Introduction

The way in which the intensive care unit (ICU) is organized and managed potentially affects quality and efficiency in critical care [112]. Yet despite an extensive literature addressing ICU organization and management, our understanding of the impact of these factors on outcomes remains limited, in part as a result of heterogeneity across organizational factors and the complex interplay among them. Additionally, acute care delivery may vary significantly within and among countries, with differing patient populations and local care practices affecting the ICU structure, process, and outcome relationship [4, 6, 1315].

Delivering affordable and high-quality critical care is particularly challenging for emerging countries. These countries account for more than 50 % of the world’s population and face massive demographic and socioeconomic transformation in the coming decades, with increased urbanization, economic development, and population growth [16]. In parallel, the demand for critical care is likely to increase, with severe implications in terms of resource allocation and costs. Yet even at present, patients’ outcomes remain comparatively worse as compared to those in developed nations [4, 17, 18]. Optimizing ICU organization is a potential target to improve patient care and reduce mortality in these countries. However, comprehensive information on ICU organization and its relation to outcomes is mostly restricted to developed countries [1, 3, 4, 712] and may not fully translate to developing countries.

To address these knowledge gaps, we sought to describe the organizational characteristics of a large sample of Brazilian ICUs and to investigate their impact on the hospital mortality and resource use.

Patients and methods

Design and setting

We performed a multicenter retrospective cohort study of critical care organization and outcomes in 78 ICUs at 51 hospitals in 11 Brazilian states. The complete list of investigators and centers appears in the Electronic Supplementary Material (ESM). The study was coordinated by the Department of Critical Care at the D’Or Institute for Research and Education (IDOR), Rio de Janeiro, Brazil, and was endorsed by the Brazilian Research in Intensive Care Network (BRICNet), an independent research network for performing investigator-initiated multicenter studies in critical care in Brazil. The Local Ethics Committee at the IDOR (Parecer: 334.835) and the Brazilian National Ethics Committee (CAAE: 19687113.8.1001.5249) approved the study and the need for informed consent was waived. The full description of patients and methods is given in the ESM.

Selection of centers, data collection, and definitions

Participating centers

We restricted the study to ICUs registered in the BRICNet database and known to use the Epimed Monitor System® (Epimed Solutions®, Rio de Janeiro, Brazil), a commercial cloud-based registry for quality improvement, performance evaluation, and benchmarking purposes. A total of 117 ICUs met these criteria and were invited to participate in the study. Of these, 88 ICUs agreed to participate. We excluded ICUs that initiated Epimed after July 2012 (n = 2) and ICUs in which more than 10 % of patients were missing core data (n = 8), leaving 78 ICUs from 51 hospitals in the final cohort (Fig. 1).

Fig. 1
figure 1

Study flowchart. ICU intensive care unit

Within each participating ICU, the ICU director and/or chief nurse completed a survey about hospital and ICU organizational, structural, and process characteristics (see ESM). Domains for the survey were based on prior studies demonstrating potential structure–outcome links in critical care [3, 57, 911], and included the following: hospital and ICU type and bed capacity, presence of residence/fellowship in critical care or training programs certified by the Brazilian Association of Intensive Care (Associação Brasileira de Medicina Intensiva, AMIB), ICU staffing patterns (excluding residents and trainees), presence of regular multidisciplinary clinical rounds and daily checklists, clinical pathways and protocols, regular debriefing and administrative multidisciplinary meetings, and family visiting policies. Detailed definitions of the organizational characteristics are given in the ESM.

In order to guarantee data accuracy, validity, and completeness, we piloted the survey among three participating centers and provided respondents with explicit definitions of the survey domains. We ensured the reliability of all data by interviewing medical and/or nurse ICU directors from every participating center on site or by phone.

Patients

We included all consecutive patients aged at least 16 years old admitted to the participating ICUs during 2013. Readmissions and patients with missing core data [age, location before ICU admission, main ICU admission diagnosis, the simplified acute physiology score (SAPS 3), ICU and hospital length of stay (LOS), and vital status at hospital discharge] were excluded (Fig. 1). We obtained de-identified patient data from the Epimed Monitor System®. Briefly, data were prospectively entered in a structured electronic case report form, most typically by a trained case manager. Key data elements included demographics, comorbidities based on the Charlson comorbidity index [19], functional status before hospital admission, location before ICU admission, SAPS 3 score and the sequential organ failure (SOFA) score, ICU admission diagnosis, use of ICU support, ICU and hospital LOS, and destination at hospital discharge.

Outcomes

The primary outcome of interest was in-hospital mortality at the patient level. The secondary outcome was efficient resource use [6]. We evaluated outcomes and resource use for each ICU by estimating the standardized mortality rates (SMR) and standardized resource use (SRU) according to the SAPS 3, as proposed by Rothen et al. [6]. The SRU estimates the average observed to expected ratio of resources (based on ICU LOS) used per surviving patient in a specific ICU adjusted for the severity of illness. On the basis of median SMR and median SRU, we assigned each ICU to one of four groups: “most efficient” (all units whose SMR and SRU were below the median SMR and SRU); “least efficient” (units with both SMR and SRU above the median); “overachieving” (low SMR and high SRU) and “underachieving” (high SMR and low SRU) (eFig. 1a–ll).

Data processing and statistical analysis

We screened data for missing information, implausible and outlying values, logical errors, and insufficient details. In cases of inconsistent or implausible data, we contacted local investigators to provide the requested information. There was no missing information regarding hospital and ICU characteristics. Because missingness for patients’ characteristics was minimal, we performed single imputation using the reference or “normal” category [20].

We investigated the association between organizational factors and hospital mortality adjusting for patients’ characteristics using multilevel multivariable logistic regression. A two-level model was fit with patient-level fixed effects at the first level and ICU-level fixed effects at the second level, as well as an ICU-specific random effect. Patient level variables of interest included age, gender, hospital LOS before ICU admission, diagnostic category, previous functional capacity, the Charlson comorbidity index, the use of mechanical ventilation (MV) on day 1, SOFA and SAPS 3 scores. ICU level variables of interest included ICU type [medical-surgical vs. specialty (i.e. ICUs dedicated to only specific groups of patients)], presence of training programs in critical care, ICU admission volume (number) during 2013, staffing patterns, organizational and process characteristics. We selected variables for the multivariable model using forward and backward stepwise regression. We considered variables for the model if they were associated with outcome with a P value of less than 0.20 on univariate analysis, and removed variables from the model if they had a P value of greater than 0.10 on multivariate analysis. We performed subgroup analyses stratifying patients according to the type of admission (medical vs. surgical) and the SAPS 3 tertiles. We used the Akaike information criterion, a measure of the relative predictive ability of a statistical model for a given set of data, to choose among the alternative models. Two-tailed P values less than 0.05 were considered statistically significant. We conducted all statistical analyses in R (http://www.r-project.org) and SPSS 21 (IBM Corp., Armonk, NY).

Results

Characteristics of participating centers and studied population

The final sample included 78 ICUs at 51 hospitals (Fig. 1). The number of ICUs varied largely among hospitals: 37 had only one ICU; nine had two ICUs, three had three ICUs, and two hospitals had seven ICUs. ICU and hospital characteristics are reported in Tables 1 and 2, and eTables 1 and 2 of the ESM. Participating ICUs were mostly medical/surgical (n = 62, 79 %) and located in private hospitals (n = 72, 93 %). Half (n = 40, 51 %) had training programs in critical care. The average physician/bed ratio was 0.15 (0.11–0.19). Board-certified physicians and nurses were present 24/7 in 16 (21 %) and 9 (12 %) ICUs. Few ICUs had fully dedicated clinical pharmacists (n = 18, 23 %), nutritionists (n = 28, 36 %), and psychologists (n = 9, 12 %). Multidisciplinary clinical rounds occurred on a regular basis in 67 (86 %) and daily checklists were applied in 36 (46 %) ICUs. In general, checklists were applied by varying healthcare providers and were used to both monitor and prompt the adherence to best practices. Median number of clinical protocols per ICU was 7 (25 %–75 % IQR, 5–9) (Fig. 2). Protocols were in use for the following domains: ventilator-associated pneumonia prevention (n = 71, 91 %), central-line associated bloodstream infection prevention (n = 71, 91 %), sepsis (n = 62, 79 %), liberation from MV (n = 58, 74 %), lung-protective ventilation in acute lung injury (n = 51, 65 %), acute coronary syndromes (n = 51, 65 %), sedation in MV patients (n = 50, 64 %), early mobilization (n = 44, 56 %), cerebrovascular accident management (n = 36, 42 %), and therapeutic hypothermia in cardiac arrest patients (n = 34, 44 %) (Fig. 2). There were several differences among ICUs according to the type of hospital (eTable 2). Of note, ICUs located at both private for-profit and philanthropic hospitals more often had 24/7 intensivists, clinical rounds, checklists, protocols, and managerial meetings.

Table 1 ICU characteristics, staffing patterns, and comparisons between medical-surgical and specialty ICUs
Table 2 Organizational characteristics, clinical protocols and processes of care, and comparisons between medical-surgical and specialty ICUs
Fig. 2
figure 2

Implementation of clinical protocols in the participating ICUs (n = 78). CPA cardiopulmonary arrest, CVA cerebrovascular accident, ACS acute coronary syndromes, MV mechanical ventilation, CLABSI central-line associated bloodstream infection, VAP ventilator-associated pneumonia

A total of 71,002 admissions occurred in the participating ICUs during 2013, of which 59,693 met inclusion criteria (Fig. 1). Of included admissions, there were 39,863 (66.8 %) medical, 16,652 (27.9 %) scheduled surgical, and 3178 (5.3 %) emergency surgical. Comorbidities were present in 32,940 (55.2 %) patients, with the most frequent being systemic arterial hypertension (50.1 %), diabetes mellitus (23.1 %), and cancer (17.0 %). Overall in-hospital mortality was 14.4 % (8581/59,693). The vast majority of the 51,112 surviving patients were discharged home (92.8 %); very few were transferred to another hospital (1.5 %) or hospice care (0.7 %). The main patient characteristics are given in Table 3.

Table 3 Main patient characteristics and outcomes, and comparisons between medical-surgical and specialty ICUs

Organizational, structural, and process factors associated with hospital mortality

In the ESM, eTables 3 and 4 depict the results of univariate analyses of patients’ and centers’ characteristics associated with hospital mortality. Patient-level factors associated with worse outcomes included older age, male gender, longer hospital LOS prior to ICU admission, compromised previous chronic health status, the use of MV on day 1, and higher Charlson comorbidity, SAPS 3, and SOFA scores. ICU characteristics associated with lower hospital mortality included higher nurse/bed ratios, the presence of physiotherapists and clinical pharmacists exclusively dedicated to the ICUs, higher numbers of implemented protocols, the occurrence of multidisciplinary clinical rounds, daily checklists, and regular debriefing and managerial meetings.

In the multivariable analysis, the number of fully implemented clinical protocols was the only organizational characteristic associated with hospital mortality [odds ratio (OR) for each additional protocol = 0.944 (95 % confidence interval (CI) 0.904–0.987), P = 0.011] (Fig. 3 and eTable 5). In a second model evaluating all patients, we observed that the availability of jointly managed clinical protocols [OR = 0.231 (95 % CI 0.083–0.645), P = 0.005] was also independently associated with lower mortality (Fig. 3). Similar results were observed in the subgroup analyses, with significant reductions in mortality (Fig. 3 and eTable 6). Finally, we forced into the final model the following covariates at center level: physiotherapists and pharmacists fully dedicated to the ICU, daily checklists, and regular debriefing and administrative meetings. None of them were statistically significantly associated with mortality.

Fig. 3
figure 3

Association of organizational characteristics and hospital mortality: results from the multilevel multivariable analyses. Upper panel final model studying all patients, including the number of protocols. Middle panel model studying all patients, including jointly managed protocols. Lower panel adjusted effects of protocols in subgroup analyses (see also eTables 5 and 6). Estimates were adjusted for SAPS 3 score, need for mechanical ventilation, previous chronic health status, admission diagnostic category, ICU type, intensivists 24/7 in the ICU, average graduate nurse/bed ratio, regular clinical rounds, and annual ICU admission volume. ICU intensive care unit

Organizational, structural, and process factors associated with efficient resource use

Median estimated SMR and SRU were 0.97 (0.72–1.15) and 1.06 (0.89–1.37). There were 28 (36 %) “most efficient”, 28 (36 %) “least efficient”, 11 (14 %) “overachieving”, and 11 (14 %) “underachieving” ICUs. Comparing key organizational factors between “most efficient” with “lowest efficient ICUs, we found that “most efficient” ICUs were usually located in private hospitals, with step-down units and training programs in critical care (eTable 7). We also observed a significant trend toward efficient resource use as the number of implemented clinical protocols increased (eTable 7 and eFig. 1k). The presence of daily checklists also tended to be associated with efficient resource use [OR = 2.89 (0.95–8.72), P = 0.057]. In addition, we observed a trend toward highest efficiency as the graduate nurse/bed ratio increased [OR = 2.88 (0.76–10.99), for ratios between 0.17 and 0.25; and OR = 4.40 (1.04–18.60), for a ratios >0.25, P = 0.116] and when board-certified nurses were available 24/7 in the ICU [OR = 5.87 (0.64–53.93), P = 0.084] (eTable 7). SMR and SRU for the different participating ICUs according to different organizational, structure, and process characteristics are given in eFig. 1a–ll.

Discussion

In this study, we report detailed information on structure, organization, and process of care for a large sample of Brazilian ICUs. We found that ICU organization as well as standardized resource use and mortality rates varied substantially among the participating ICUs. Adoption of clinical protocols was associated with both improved patient survival and more efficient resource use. Other structural variables thought to be associated with mortality and resource use in some settings, such as 24/7 coverage by intensivists and regular multidisciplinary rounds, were not associated with mortality and resource use in this sample.

Although many studies evaluating ICU organization and process of care are available in the literature, the understanding of their effect on outcomes and resource use is limited by the complex interplay among them. Moreover, data from emerging countries remain scarce [18, 2124]. The present study adds to this literature by examining the relationship between ICU organization, patient outcomes, and resource use in a large sample of Brazilian ICUs. Recent studies demonstrate that there is significant variation in the capacity and outcomes in critical care around the globe [13, 14]. An increasing challenge in emerging countries is to provide access and high-quality and affordable care for the large urban populations [14, 25].

A higher number of protocols was associated with improved patient outcomes and more efficient resource use—a finding that was significant for both medical and surgical patients and across all severity of illness strata. This finding supports that the implementation of clinical protocols may be a potentially valuable strategy to achieve high-quality ICU outcomes, at least in some ICUs [12, 26]. Previous experiences on implementation of sepsis management strategies and sedation protocols carried out in Asian countries and Brazil were cost-effective and associated with better outcomes and less use of resources (as evaluated by ICU LOS and MV duration) [18, 21, 27]. In addition, our results suggest that collaborative multidisciplinary work among ICU care providers impacts favorably on the patients’ outcomes, since hospital mortality was lower in ICUs where protocols were jointly managed by different care providers. Conversely, in a recent study performed in the USA, protocols were not associated with outcomes [26]. A better understanding of which organizational and structural factors influence the successful implementation of protocols in ICUs requires further investigation.

In contrast to prior work, we did not observe a significant impact of physician or nurse staffing patterns on patients’ outcomes. In the early 2000s, a meta-analysis reported that high-intensity intensivist staffing (mandatory intensivist consultation or closed ICU) was associated with both reduced ICU and hospital mortality rates and LOS [11]. However, more recently, four multicenter studies investigated the association between the presence of 24-h intensivists and patients’ outcomes with different results, suggesting that this relationship seems less relevant that previously reported [2, 3, 10, 12]. On the other hand, the number of critical care nurses may play an important role in clinical outcomes of severely ill patients [2, 3, 28]. However, this was not the case for the present study. We can hypothesize that Brazilian ICUs have developed systems to make patient outcomes robust to variation in nurse staffing levels. Alternatively, our results might also be explained in part because graduate-nurse bed ratios higher than 1:1.5 were present in only two ICUs. Despite the introduction in almost all ICUs, regular multidisciplinary rounds were not associated with mortality and resource use, corroborating the findings of two recent studies carried out in the USA [3, 12].

We found that ICU admission volume was not independently associated with mortality, which contrasts early work [7] but is in agreement with contemporary studies [2, 3, 29, 30]. One of the potential explanatory mechanisms in volume–outcome associations is that organizational factors associated with outcomes are common at high-volume centers [31]. Therefore, our results are not unexpected, as we have controlled for these confounders in multivariable analyses. Additionally, some investigators hypothesized that even in scenarios with wide variation in admission volumes, the use of evidence-based practices and implementation of quality improvement strategies may be of help to allow low-volume centers to achieve good outcomes [30].

Our study has several limitations. First, we did not audit the implementation of protocols. The reported presence of a protocol does not necessarily imply that it was successfully implemented. Moreover, we cannot guarantee that protocols were similar in all units. Second, although we have evaluated a large number of Brazilian ICUs, as we used a convenience sample, our data may not be representative of the entire nation. The average number of ICU beds in Brazil is comparable to Europe (13 vs. 11.5 per 100,000 people), but wide variation within the country exists in terms of ICU beds and provision of healthcare [32, 33]. Third, we cannot rule out the possibility that the study was underpowered to test the association between some covariates at center level (e.g., nurse to bed ratios) and the dependent variable, nor that our results can be explained by unmeasured confounders. Fourth, the evaluation of SRU is beyond the original scope of SAPS 3. However, it has been used to assess variability in resource use between ICUs [6]. Fifth, as with most of the previous studies using patients’ data from already existing standardized databases [710, 29, 30], we were able to only evaluate hospital mortality. When evaluating ICU performance, assessments based on hospital mortality can be subject to biases related to discharge practices [34]. Nonetheless, the proportion of patients discharged to other hospitals and hospice care facilities in our study was relatively low. Sixth, we studied a population with an overall relatively low severity of illness and some caution is needed in the generalization of our results. Finally, we only assessed resource use in the ICU. Although we acknowledge that organization and process of care outside the ICU can potentially affect the overall resource use, such information was not collected.

In conclusion, specific organizational and process factors, including the implementation of protocols, were associated with patient outcomes and efficient resource use. These observations can assist in policies and interventions to bridge the current quality gap in the delivery of critical care in emerging countries.