Introduction

In 2001, Rivers et al reported on early resuscitation based on central venous pressure and oxygenation in a single urban US site. This approach, termed “Early Goal Directed Therapy” (EGDT),1 decreased mortality from septic shock. Despite subsequent non-randomised studies showing benefit2 3 or tendencies towards benefit,4 5 feasibility6 7 and incorporation into the international Surviving Sepsis Campaign (SSC) 6-hour bundle guidelines,8 implementation of EDGT is not yet widespread.9 10 Postulated explanations include the difficulty of recognising patients with early severe sepsis in busy emergency departments, the complex nature of the intervention, the uncertain need for each component, resource implications11 and doubts about the validity of the original trial. Effectiveness may be context-dependent, and some believe that equivalent or better outcomes can be achieved with care tailored to the patient.12

As a preliminary to sepsis resuscitation trials in the USA, Australia and the UK, we sought to assess clinicians' beliefs on the early resuscitation of patients with severe sepsis. We deliberately did not mention EGDT or the SSC, as we wished to investigate intended practice rather than knowledge or opinion of these guidelines. We also sought to identify barriers to the implementation of the EGDT component of the SSC 6-hour bundle guidelines. Our study was designed to be purely observational, making no hypothesis as to the rate or desirability of guideline adoption; in so doing, we hoped to design a questionnaire with as little bias as possible.

Methods

Study design

We designed a scenario-based self-administered questionnaire survey asking about management of a 65-year-old previously healthy male patient with suspected pneumonia, signs of shock and organ dysfunction. We asked: “How would you manage this patient presenting to your emergency department today?” to uncover resource limitations, institutional policies and personal attitudes that could influence behaviour. In the case presentation we made no mention of the SSC guidelines, but clinical questions (see table 1 in online supplement) corresponded to the 6-hour resuscitation bundle recommendations. We did not explore other recommendations for the care of sepsis and avoided using the term “sepsis” as a 2004 survey of 1058 doctors found fewer than 17% agreed on any one definition.13

Respondents were required to answer all questions. If earlier choices logically would have precluded asking later questions, the respondent was told to imagine having taken over care from another doctor who had selected the required option. We also asked respondents to identify barriers to the full implementation of the 6-hour bundle in their institution. These questions centred on specific examples of the described barriers to adherence to clinical practice guidelines.14

The survey was pretested among members of the Departments of Critical Care Medicine and Emergency Medicine at the University of Pittsburgh and the Australian and New Zealand Intensive Care Society Clinical Trials Group. Pretesting identified questions that did not provide all possible options; these questions were modified and the option of free-text responses was added. The University of Pittsburgh Institutional Review Board and each participating specialist organisation approved the survey design.

Selection of participants

We aimed to collect responses from doctors undertaking independent practice in emergency or intensive care medicine and, in the UK, in acute internal medicine. Acute internal medicine is an evolving UK specialty that takes primary responsibility for a range of medical inpatients during the acute phase of illness, usually up to 48–72 h. We used membership lists from the specialist societies shown in figure 1. Responses were received from doctors in the Republic of Ireland who were members of UK societies. Rather than exclude these, we grouped them with those from the UK, as clinical training in these two countries is broadly comparable and mutually recognised. In the USA we targeted the Pennsylvania Chapter of the American College of Emergency Physicians for logistic reasons and the large pool of potential respondents.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Survey respondents with exclusions. PaACEP, Pennsylvania Chapter of the American College of Emergency Physicians; SCCM, Society of Critical Care Medicine; ACEM, Australasian College for Emergency Medicine; JFICM, Joint Faculty of Intensive Care Medicine of the Australian and New Zealand College of Anaesthetists and the Royal Australasian College of Physicians; BAEM, British Association for Emergency Medicine; ICS, Intensive Care Society; SAM, Society for Acute Medicine. *Total invitations sent=total number of emails sent − number returned from invalid addresses. **The number of total responses is known only by country as respondents answered country-specific surveys. Note that the “eligible completed response rate” has not been calculated as it is not possible to know how many invitations were sent to ineligible doctors.

We excluded from the primary analysis responses from doctors in the USA and Australasia without specialty certification. Specialist training is longer in the UK and Ireland, with senior trainees often assuming independent responsibility for patient care. We therefore included UK/Irish doctors who had passed their final specialty examination. A proportion of respondents started but did not complete the survey. As the demographic questions confirming eligibility were at the end of the survey, these incomplete responses were excluded from the primary analysis. Supplementary analyses compared non-certified and incomplete responses with those in the primary analysis. We also excluded the small number who did not practise in a target region or clinical field (figure 1).

Data collection, processing and analysis

Survey invitations were emailed between 18 December 2006 and 18 July 2007, with a follow-up reminder sent 2 weeks later. A proprietary internet-based survey instrument was used. We amalgamated the responses to alternative forms of equivalent questions, leaving a total of 12 questions for analysis. Using Stata Version 9 (StataCorp), we employed multinomial logistic regression models or binomial regression models modified using the “firthlogit” method (as appropriate) to assess the independent effects of experience, hospital type, specialty and region of practice on the responses to each question. All of these dichotomous predictors were forced into the models. The “firthlogit” method is a penalised form of maximum likelihood estimation that reduces bias in logistic regression where the conventional approach yields multiple instances of “perfect prediction”.15 Incomplete responses and responses from non-certified doctors were compared with those in the primary analysis using χ² tests. To protect against spurious conclusions due to multiple testing, α was set at 0.01.

Results

We distributed 11 882 invitations, eliciting 2137 full and 324 partial responses (response rate 21%). Following application of the predefined exclusion criteria (figure 1), the primary analysis was performed on the full responses from 1692 doctors.

Experience (43% with ≤10 years practice in their specialty) and type of hospital (53% in university hospitals) were evenly spread (see table 2 in online supplement). Detailed replies to each clinical question are shown in figures 1–10 in the online supplement. The proportions of respondents who selected responses consistent with the SSC guidelines are summarised in figures 2 and 3.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2

Initial management of severe sepsis: percentage of responses consistent with the Surviving Sepsis Campaign (SSC) guidelines. A 65-year-old man weighing 80 kg who was previously well presents with presumed pneumonia (heart rate 100, blood pressure 125/50 (mean arterial pressure 75), respiratory rate 22, SpO₂ 95% on room air, temperature 38.7°C). The patient is awake and alert. (1) Which tests would you order to help determine illness severity? (a variety of laboratory tests were presented). (2) Does a lactate concentration of 4 mmol/l influence your management plan? (yes/depends on physical examination/no). Now consider a different patient, again a 65-year-old previously healthy man weighing 80 kg with presumed pneumonia. This patient is hypotensive (heart rate 120, blood pressure 80/35 (mean arterial pressure 50), respiratory rate 22, Spo₂ 95% on room air, temperature 38.7°C. The patient is awake and alert. (3) How would you first treat the low blood pressure? (no treatment/start a vasopressor/infuse various amounts of fluid). (4) What monitoring devices would you order? (a variety of monitoring devices were presented). The percentage of respondents who gave answers consistent with the SSC guidelines for each question is shown. The complete set of responses is contained in the online supplement. *Significant effect compared with US ICU on selecting this response in logistic regression model (p<0.001).

• Download figure
• Open in new tab
• Download powerpoint

Figure 3

Subsequent management of severe sepsis: percentage of responses consistent with the Surviving Sepsis Campaign (SSC) guidelines. (1) What would you order next to treat the low blood pressure? (fluid/vasopressor/nothing else). (2) If you need to use a vasopressor in this patient, which would you choose? (a variety of vasopressors/inotropes were presented). (3) How would you decide how much more fluid to give? (no more intravenous fluid/target central venous pressure (CVP) 8–12/other CVP targets were listed/estimate a specific fluid volume/titrate to physical examination). (4) After adequate fluid resuscitation, vital signs are blood pressure 125/50 (mean arterial pressure 75), heart rate 100 on a moderate rate (0.1 μg/kg/min) norepinephrine infusion. Cardiac output is not monitored. The haemoglobin is 8.5 g/dl and the ScvO₂ is 50%. What would you do next? (nothing/transfuse packed red cells/increase norepinephrine/add an inotrope/insert a cardiac output monitor/decide only after physical examination). (5) Let's say the haemoglobin is 10.5 g/dl, blood pressure 125/50 (mean arterial pressure 75) after fluid+moderate rate (0.1 μg/kg/min) norepinephrine and the ScvO₂ is 50%. Would you start an inotrope (eg, epinephrine, dobutamine, dopexamine, dopamine)? (no/only if indicated by a monitor/only if indicated by clinical examination/yes). (6) Let's say that, after appropriate fluid, vasopressor, inotropic and blood product support, the patient has improved, with a ScvO₂ of 60%, blood pressure 100/40 (mean arterial pressure 60), pulse 90, central venous pressure 11. The patient is alert and there are minimal respiratory secretions. However, the respiratory rate is 25 and the SpO₂ is 99% on 6 l/min oxygen. What change in treatment would you order now? (no change/reduce Fio₂/increase Fio₂/use NIPPV/intubate). The percentage of respondents who gave answers consistent with the SSC guidelines for each question is shown. The complete set of responses is contained in the online supplement. *Significant effect (compared with ICU physicians in the USA) on selecting this response in logistic regression model (p<0.001).

Overall guideline compliance

Guideline compliance was low: only two respondents (0.1%) gave answers consistent with the entire EGDT protocol and 32 (2%) implemented all but intubation for a persistently low central venous oxygen saturation (ScvO₂), a component of EGDT but not the SSC 6-hour bundle. Only 21% (n=356) complied with all but intubation plus any one of the other protocol elements.

Compliance with individual components of the guidelines

Lactate measurement and interpretation

Overall, less than half (46.5%) of the respondents measured blood lactate concentration. When presented with a lactate level of 4 mmol/l, only 41.9% indicated this would definitely influence their management. There were marked differences among the groups: 29.5% of US EM physicians would check lactate levels compared with 78.8% of UK EM physicians. UK EM physicians were most likely (73.7%) to be influenced by a lactate level of 4 mmol/l. ANZ EM physicians were least (26.5%) influenced; 57.3% felt the importance of the result would be determined by their findings on physical examination.

Fluid resuscitation and vasopressor for hypotension

Less than one-third (27.4%) of respondents gave the recommended 1–1.5 l (12–20 ml/kg) fluid bolus as initial treatment for hypotension. All groups tended to give less fluid than recommended, but this was particularly so for intensivists in ANZ (61.3%) and the UK (69.3%) and acute internal medicine physicians in the UK (68.6%). When the patient remained hypotensive following an initial 1.5 l fluid bolus, most respondents (67.7%) chose to give more fluid as recommended. Overall, 44.4% chose the recommended fluid goal of central venous pressure 8–12 mm Hg. Most respondents chose norepinephrine when optimal fluid resuscitation (by their own definition) had not normalised the blood pressure, although dopamine was a more common choice among US EM physicians (55.5%).

Invasive monitoring

Most respondents would insert arterial (61.1%) and central venous (71.5%) catheters. However, US EM physicians reported a substantially different practice: only 19.7% inserted an arterial catheter and 43.9% a central venous line. Many US EM physicians (17.3%) and intensivists (22.0%) who inserted a central venous catheter said they would use a continuous monitor of ScvO₂, whereas only 1.7% in ANZ and 5.9% in the UK did so. More respondents chose to use intermittent ScvO₂ measurement such that, overall, 51.5% chose to measure ScvO₂ or mixed venous oxygen (SvO₂) by any means, although again there was a greater than twofold variation between groups (figure 2).

Optimisation of ScvO₂ using blood transfusion, inotropes and mechanical ventilation

Once fluid resuscitation and blood pressure had been optimised, 51.5% did not report transfusing packed red blood cells for an ScvO₂ <70% and haemoglobin <10 g/dl. Transfusion practice was highly variable, with 14.8% of ANZ intensivists and >65% of EM physicians in all three regions following the SSC guideline. Few (39.0%) said they would commence an inotrope specifically for an ScvO₂ <70%, relying instead on clinical examination (31.7%) or a monitor of cardiac output (21.9%). Only 4.5% of respondents indicated they would intubate the patient for an ScvO₂ <70% (as in EGDT) once all the SSC guidelines had been implemented.

Barriers to guideline implementation

Respondents identified numerous reasons for not fully employing the 6-hour resuscitation bundle (figure 4). The principal concern of emergency physicians (40%) was time pressure, while the same proportion of intensivists (40%) felt the threshold for blood transfusion was too low. ANZ intensivists in particular were sceptical of the evidence supporting the protocol; 62% felt there was insufficient evidence and 55% believed that equally good or better results were achieved by care tailored to the patient. One-fifth of US and ANZ EM physicians reported no knowledge of Rivers' EGDT study on which the SSC guidelines are based.

• Download figure
• Open in new tab
• Download powerpoint

Figure 4

Responses to the question: “If you do not currently aim to implement ALL of the Surviving Sepsis Campaign (Rivers' “Early Goal-Directed Therapy” (EGDT)) protocol1 for ALL patients with sepsis, which of the following best explain why not?” *Significant effect (compared with ICU physicians in the USA) on selecting this response in multinomial logistic regression model (p<0.001).

Influence of doctor and hospital characteristics

Years of experience and hospital type influenced the responses to some questions. Experienced doctors used dopamine more frequently (p<0.001). Doctors outside university hospitals were less likely to insert an arterial line and were more likely to attach no significance to a lactate concentration of 4 mmol/l and to be guided by physical examination rather than a central venous pressure target. They were more likely to be unaware of the study by Rivers et al and more likely to report that in their hospital there would often be no-one present who would be capable of inserting the required monitoring devices.

Incomplete responses and responses from doctors without specialty certification

Two hundred and ninety-nine doctors in the regions of interest started but did not complete the survey (12.1%). None gave answers consistent with all of the EGDT protocol. The only differences from those with complete responses were less use of lactate as a screening tool among US and UK partial responders. Three hundred and twenty-one doctors (13.0%) were excluded from the primary analysis because they did not hold specialty certification. One of these, a US doctor practising intensive care, did comply with all aspects of the protocol. Non-board certified doctors in the USA generally reported higher rates of guideline compliance, but this was significant only for the monitoring of ScvO₂. There were no differences in compliance between trainees and specialists in the UK or ANZ.

Discussion

In the largest most geographically comprehensive and most multidisciplinary survey of critical care management to date, almost no doctor who answered the survey stated that he/she would implement the entire SSC 6-hour resuscitation bundle, for reasons that varied by region and specialty. This suggests that any intervention aiming to alter sepsis management would be more efficient if targeted to particular audiences and if focused on the particular barriers identified in each group.

We can only speculate on why doctors in different regions and specialities had such different practice intentions, as we collected no data on this subject and there is little relevant published literature. Emergency medicine doctors in the USA and ANZ were less likely to check lactate concentrations on presentation, perhaps because they have less ready access to lactate point-of-care testing than EM colleagues in the UK who are more than twice as likely to check lactate and attribute significance to a raised level. Intensivists in the UK and ANZ were particularly conservative with initial fluid boluses, perhaps because higher nurse-patient ratios in their environments led them to favour a more titrated approach. In the absence of evidence, tradition probably guides decisions, potentially explaining the marked preference for dopamine over norepinephrine among EM physicians, particularly in the USA and among older doctors. US EM physicians were very reluctant to insert invasive monitoring devices, as were doctors in non-university hospitals. In free-text comments many reported insufficient trained staff to allow this. Intensivists were especially reluctant to transfuse red cells to achieve a target haemoglobin of >10 g/dl, which may reflect the fact that much of the primary evidence concerning the adverse effects of blood transfusion has appeared outside the emergency medicine literature, or perhaps that patients are present in ICUs long enough for complications such as transfusion-related acute lung injury to be appreciated.

The main limitation of our survey is its low response rate, which raises the possibility that our samples do not represent the opinions of the populations from which they are drawn. The specialty societies could not provide demographic data of their membership, so we cannot test whether our samples were representative in terms of age, sex, experience or hospital type. Previously published email surveys of doctors' attitudes and practices have response rates similar to ours.16 17 We aimed to survey opinions of the entire community of doctors primarily responsible for early sepsis care, and not all of the relevant specialty societies could provide the information required to conduct more targeted sampling. In accordance with accepted survey methodology,18 we asked demographic questions last. As we needed this information to categorise respondents into specialty and country groups, it was necessary to exclude partly completed surveys from the primary analysis. If we had asked these questions first, all responses could have been counted. Even if the survey respondents are biased samples of the target populations, their opinions represent delivery of care by a very large number of doctors. Overall guideline compliance and compliance with individual components of the guidelines in many regions is strikingly low. It is unlikely that the sample is so unrepresentative that true compliance is markedly higher and closer to that targeted by the SSC.

We confined our primary analysis to board-certified (or equivalent) doctors who were members of their national specialty societies. Many patients with sepsis will not be treated by such a doctor.19 20 Other doctors may be more or less likely to follow published guidelines; however, of the 13.0% excluded from our primary analysis because they did not hold specialty certification, only those in the USA had different responses to board-certified doctors and for a singular element. Again, based on these data, the effect of certification seems modest at best.

We can never be certain that respondents would act as they describe. However, we used well-established features of good survey design, including asking about intended practice in a specific situation rather than global feelings.18 Importantly, we mentioned neither EGDT nor the SSC guidelines until the final question regarding barriers to guideline implementation. In so doing, we hoped to avoid respondents either simply attempting to demonstrate their knowledge of the recommendations or their attitude towards them. We used validation questions to ensure the survey was identifying known trends, avoided “double-barrelled” questions and incorporated implausible options to detect spurious responses. Expected patterns, such as the preference for dopamine among US emergency physicians, are consistent with anecdotal experience. Comparison with observed practice is not necessarily appropriate, as the Hawthorne effect may alter the behaviour of doctors who know they are being observed. Case vignettes have been found to accurately represent the behaviour of doctors, being superior to data collection from the clinical chart.21 Other studies have found that questionnaires overestimate adherence to guidelines.22 If our study overestimates guideline compliance, the low use of the SSC resuscitation recommendations becomes more striking.

Smaller studies have examined attitudes to and implementation of the SSC 6-hour bundle. In 2004, a survey of emergency physicians in 30 US academic medical centres23 demonstrated a 7% reported frequency of EGDT use. A postal survey of English emergency departments in 200610 found only 18.8% of 117 departments had a written guideline conforming to the EGDT protocol. Of 53 patients with sepsis in Spain in 2004,24 only two had blood lactate measured, 41% were incorrectly classified as having no organ system dysfunction and only 46.6% with severe sepsis or septic shock had the recommended fluid management. No patient had measurements of central venous pressure or oxygen saturation. A scenario-based questionnaire of Canadian intensivists9 reported only 52.5% would use inotropes for a low ScvO₂ “often or always”, and only 7.6% would transfuse red cells if the haemoglobin concentration was <10 g/dl. Unlike our survey, EM physicians were not included and participants were asked to report their intended behaviour on the assumption that there were no resource limitations. The reasons for not using EGDT were hypothesised but not quantified.

Despite SSC guideline dissemination, our survey demonstrates lack of support from, or awareness of, potential users. We agree with the recent analysis25 that multicentre trials are an important step towards bringing the most effective care to patients with sepsis.