Rapid response systems: a systematic review and meta-analysis By Ritesh Maharaj, Ivan Raffaele, Julia Wendon Published in: Critical Care DOI: 10.1186/s13054-015-0973-y ABSTRACT Introduction Although rapid response system teams have been widely adopted by many health systems, their effectiveness in reducing hospital mortality is uncertain. We conducted a meta-analysis to examine the impact of rapid response teams on hospital mortality and cardiopulmonary arrest. Method We conducted a systematic review of studies published from January 1, 1990, through 31 December 2013, using PubMed, EMBASE, CINAHL (Cumulative Index to Nursing and Allied Health Literature) and the Cochrane Library. We included studies that reported data on the primary outcomes of ICU and in-hospital mortality or cardiopulmonary arrests. Results Twenty-nine eligible studies were identified. The studies were analysed in groups based on adult and paediatric trials that were further sub-grouped on methodological design. There were 5 studies that were considered either cluster randomized control trial, controlled before after or interrupted time series. The remaining studies were before and after studies without a contemporaneous control. The implementation of RRS has been associated with an overall reduction in hospital mortality in both the adult (RR 0.87, 95 % CI 0.81–0.95, p<0.001) and paediatric (RR=0.82 95 % CI 0.76–0.89) in-patient population. There was substantial heterogeneity in both populations. The rapid response system team was also associated with a reduction in cardiopulmonary arrests in adults (RR 0.65, 95 % CI 0.61–0.70, p<0.001) and paediatric (RR=0.64 95 % CI 0.55–0.74) patients. Conclusion Rapid response systems were associated with a reduction in hospital mortality and cardiopulmonary arrest. Meta-regression did not identify the presence of a physician in the rapid response system to be significantly associated with a mortality reduction. Electronic supplementary material The online version of this article (doi:10.1186/s13054-015-0973-y) contains supplementary material, which is available to authorized users. INTRODUCTION Many ward patients may deteriorate to the point of unexpected ICU admission or even cardiac arrest and death. About one-half of the serious adverse events are deemed to be preventable [1]. Patients often show some signs of physiological deterioration for several hours (median 6 hours) before cardiac arrest [2, 3]. Theoretically, this would appear to be sufficient time to deliver interventions that would alter the trajectory of deterioration. This ‘failure to rescue’ is the context in which rapid response systems (RRSs) have been introduced [4]. Most RRS trials have used similar criteria for activation of the team. These include various thresholds for respiratory rate, heart rate, blood pressure and mental state as well as the ‘staff worried’ criterion [5]. Unfortunately, the determination of respiratory rate and mental state on the wards is error prone and vital signs are unlikely to be measured more frequently than five times per day outside the ICU [6–8]. Variations in activation potentially influence the effectiveness of RRSs. Non-activation and delays in activation may be associated with harm and too liberal activation may cause system fatigue [9–11]. Consequently, the time spent implementing and maintaining the concept of a RRS and the rate of RRS activation may be a contributory factor to its success. The RRS team is usually multidisciplinary in nature and tailored to meet the institutional needs and resources. In this respect there is some variation in the constitution of a RRS. In the UK the RRS team may be nurse led, and in the USA nurse or respiratory therapists may lead [12, 13]. In Australia, New Zealand and Scandinavia, a physician-led RRS is favoured [5, 14]. The optimal composition of a RRS team is unknown and may depend upon organizational structure. Organizations such as the Institute for Healthcare Improvement, the 2009 Joint Commission’s National Patient Safety Goal in the USA, the National Institute of Clinical Excellence (NICE) in the UK as well as numerous other organizations have driven the implementation of RRS teams [15, 16]. Despite high face validity, the effectiveness of RRS teams in reducing hospital mortality remains controversial. A high-quality meta-analysis in 2010 did not find evidence of RRS teams reducing hospital mortality in adults [17]. A more recent review by Winters et al. [18] included studies of varying methodological quality and design, with inconsistent findings across outcomes. Both analyses aggregated studies of varying methodology and quality, further limiting the interpretation of the results [18]. Accordingly, a systematic review and meta-regression was undertaken to assess the effect of the RRS on hospital mortality and cardiopulmonary arrest outside the ICU and to evaluate the potential relationship between the number of RRS team activations per 1000 admissions, the presence of a physician in the RRS team and the duration of the implementation phase and the effectiveness of RRS teams. METHODS A systematic review of studies published between 1 January 1990 and 31 December 2013 was conducted in accordance with published guidelines [19, 20]. We used the PubMed, EMBASE, CINAHL (Cumulative Index to Nursing and Allied Health Literature) and Cochrane Register of Controlled trials databases. Additionally, a hand search of bibliographies of key publications was performed. Search terms included ‘rapid response team’, medical emergency team’ and ‘critical care outreach’. Details of the electronic search are described in Fig. 1 with additional information provided in the supplementary appendix. Study eligibility criteria and selection The inclusion criteria for this meta-analysis was that studies had to be a study that described the effect of RRS teams in a population of hospital in-patients that included a comparison between a control cohort and intervention cohort, and provided quantitative data about mortality rates or cardiopulmonary arrests. There was no country restriction but only English language studies were included. A total of 2935 abstracts were identified by the search strategy. The titles and abstracts were independently assessed for eligibility by two investigators (RM and IR). Eight hundred and eighty-one duplicate studies were removed and a further 1994 studies were removed because they were not relevant or did not report sufficient data (Fig. 1). In cases with multiple articles with overlapping data from the same population, we included data from the most comprehensive study and excluded the other studies [21–24].Fig. 1Literature search flow diagram Data extraction process Two reviewers independently, using a standardized format, extracted the data. For each study the following data elements were extracted: year of study, type of study design (randomized controlled trial (RCT), before–after, interrupted time series), type of hospital, number of centres, type of RRS, types of interventions, period pre and post implementation, demographics, number of activations per 1000 admissions, effect on do-not-resuscitate orders, non-ICU cardiopulmonary arrest, hospital mortality, ICU admissions as well as study quality indicators. Authors were contacted for additional data as required. Risk of bias in individual studies Reviewers worked independently to assess study quality. The Newcastle Ottawa Scale (NOS) was used for assessing non-randomized studies [25]. The NOS uses a star system to evaluate the selection of study groups, the comparability of groups and the ascertainment of either the exposure or outcome of interest. The interrupted time series, controlled before–after and cluster randomized studies were evaluated using the criteria recommended by the Cochrane Effective Practice and Organisation of Care Group [26]. This included recruitment bias, baseline imbalance, loss of clusters, incorrect analysis or selective outcome reporting. To explore variability in study results (heterogeneity) we specified the following hypotheses a priori. We hypothesized that effect size may differ according to the number of activations per 1000 admissions, the duration of implementation in months and the presence of a physician on the team. The proposed standardized method to report the dose of RRS teams is activations per 1000 patient admissions [27]. Data analysis The primary outcome was hospital mortality. The secondary outcomes were non-ICU cardiopulmonary arrest and ICU admission rates. The meta-analyses were performed by computing the risk ratio (RR) using a random effects model. The random effects model provides more conservative estimates of treatment effects in the face of heterogeneity. The analysis was performed in subgroups of adults and paediatrics with further subgroups by study design (cluster randomized control trial (CRCT), interrupted time series and controlled before–after versus before–after studies with no contemporaneous controls and observational studies). Quantitative analysis was performed using an intention-to-treat analysis, and the RR and 95 % confidence intervals (CIs) were calculated [28]. We used the method proposed by Higgins et al. [29] to measure inconsistency between study results, reported as the I2 statistic as well as the 95 % CI. Publication bias was evaluated using the contour funnel plot asymmetry and the Harbord modification of the Egger test [30, 31]. We performed both analyses because graphical evaluation can be subjective. Factors other than publication bias can cause asymmetry in the funnel plot. These include factors such as study quality or true study heterogeneity. A meta-regression was undertaken to explore the association between treatment effect and study characteristics [32]. These predefined factors included the number of activations per 1000 admissions, the presence of a physician in the RRS team and the time period for implementation. The study by Hillman et al. [5] was a cluster randomized trial. The approach recommended by the Cochrane Collaboration to approximate the effective sample size is to divide by the design effect [20]. The design effect is calculated as:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \mathrm{Design}\ \mathrm{effect} = 1 + \left(M\hbox{--}\ 1\right) \times \mathrm{I}\mathrm{C}\mathrm{C} $$\end{document}Designeffect=1+M–1×ICC where M is the average cluster size and ICC is the intracluster correlation coefficient. The ICC compares the variance within clusters with the variance between clusters. Mathematically this is the between-cluster variability divided by the sum of the within-cluster and between-cluster variabilities. An ICC value of 0 gives a design effect of 1. This value indicates that individuals within a cluster have no similarity and there is no adjustment for sample size. An ICC of 1 indicates that all individuals within a cluster are identical and the sample size is the number of clusters. Larger cluster sizes are associated with smaller ICC values. In the study by Hillman et al. [5], the number of clusters was 23 and the ICC used was 0.02. The study by Priestley et al. [12] was described as a step wedge CRCT. This involves a sequential roll-out of and intervention to clusters over a number of time periods [33]. The order in which clusters receive the intervention is random. The measurement of data from all clusters and at each step is a key feature that distinguishes the step wedge design from a classic cluster RCT [33, 34]. This trial did not measure at each step and is actually reported more like a cluster RCT [34]. We therefore calculated the design effect according to the procedure recommended by the Cochrane Collaboration [20]. Sensitivity analyses were pre-specified. The treatment effect was reported using a cumulative meta-analysis method fixed-effects model and a graphical augmentation to the funnel plot to assess the impact of a new study on the existing meta-analysis. An influence analysis was performed in which the pooled estimates are computed omitting one study at a time. Additionally a sensitivity analysis was conducted using an ICC of 0.01 and of 0.05 to establish whether this qualitatively changed results. The details of this analysis are included in Additional file 1. The analysis was performed using STATA statistical software (version 12.0; Statacorp, College Station, TX, USA) and Revman (version 5.1; Copenhagen: The Nordic Cochrane Centre, The Cochrane Collaboration). RESULTS A total of 29 publications were included in the analysis (Fig. 1). No unpublished studies were obtained. Study characteristics The RRS studies had an effective sample size of 2,160,213 patients (1,107,492 in the intervention group and 1,108,380 in the control group) (Table 1). Nineteen studies (65.5 %) reported physicians as part of the RRS team for 24 hours per day and 7 days per week, two studies only had physician presence for office hours Monday to Friday, seven studies had no physician presence and one study did not report on the composition of the team. All of the studies have been published since 2000 and 13 studies have been published after 2008 (the end date for systematic review by Chan et al. [17]). Twenty-five studies were single centre. Twenty-one studies were conducted in academic hospitals, seven in community hospitals and one study used multiple sites that included both academic and community hospitals. The characteristics of the RRS intervention are described in Table 2. The number of RRS team activations per 1000 admissions was reported in 23 studies and varied substantially across studies. The mean and 95 % CI for the adult and paediatric activations per 1000 admissions were 16.3 (9.0–23.7) and 16.8 (6.0–27.6), respectively. About 33 % (95 % CI 23–43 %) of referrals were admitted to the ICU immediately after a RRS team consultation and 9.7 % (95 % CI 4.5–14.9 %) acquired a new designation of do not attempt resuscitation.Table 1Characteristics of included studiesIDAuthorCountryYearStudy designNumber and type of sites and populationType of team1Al-Qahtani et al. [40]Saudi Arabia2013Before/after without contemporaneous controls1/academic/adult1 ICU doctor1 ICU nurse1 ward nurse1 respiratory therapist2Baxter et al. [41]Canada2008Before/after without contemporaneous controls2/community/adult1 ICU doctor1 ICU nurse1 respiratory therapist3Beitler et al. [42]USA2011Before/after without contemporaneous controls1/academic/adultMedical doctorICU nurseRespiratory therapistPatient transporter4Bellomo et al. [22]Australia2004Before/after study without contemporaneous controls1/academic/adult1 ICU doctor1 Nurse5Bristow et al. [43]Australia2000Controlled before/after3/community/adult1 ICU doctor1 general medicine doctor1 nurse6Buist et al. [44]Australia2002Before/after without contemporaneous controls1/academic/adult2 doctors1 ICU nurse7Campello et al. [45]Portugal2009Before/after without contemporaneous controls1/community/adult1 ICU doctor1 nurse8Chan et al. [13]USA2008Before/after without contemporaneous controls1/academic/adult2 ICU nurses1 respiratory therapist9Dacey et al. [46]USA2007Before/after without contemporaneous controls1/community/adult1 ICU or hospitalist doctor1 ICU nurse1 respiratory therapist1 physician assistant10De Vita et al. [47]USA2004Before/after study without contemporaneous controls1/academic/adult1 ICU doctor1 anaesthetist2 physician2 ICU nurse1 ward nurse11Hayani et al. [48]Canada2011Before/after without contemporaneous controls1/academic/adult1 ICU doctor1 ICU nurse1 respiratory therapist12Hillman et al. [5]Australia2004Cluster RCT23/mixed/adultICU or ED doctorICU or ED nurse13Howell et al. [49]USA2012Interrupted time series1/academic/adultWard doctor2 nurses1 respiratory therapist14Jones et al. [50]Australia2005Before/after without contemporaneous controls1/academic/adultICU doctorICU nurse15Kenward et al. [51]UK2004Before/after study without contemporaneous controls1/community/adultNR16Konrad et al. [14]Sweden2010Before/after without contemporaneous controls1/academic/adult1 ICU doctor1 ward doctor1 ICU nurse1 ward nurse17Lim et al. [52]South Korea2011Before/after without contemporaneous controls1/academic/adult1 ICU doctor1 ICU nurse1 respiratory therapist18Priestley et al. [12]UK2004Step wedge cluster RCT1/community/adult2 nurses19Santamaria et al. [53]Australia2010Before/after without contemporaneous controls1/academic/adult1 ICU doctor1 general medicine doctor1 ICU nurse20Shah et al. [54]USA2011Before/after without contemporaneous controls2/academic/adult1 ICU nurse 1 respiratory therapist21Simmes et al. [55]The Netherlands2012Before/after without contemporaneous controls1/academic/adult1 ICU physician1 ICU nurse22Brilli et al. [56]USA2007Before/after without contemporaneous controls1/academic/paediatric1 ICU doctor1 nurse1 respiratory therapist23Hanson et al. [57]USA2010Interrupted time series1/academic/paediatric1 PICU doctor1 nurse1 respiratory therapist24Anwar ul Haque et al. [58]Pakistan2010Before/after without contemporaneous controls1/academic/paediatric1 ICU doctor25Hunt et al. [59]USA2008Before/after without contemporaneous controls1/academic/paediatric3 PICU doctors1 PICU nurse1 PICU respiratory therapist26Kotsakis et al. [60]Canada2011Prospective before/after without contemporaneous controls4/academic/paediatric1 doctor (PICU daytime ICU night-time)1 PICU nurse1 respiratory therapist27Sharek et al. [61]USA2007Before/after without contemporaneous controls1/academic/paediatric1 PICU doctor1 PICU or cardiac nurse1 respiratory therapist28Tibbals and Kinney [23]Australia2009Before/after without contemporaneous controls1/academic/paediatric1 ICU doctor1 ED doctor1 general medicine doctor1 ICU nurse29Zenker et al. [62]USA2007Before/after without contemporaneous controls1/academic/paediatric1 doctor1 PICU nurse1 respiratory therapist ED emergency department, NR not reported, PICU paediatric intensive care unit, RCT randomized controlled trialTable 2Characteristics of rapid response system implementation and interventionsIDAuthorCalls per 1000 admissionsDNAR (%)Control period/implementation period/intervention period (months)ICU disposition (%)Types of interventions (%)Study definition of cardiac arrest and mortality1Al-Qahtani et al. [40]18.29.324/0/3640.2Intubation 4Non-ICU cardiopulmonary arrests, hospital mortality, including patients with DNAR designationNIV 8IV fluids 48Diuretics 13Vasoactive infusions 52Baxter et al. [41]40.3824/12/1227Intubation 5All cases of arrest Hospital-wide deaths, patients with DNAR designation includedNIV 6IV fluids 32Diuretics 10Vasopressors 83Beitler et al. [42]10.811.236/0/3643.4NRNon-ICU cardiopulmonary arrests, hospital mortality, patients with DNAR designation included4Bellomo et al. [22]4.7104/14/418.2Intubation 3All cardiac arrests, hospital mortality, patients with DNAR designation includedNIV 9IV fluids 18Diuretics 11Vasopressors 55Bristow et al. [43]NRNRNR/NR/6NRNRAll cardiac arrests, hospital mortality, patients with DNAR designation included6Buist et al. [44]6.710.512/24/1210.5NRAll cardiac arrests, hospital mortality, patients with DNAR designation included7Campello et al. [45]7.8NR12/0/48NRNRAll cardiac arrests, hospital mortality, patients with DNAR designation included8Chan et al. [13]15.12.120/4/2041.2Intubation 7Hospital-wide cardiopulmonary arrest and mortality with DNAR designation includedNIV 11IV fluids 16Diuretics 7Vasopressors 19Dacey et al. [46]20.1104/1/1224Intubation 11All cardiac arrests, and hospital mortality with DNAR designation includedNIV 10IV fluids 3210De Vita et al. [47]25.8NR60/0/20NRHospital-wide cardiopulmonary arrest mortality not reported11Hayani et al. [48]NRNR60/0/36NRNRMortality at 100 days after transplant12Hillman et al. [5]8.782/4/630NRNon-ICU cardiopulmonary arrest Cardiac arrest and non-ICU mortality with DNAR designation excluded13Howell et al. [49]53NR22/6/3120NRHospital mortality with DNAR designation excluded14Jones et al. [50]25.2NR12/14/50NRNRAll cardiac arrests, and hospital mortality with DNAR designation included15Kenward et al. [51]532512/0/1220Intubation 23All cardiac arrests, and hospital mortality with DNAR designation includedIV fluids 2516Konrad et al. [14]2.52660/3/2427NRAll cardiac arrests, and hospital mortality DNAR designation included17Lim et al. [52]NRNR6/6/6NRNRAll cardiac arrests, and hospital mortality including DNAR designation18Priestley et al. [12]NRNR0/0/8NRNRCardiac arrest NR; hospital mortality including DNAR designation19Santamaria et al. [53]8.7NR30/3/18NRNRAll cardiac arrest and hospital mortality including DNAR designation included20Shah et al. [54]26.779/6/2750NRAll cardiac arrest and hospital mortality with DNAR designation included21Simmes et al. [55]56NR12/3/2553NRAll cardiac arrest and hospital mortality with DNAR designation excluded23Hanson et al. [57]NRNR24/10/1257NRNon-ICU cardiac arrest and hospital mortality with DNAR designation included24Anwar ul Haque et al. [58]21NR10/0/939Intubation 18All cardiac arrest and hospital mortality with DNAR designation included25Hunt et al. [59]11.9NR12/0/12NRNRNon-ICU cardiac arrest, hospital mortality NR26Kotsakis et al. [60]44.2NR24/9/2430NRAll cardiac arrest and hospital mortality with DNAR designation included27Sharek et al. [61]19.70.754/2/1957Intubation 0.7Non-ICU cardiac arrest and hospital mortality with DNAR designation includedIV fluids 16Vasopressors 428Tibballs and Kinney [23]5.1NR41/3/4847Intubation 8All cardiac arrest and hospital mortality with DNAR designation includedNIV 7IV fluids 23Vasopressors 429Zenker et al. [62]12.8NR23/0/1236Intubation 7All cardiac arrest and hospital mortality with DNAR designation includedNIV 6IV fluidsVasopressors 2 IV intravenous fluids, DNAR do not attempt resuscitation, NIV non-invasive ventilation, NR not reported Risk of bias within studies Two investigators (RM and IR) assessed study quality independently. The risk of bias for each study is presented in Tables 3, 4 and 5. The Newcastle Ottawa Quality Assessment Scale was used to evaluate the before–after studies without contemporaneous controls [35]. The interrupted time series, controlled before–after and cluster randomized studies were evaluated using the criteria recommended by the Cochrane Effective Practice and Organisation of Care Group [26]. This included recruitment bias, baseline imbalance, loss of clusters, incorrect analysis or selective outcome reporting.Table 3Risk of bias table for cluster randomized control trials and controlled before–after trialsStudyAllocation sequence generationAllocation concealmentBaseline comparabilityComplete outcome dataOutcome variables assessed blindlyProtection from contaminationSelective outcome reportingFree from other biasesBristow et al. [42]High riskHigh riskLow riskLow riskLow riskLow riskLow riskLow riskPriestley et al. [12]High riskHigh riskHigh riskHigh riskUnclearUnclearHigh riskLow riskHillman et al. [5]Low riskLow riskLow riskLow riskLow riskLow riskLow riskLow riskTable 4Risk of bias table using the Newcastle Ottawa Quality Assessment Scale for cohort studiesStudySelectionComparabilityOutcomeRepresentativeness of exposed cohortSelection of non-exposed cohortAscertainment of exposureOutcome of interest was not present at the start of the studyComparability of cohorts on the basis of design or analysisAssessment of outcomeWas follow-up long enough for the events to occur?Adequacy of follow-up cohortsAl-Qahtani et al. [40]*******Baxter et al. [41]*******Beitler et al. [42]*****Bellomo et al. [22]********Buist et al. [44]******Campello et al. [45]******Dacey et al. [46]******DeVita et al. [47]*****Hayani et al. [48]****Jones et al. [50]******Kenward et al. [51]*******Konrad et al. [14]********Lim et al. [52]******Santamaria et al. [53]********Shah et al. [54]********Simmes et al. [55]***Brilli et al. [56]********Anwar ul Haque et al. [58]****Hunt et al. [59]*******Kotsakis et al. [60]****Sharek et al. [61]********Tibballs and Kinney [23]******Zenker et al. [62]*****Table 5Risk of bias table for interrupted time series studiesStudyWas the intervention independent of other changes?Was the shape of the intervention effect pre-specified?Was the intervention unlikely to affect data collection?Was knowledge of the allocated interventions adequately prevented during the study?Were incomplete outcome data adequately addressed?Was the study free from selective outcome reporting?Was the study free from other risks of bias?Howell et al. [49]Low riskLow riskLow riskLow riskLow riskLow riskLow riskHanson et al. [57]Low riskUnclear riskLow riskLow riskLow riskLow riskLow risk In general terms, the number of stars denotes study quality. A study can earn one star for each component of ‘Selection’, ‘Outcome’ and ‘Comparability’. Representativeness is awarded a star if the cohort is truly or somewhat representative of the population of interest. A star is awarded for selection of the non-exposed cohort, if it is drawn from the same population as the exposed cohort. Exposure is satisfactorily ascertained if data were acquired from secure records. A maximum of two stars can be given for ‘Comparability’. Either the exposed or non-exposed were matched in design or confounders adjusted in analysis. A maximum of three stars can be given for ‘Outcome’. Assessment of outcome is awarded a star if the outcomes were assessed by independent or blind assessment, confirmation of records by reference to secure records or record linkage. The adequacy of the duration of follow-up should be awarded a star if it was long enough for the outcomes to occur. Completeness of follow-up was considered adequate if losses were not related to the exposure or the outcome and was sufficiently low to be unlikely to introduce bias. Syntheses of results Primary outcome Hospital mortality The implementation of RRS in the adult population has been associated with an overall reduction in hospital mortality (RR 0.87, 95 % CI 0.81–0.95, p <0.001) (Fig. 2). There was evidence of considerable heterogeneity (I2 = 86 %, p <0.001). The treatment effect in the cluster randomized trials, controlled before–after and interrupted time series studies was RR 0.91 (95 % CI 0.85–0.97) with less heterogeneity (I2 = 3 %). In the paediatric population, RRS also showed a reduction in mortality (RR 0.82, 95 % CI 0.76–0.89) with significant heterogeneity (I2 = 78 %) (Fig. 3). There was only one study in the cluster randomized control study, controlled before–after and interrupted time series subgroup so no subgroup analysis based on study design could be performed.Fig. 2Forest plot of the effect of rapid response system teams on hospital mortality in adult in-patients. Weights are calculated from random-effects analysis. CBA controlled before–after, CCO critical care outreach, CI confidence interval, ITS interrupted time series, RCT randomized controlled trialFig. 3Forest plot of the effect of rapid response system teams on hospital mortality in paediatric in-patients. Weights are calculated from random-effects analysis. CBA controlled before–after, CI confidence interval, ITS interrupted time series, RCT randomized controlled trial Secondary outcomes Cardiopulmonary arrests The implementation of RRS in the adult population has been associated with an overall reduction in cardiopulmonary arrests (RR 0.65, 95 % CI 0.61–0.70, p <0.001) with substantial heterogeneity (I2 = 70 %, p <0.001) (Figure S1 in Additional file 1). The treatment effect in the cluster randomized trials, controlled before–after and interrupted time series studies subgroup was RR 0.74 (95 % CI 0.56–0.98) with less heterogeneity (I2 = 0 %). In the paediatric population, RRS also showed a reduction in cardiopulmonary arrests (RR = 0.64, 95 % CI 0.55–0.74) with minimal heterogeneity (I2 = 7 %) (Figure S2 in Additional file 1). ICU admissions Only 10 of the adult studies reported the effect of RRS teams on the number ICU admissions. The implementation of RRS in the adult population has not been associated with a significant effect on the number of ICU admissions (RR 0.90, 95 % CI 0.70–1.16, p = 0.43). None of the paediatric studies reported the effect of RRS teams on the number of ICU admissions. Assessment of publication bias Publication bias refers to the phenomenon in which studies with less favourable results are less likely to be published than those with favourable results. Funnel plots appear asymmetric because of systematic suppression of studies. There are many factors other than publication bias that may explain funnel plot asymmetry, such as differential study quality or small study effects. A contour-enhanced funnel plot aims to disentangle these causes of funnel plot asymmetry. Generally, the level of statistical significance may drive publication bias, with studies that do not reach the perceived milestones of significance (e.g. p <0.05) less likely to be published. A contour-enhanced funnel plot overlays the contours of statistical significance on a funnel plot. This provides a novel method to assess whether the studies that exist are areas of statistical significance or whether there are areas where studies are missing that correspond to areas of low statistical significance. If studies are missing in areas of low statistical significance, then there may be publication bias. If studies are perceived to be missing, then publication bias is less likely. The contour-enhanced funnel plot was asymmetric but the perceived missing studies were in areas of high statistical significance, making publication bias a less likely cause of funnel asymmetry (Fig. 4).Fig. 4Contour-enhanced funnel plot. If studies appear to be missing in areas of low statistical significance, then it is possible that the asymmetry is due to publication bias. Conversely, if the area in which studies are perceived to be missing are of high statistical significance, then publication bias is a less likely cause of the funnel asymmetry The Harbord modification of the Egger test was used to assess funnel plot asymmetry. The estimated intercept was −0.207 with a standard error of 0.897 and a p value of 0.819. This result suggests little evidence for small study effects. Assessment of heterogeneity A meta-regression was performed to explore the heterogeneity between studies. The covariates used were the number of activations per 1000 admissions, the duration of implementation in months and the presence of a physician in the RRS team. The model was adjusted for multiplicity using the Monte-Carlo permutation test for meta-regression. There was no significant relationship between any of the covariates (activation dose (p = 0.112), implementation time (p = 0.999) or physician presence (p = 0.992)) and hospital mortality. The model accounted for 18 % of the observed heterogeneity. The residual 82 % heterogeneity was probably due to persisting between study effects. Sensitivity analysis Assumptions about the effects of clustering were explored using an ICC of 0 and of 0.01 (Figures S3 and S4 in Additional file 1). Neither of these assumptions changed the treatment effect and is included in the supplementary data. A sensitivity analysis was performed by systematically omitting individual studies to establish the influence on outcome (Figure S5 in Additional file 1). This showed that there was no study whose removal would materially change the pooled estimate of hospital mortality. A cumulative meta-analysis was performed in which the cumulative evidence at the time of each study is calculated (Figure S6 in Additional file 1). This analysis showed that the effect estimate has been consistent over time. A further analysis examined the potential impact a new study would have on the existing meta-analysis, providing an indication of the robustness of the results to the addition of new evidence. The graph shows that all studies lie in the region with a beneficial treatment effect, which dominates the graph (Figure S7 in Additional file 1). This result suggests that the meta-analysis is relatively robust to the addition of a single new trial. DISCUSSION In a systematic review and meta-analysis of 29 studies we found that a RRS team was associated with a reduction in hospital mortality in both adult and paediatric hospital populations. Our study reveals a striking degree of variation in how RRS teams were constituted, delivered and evaluated. The term ‘rapid response system’ may refer to rapid response teams, medical emergency teams or critical care outreach teams. Critical care outreach teams have the functionality of rapid response teams together with a surveillance function as well as ICU discharge follow-up. A number of findings merit further discussion. There was a high degree of between-study heterogeneity in the included studies. The analysis included studies with different methodologies and could account for this observation. The adult CRCTs, controlled before–after and interrupted time series studies showed minimal heterogeneity (I2 = 3 %), compared with observational and before–after studies without a contemporaneous control (I2 = 88 %). This may be due to a small number of studies in the former subgroup. This study did not find any dose–response relationship between the duration of the implementation phase, the presence of a physician on the team or the number of activations per 1000 and hospital mortality. These covariates were chosen because of pre-existing favourable reports [4]. Previous reports have suggested that a longer duration of implementation may lead to higher levels of support and engagement with the RRS [4, 36]. The optimal composition of the RRS team is uncertain. Two previous single-centre reports did not show the benefits of intensivist-led teams compared with registrar or resident-led teams [37, 38]. The majority of RRS interventions did not require the presence of a physician (fluids, oxygen and diuretics). Decisions around end-of-life planning may require physician involvement but would not necessarily manifest as changes in hospital mortality, although they could affect the number of cardiopulmonary resuscitation codes on the wards. It is possible that the presence of a physician in the team may have a differential effect in university hospitals compared with community hospitals, but there are insufficient data to establish this. An increase in RRS team activations per 1000 admissions had previously been associated with reduction in cardiac arrest [39]. We were unable to show a significant relationship between RRS team activation and hospital mortality. This may suggest that the mechanism by which RSS teams reduce mortality is not through reductions in cardiac arrest. Very sensitive calling criteria may overactivate the RSS team, causing system fatigue with no tangible benefit. A final caveat is that the interpretation of meta-regression should always be undertaken cautiously. Meta-regression has limitations: the small number of studies, correlation between covariates and unmeasured characteristics, differences in the relationships that occur at a patient level and that may not be detected at a study level, and the tendency of regression to the mean. The effect on the RRS team on the ICU workflow is important. The study was not able to show any effect on overall numbers of ICU admissions. This study has several limitations. The vast majority of studies were observational studies without a contemporaneous control. Whilst there are several guidelines for the reporting of these studies, valuable information was often missing. The subgroup analysis did not find any significant difference in treatment effect in the different study methodologies. The outcomes of studies were reported variably. Some studies reported all hospital mortality and others reported only non-DNAR designated hospital mortality We used all hospital mortality reported because this offers the most conservative estimate of treatment. The major strength of our study is that the treatment effect has been consistent over time, is not influenced by any single study, and is robust to assumptions about clustering and to a further study being conducted. CONCLUSION This study found that RRS teams associate with a reduction in hospital mortality and cardiac arrest. These findings did not show any significant publication bias. A sensitivity analysis showed that the study findings were robust to addition of a new study. We were unable to show any benefit from the presence of a physician on the RRS team, the duration of implementation or the number of activations. Whilst RRS teams are very much part of the landscape in many health systems, further work is needed to understand the specific factors that are likely to mitigate their effectiveness in given operational contexts. Key messages RRS teams are effective in reducing hospital mortality in both adult and paediatric in-patients.RRS teams also reduce hospital cardiac arrest.The vast majority of rapid response interventions do not require a physician and the presence of a physician was not associated with improved outcomes. References1.NealeGWoloshynowychMVincentCExploring the causes of adverse events in NHS hospital practiceJ R Soc Med.200194322301141870010.1177/014107680109400702PMC12815942.BuistMDJarmolowskiEBurtonPRBernardSAWaxmanBPAndersonJRecognising clinical instability in hospital patients before cardiac arrest or unplanned admission to intensive care. A pilot study in a tertiary-care hospitalMed J Aust19991712251045166710.5694/j.1326-5377.1999.tb123492.x3.FranklinCMathewJDeveloping strategies to prevent inhospital cardiac arrest: analyzing responses of physicians and nurses in the hours before the eventCrit Care Med.199422244710.1097/00003246-199402000-0001483066824.JonesDADeVitaMABellomoRRapid-response teamsN Engl J Med.20113651394610.1056/NEJMra0910926217519065.HillmanKChenJCretikosMBellomoRBrownDDoigGIntroduction of the medical emergency team (MET) system: a cluster-randomised controlled trialLancet.20053652091710.1016/S0140-6736(05)66733-5159644456.LaurilaJVPitkalaKHStrandbergTETilvisRSDetection and documentation of dementia and delirium in acute geriatric wardsGen Hosp Psychiatry.20042631510.1016/j.genhosppsych.2003.08.003147573007.BuistMBernardSNguyenTVMooreGAndersonJAssociation between clinically abnormal observations and subsequent in-hospital mortality: a prospective studyResuscitation.2004621374110.1016/j.resuscitation.2004.03.005152943988.SmithGBPrytherchDRSchmidtPFeatherstonePIKnightDClementsGHospital-wide physiological surveillance—a new approach to the early identification and management of the sick patientResuscitation.200671192810.1016/j.resuscitation.2006.03.008169454659.DowneyAWQuachJLHaaseMHaase-FielitzAJonesDBellomoRCharacteristics and outcomes of patients receiving a medical emergency team review for acute change in conscious state or arrhythmiasCrit Care Med.2008364778110.1097/01.CCM.0000300277.41113.461809153510.ChenJBellomoRFlabourisAHillmanKFinferSCentre MSIftS and Group ACT. The relationship between early emergency team calls and serious adverse eventsCrit Care Med2009371485310.1097/CCM.0b013e3181928ce31905062511.CretikosMChenJHillmanKBellomoRFinferSFlabourisAThe objective medical emergency team activation criteria: a case–control studyResuscitation.200773627210.1016/j.resuscitation.2006.08.0201724173212.PriestleyGWatsonWRashidianAMozleyCRussellDWilsonJIntroducing Critical Care Outreach: a ward-randomised trial of phased introduction in a general hospitalIntensive Care Med.200430139840410.1007/s00134-004-2268-71511203313.ChanPSKhalidALongmoreLSBergRAKosiborodMSpertusJAHospital-wide code rates and mortality before and after implementation of a rapid response teamJAMA.200830025061310.1001/jama.2008.7151905019414.KonradDJaderlingGBellMGranathFEkbomAMartlingCRReducing in-hospital cardiac arrests and hospital mortality by introducing a medical emergency teamIntensive Care Med.201036100610.1007/s00134-009-1634-x1976020615.The Joint Commission announces the 2009 National Patient Safety Goals and requirements. Jt Comm Perspect. 2008;28:1, 11–51879572116.BerwickDMCalkinsDRMcCannonCJHackbarthADThe 100,000 lives campaign: setting a goal and a deadline for improving health care qualityJAMA.2006295324710.1001/jama.295.3.3241641846917.ChanPSJainRNallmothuBKBergRASassonCRapid response teams: a systematic review and meta-analysisArch Intern Med.2010170182610.1001/archinternmed.2009.4242006519518.WintersBDWeaverSJPfohERYangTPhamJCDySMRapid-response systems as a patient safety strategy: a systematic reviewAnn Intern Med.20131584172510.7326/0003-4819-158-5-201303051-0000923460099PMC469599919.LiberatiAAltmanDGTetzlaffJMulrowCGotzschePCIoannidisJPThe PRISMA statement for reporting systematic reviews and meta-analyses of studies that evaluate healthcare interventions: explanation and elaborationBMJ.2009339b270010.1136/bmj.b270019622552PMC271467220.Higgins JPT, Green S, editors. Cochrane handbook for systematic reviews of interventions. Version 5.1.0 (updated March 2011). Oxford, UK: The Cochrane Collaboration;201121.BellomoRGoldsmithDUchinoSBuckmasterJHartGKOpdamHA prospective before-and-after trial of a medical emergency teamMed J Aust.200317928371296490910.5694/j.1326-5377.2003.tb05548.x22.BellomoRGoldsmithDUchinoSBuckmasterJHartGOpdamHProspective controlled trial of effect of medical emergency team on postoperative morbidity and mortality ratesCrit Care Med.2004329162110.1097/01.CCM.0000119428.02968.9E1507137823.TibballsJKinneySReduction of hospital mortality and of preventable cardiac arrest and death on introduction of a pediatric medical emergency teamPediatr Crit Care Med.2009103061210.1097/PCC.0b013e318198b02c1930780624.TibballsJKinneySDukeTOakleyEHennessyMReduction of paediatric in-patient cardiac arrest and death with a medical emergency team: preliminary resultsArch Dis Child.20059011485210.1136/adc.2004.06940116243869PMC172017625.StangACritical evaluation of the Newcastle-Ottawa scale for the assessment of the quality of nonrandomized studies in meta-analysesEur J Epidemiol.201025603510.1007/s10654-010-9491-z2065237026.Effective Practice and Organisation of Care (EPOC). EPOC Resources for review authors. Oslo: Norwegian Knowledge Centre for the Health Services; 2015. Available at: http://epoc.cochrane.org/epocspecific-resources-review-authors.27.DevitaMABellomoRHillmanKKellumJRotondiATeresDFindings of the first consensus conference on medical emergency teamsCrit Care Med.20063424637810.1097/01.CCM.0000235743.38172.6E1687803328.LiberatiAAltmanDGTetzlaffJMulrowCGotzschePCIoannidisJPThe PRISMA statement for reporting systematic reviews and meta-analyses of studies that evaluate health care interventions: explanation and elaborationPLoS Med.2009610.1371/journal.pmed.100010019621070PMC270701029.HigginsJPThompsonSGDeeksJJAltmanDGMeasuring inconsistency in meta-analysesBMJ.20033275576010.1136/bmj.327.7414.55712958120PMC19285930.HarbordRMEggerMSterneJAA modified test for small-study effects in meta-analyses of controlled trials with binary endpointsStat Med.20062534435710.1002/sim.23801634503831.PalmerTMPetersJLSuttonAJMorenoGGContour-enhanced funnel plots for meta-analysisStata J.200882425432.HarbordRMHigginsJPTMeta-regression in StataStata J.2008849351933.WoertmanWde HoopEMoerbeekMZuidemaSUGerritsenDLTeerenstraSStepped wedge designs could reduce the required sample size in cluster randomized trialsJ Clin Epidemiol.201366752810.1016/j.jclinepi.2013.01.0092352355134.de HoopEWoertmanWTeerenstraSThe stepped wedge cluster randomized trial always requires fewer clusters but not always fewer measurements, that is, participants than a parallel cluster randomized trial in a cross-sectional designIn reply. J Clin Epidemiol.201366142810.1016/j.jclinepi.2013.07.0082402161235.HartlingLMilneAHammMPVandermeerBAnsariMTsertsvadzeATesting the Newcastle Ottawa Scale showed low reliability between individual reviewersJ Clin Epidemiol.2013669829310.1016/j.jclinepi.2013.03.0032368384836.DeVitaMABellomoRHillmanKIntroduction to the rapid response systems seriesJt Comm J Qual Patient Saf.200632359601688412110.1016/s1553-7250(06)32046-637.MorrisDSSchweickertWHolenaDHandzelRSimsCPascualJLDifferences in outcomes between ICU attending and senior resident physician led medical emergency team responsesResuscitation.2012831434710.1016/j.resuscitation.2012.07.01722841611PMC403784138.KarvellasCJde SouzaIAGibneyRTBagshawSMAssociation between implementation of an intensivist-led medical emergency team and mortalityBMJ Qual Saf.201221152910.1136/bmjqs-2011-0003932219054039.JonesDBellomoRDeVitaMAEffectiveness of the medical emergency team: the importance of doseCrit Care.20091331310.1186/cc799619825203PMC278434040.Al-QahtaniSAl-DorziHMTamimHMHussainSFongLTaherSImpact of an intensivist-led multidisciplinary extended rapid response team on hospital-wide cardiopulmonary arrests and mortalityCrit Care Med.2013415061710.1097/CCM.0b013e318271440b2326361841.BaxterADCardinalPHooperJPatelRMedical emergency teams at The Ottawa Hospital: the first two yearsCan J Anaesth.2008552233110.1007/BF030215061837896742.BeitlerJRLinkNBailsDBHurdleKChongDHReduction in hospital-wide mortality after implementation of a rapid response team: a long-term cohort studyCrit Care.201115R26910.1186/cc1054722085785PMC338866643.BristowPJHillmanKMCheyTDaffurnKJacquesTCNormanSLRates of in-hospital arrests, deaths and intensive care admissions: the effect of a medical emergency teamMed J Aust.2000173236401113034610.5694/j.1326-5377.2000.tb125627.x44.BuistMDMooreGEBernardSAWaxmanBPAndersonJNNguyenTVEffects of a medical emergency team on reduction of incidence of and mortality from unexpected cardiac arrests in hospital: preliminary studyBMJ.20023243879010.1136/bmj.324.7334.38711850367PMC6553045.CampelloGGranjaCCarvalhoFDiasCAzevedoLFCosta-PereiraAImmediate and long-term impact of medical emergency teams on cardiac arrest prevalence and mortality: a plea for periodic basic life-support training programsCrit Care Med.20093730546110.1097/CCM.0b013e3181b021831977075446.DaceyMJMirzaERWilcoxVDohertyMMelloJBoyerAThe effect of a rapid response team on major clinical outcome measures in a community hospitalCrit Care Med.20073520768210.1097/01.CCM.0000281518.17482.EE1785582147.DeVitaMABraithwaiteRSMahidharaRStuartSForaidaMSimmonsRLUse of medical emergency team responses to reduce hospital cardiopulmonary arrestsQual Saf Health Care.200413251410.1136/qshc.2003.00658515289626PMC174386548.HayaniOAl-BeihanyAZarychanskiRChouAKharabaABaxterAImpact of critical care outreach on hematopoietic stem cell transplant recipients: a cohort studyBone Marrow Transplant.20114611384410.1038/bmt.2010.2482097246549.HowellMDNgoLFolcarelliPYangJMottleyLMarcantonioERSustained effectiveness of a primary-team-based rapid response systemCrit Care Med.2012402562810.1097/CCM.0b013e318259007b22732285PMC363807950.JonesDOpdamHEgiMGoldsmithDBatesSGutteridgeGLong-term effect of a medical emergency team on mortality in a teaching hospitalResuscitation.2007742354110.1016/j.resuscitation.2006.12.0071736791351.KenwardGCastleNHodgettsTShaikhLEvaluation of a medical emergency team one year after implementationResuscitation.2004612576310.1016/j.resuscitation.2004.01.0211517270352.LimSYParkSYParkHKKimMParkHYLeeBEarly impact of medical emergency team implementation in a country with limited medical resources: a before-and-after studyJ Crit Care.201126373810.1016/j.jcrc.2010.08.0192103652753.SantamariaJTobinAHolmesJChanging cardiac arrest and hospital mortality rates through a medical emergency team takes time and constant reviewCrit Care Med.2010384455010.1097/CCM.0b013e3181cb0ff12002934154.ShahSKCardenasVJJrKuoYFSharmaGRapid response team in an academic institution: does it make a difference?Chest.20111391361710.1378/chest.10-05562086461855.SimmesFMSchoonhovenLMintjesJFikkersBGvan der HoevenJGIncidence of cardiac arrests and unexpected deaths in surgical patients before and after implementation of a rapid response systemAnnals Intensive Care.201222010.1186/2110-5820-2-20PMC34251342271630856.BrilliRJGibsonRLuriaJWWheelerTAShawJLinamMImplementation of a medical emergency team in a large pediatric teaching hospital prevents respiratory and cardiopulmonary arrests outside the intensive care unitPediatr Crit Care Med.200782364610.1097/01.PCC.0000262947.72442.EA1741711357.HansonCCRandolphGDEricksonJAMayerCMBruckelJTHarrisBDA reduction in cardiac arrests and duration of clinical instability after implementation of a paediatric rapid response systemPostgrad Med J.201086314810.1136/qshc.2007.0260542044822658.HaqueA uSaleemAFZaidiSHaiderSRExperience of pediatric rapid response team in a tertiary care hospital in PakistanIndian J Pediatr201077273610.1007/s12098-010-0032-22017783059.HuntEAZimmerKPRinkeMLShilkofskiNAMatlinCGargerCTransition from a traditional code team to a medical emergency team and categorization of cardiopulmonary arrests in a children's centerArch Pediatr Adolesc Med.20081621172210.1001/archpediatrics.2007.331825023460.KotsakisALobosATParshuramCGillelandJGaiteiroRMohseni-BodHImplementation of a multicenter rapid response system in pediatric academic hospitals is effectivePediatrics.201112872810.1542/peds.2010-07562169011361.SharekPJParastLMLeongKCoombsJEarnestKSullivanJEffect of a rapid response team on hospital-wide mortality and code rates outside the ICU in a children’s hospitalJAMA.200729822677410.1001/jama.298.19.22671802983062.ZenkerPSchlesingerAHauckMSpencerSHellmichTFinkelsteinMImplementation and impact of a rapid response team in a children’s hospitalJt Comm J Qual Patient Saf.200733418251771114410.1016/s1553-7250(07)33048-1