such tools may decrease the number of days of mechanical ventilation, length of ICU stay, number of hospital days, and financial costs of care. These results were not seen in the exploratory subgroup analysis of ICU non-survivors, in which similar reductions in health care utilization were not demonstrated.  The mechanisms by which these interventions could decrease health care resource utilization without increasing the number of patients with documented code status or withholding or withdrawal of treatments is unclear. The most obvious possibility is that the studies that demonstrated a decrease in resource use were not the same studies that failed to find improvements in documented goals-of-care discussions, code status documentation, and decisions to withdraw or withhold therapies, and thus, heterogeneity between the studies may account for the differences. Or, it may be that in the studies evaluating these tools, while there actually were increases in the number of decisions to withhold or withdraw life-sustaining therapies, such decisions were poorly documented. If there is a clear consensus between SDMs and clinicians that the current ICU level of care is futile or unwanted, clinicians may fail to document resuscitation status and decisions to withdraw or withhold treatment, whereas in conflict-charged situations, documentation may be extensive. Another possible explanation for decreased resource use is that the use of communication tools may result in less aggressive, less expensive care, in ways that are not always considered to be withdrawing or withholding of lifesustaining therapies (e.g., fewer diagnostic tests and invasive procedures, or the use of one-way extubation, with rapid transfer out of ICU).
Overall, the low quality of evidence suggests that more high-quality randomized trials are needed to determine whether the use of structured communication tools to assist SDMs and clinicians with EoL decision-making has a major effect upon outcomes that are important to patients. Given the considerable experience ICU clinicians have in assisting SDMs with EoL decision-making, it may be reasonable to focus such studies on patient populations for which extra assistance would be needed. For instance, high-quality randomized trials of simple structured communication tools would be of value in Fig. 6 Quality of communication between family/substitute decision-maker and health care providers Fig. 7 Health care resource utilization -duration of mechanical ventilation (days). a Studies including both intensive care unit survivors and non-survivors. b Studies including intensive care unit non-survivors only ICU patients with conflict over goals of care, or in longstay ICU patients. We would recommend that such studies report outcomes related to three domains: patient-level outcomes (e.g., documented code status, withdrawal or withholding of treatment, and mortality); family-level outcomes (satisfaction, and long-term mental health outcomes, including depression and anxiety), and system-level outcomes (resource use, including bed use), given that decisions about whether ICUs should implement structured communication tools to assist EoL

Strengths
The strength of our study lies in its broad search strategy to find communication tools to assist in EoL decision-making. This allowed us to capture the full range of tools available to ICU clinicians, including traditional decision aids, structured meeting plans, educational interventions, the use of consulting services, and complex quality improvement programs. Further strengths of our study include: the rigorous search strategies; the input of two separate authors in assessing studies for screening, eligibility, and risk of bias, with secondary checking and verification of data extraction; and our use of GRADE to assess the overall quality of evidence for each outcome.

Limitations
Our study has limitations related both to its methods, and to the studies we retrieved. First, our use of very broad inclusion criteria resulted in difficulty in identifying studies that evaluated interventions related to EoL decision-making. Extensive review of abstracts, study methods, and outcomes was often required to determine whether or not a study intervention was indeed a communication tool designed to assist in EoL decisionmaking, or a communication tool for another purpose (e.g., not all palliative care consultations have the purpose of assisting decision-making; instead, many have the purpose of providing emotional comfort and symptom control). By not limiting our review to one specific type of intervention, it became more difficult to identify the studies we wanted to include; given this difficulty, it is possible that our review failed to identify some potentially relevant articles, despite our rigorous search.
Our second limitation was related to the outcomes reported in the studies we found. Although documented code status and decisions to withdraw and withhold treatments were commonly reported, they are ultimately surrogate measurements for unknown patient wishes. Ideally, studies would be able to directly assess patient wishes for care and assess concordance between patient preferences and care plans, and patient preferences for care and the care actually received. We recognize the fact that many patients may not have a completed advance directive or may not have discussed EoL preferences with their SDMs; however, we believe that reporting concordance between patient wishes and the actual care received at the EoL would still be an important, even if imperfect, outcome for future studies to report.
The third limitation of our review is the limited number of studies retrieved, which were varied in quality and intervention type. We found very few RCTs that reported on our outcomes of interest, with the result that one trial by Curtis et al., considered to be at substantial risk of bias, given the variability between clusters, lack of adjustment for cluster design for some outcomes, and loss of a cluster, was responsible for much of the randomized data in our primary outcomes. The quality of evidence for all of our outcomes is of very low to low quality according to GRADE, meaning our estimated effects are very likely to be affected by future studies.

Conclusions
We found very low-quality evidence that the use of structured communication tools increases the number of documented goals-of-care discussions and low-quality evidence that they do not affect the number of patients with documented code status/DNR forms or decisions to withdraw/withhold life-sustaining treatments. We also found very low-quality evidence that the use of structured communication tools results in reduced health care resource utilization compared to usual care. More high-quality RCTs are required to evaluate the effects of structured communication tools to facilitate EoL decision-making in the ICU upon system-level, family-level, and patient-level outcomes, including concordance between patient wishes for care and the care received at the end of life.