Collection Mémoires et thèses électroniques
AccueilÀ proposNous joindre

Evaluation of Etomidate’s Effect on Cortisol Secretion in Intubated Traumatic Brain Injury Victims: A Prospective Cohort Study

Table des matières

Etomidate is one of the most frequently used anesthetic induction agents for intubating head trauma patients, although its possible adverse effects on adrenal function are still debated. Marketed in 1972 in Europe, etomidate was largely used initially because of its advantageous cardiovascular, respiratory and neuroprotective effects. However in 1983, Ledingham and Watt [39] showed that a continuous etomidate infusion for more than 5 days increased mortality. Mortality increased from 25% to 44% in the trauma unit where it was used. This increase in mortality was attributed later to adrenal suppression induced by the use of etomidate. Nevertheless, many other studies have shown that etomidate’s adrenal suppression is temporary, lasting between 4 to 12 hours, and does not cause any clinically significant problems when used in healthy elective surgical patients [40, 42, 54, 60, 75].

In 1999, Absalom et al. [55] studied the effect of etomidate in 35 critically ill patients. This small randomized controlled clinical trial showed that adrenal suppression lasted up to 24 hours after a single bolus dose was used to anaesthetize patients undergoing urgent surgical procedures. This population did not include any traumatic brain injury patients. In addition, no clinically relevant adverse effect of this adrenal suppression was identified.

Schenarts et al. (2001) [56] also conducted a randomized control trial in 31 critically ill patients that compared etomidate to midazolam for intubating patients in the emergency department (ED). They showed, once again, that adrenal suppression lasted between 4 to 12 hours without any clinically significant adverse outcomes from this suppression.

Hoen et al. (2002) [92] presented their results on the impact of relative adrenal insufficiency (RAI) in a cohort of hemorrhagic shock patients in which 47% presented RAI even after adequate resuscitation. RAI was defined as a delta cortisol inferior to 248.4 nmol/l (9mcg/dl) after a 250 mcg ACTH stimulation test. The small sample size of this study precludes any definite conclusion as to the effect of etomidate on mortality, but there was a trend to increased mortality in the group with RAI and a significant increase in vasopressor use in the etomidate group.

Cohan et al. [99] recently showed that approximately 50% of patients with moderate or severe traumatic brain injury (TBI) had at least transient adrenal insufficiency (AI). The definition of AI used in this study was based on two consecutive cortisol measurements under 414 nmol/l (15 mcg/dl) or one single measurement under 138 nmol/l (5 mcg/dl). Unfortunately, this definition has not been validated in a prospective cohort study that would have assessed the relationship between this specific definition for AI and mortality. These authors showed that lower cortisol levels were associated with lower blood pressure and higher vasopressor use and they recommended that consideration should be given to monitoring cortisol levels in intubated TBI patients. They also showed that young age, greater injury severity, early ischemic insults, and the use of etomidate and metabolic suppressive agents, like propofol and pentobarbital, are associated with their definition of AI. A randomized controlled trial studying the effect of hydrocortisone replacement in patients with adrenal insufficiency defined by their methods is currently being conducted [99]. Unfortunately, this study might add to the confusion surrounding the diagnosis in AI and its clinical impact. Using Cohan et al.’s definition of AI [99], etomidate’s effect on adrenal gland production of cortisol lasts up to 10 hours on average. Surprisingly, they also show that in all their subjects, AI appeared on average 2.4 days after intubation, which is not consistent with the pharmacokinetics of etomidate. Etomidate actually has been shown to have a time-dependent inhibition on the synthesis of cortisol [40, 41, 55-57]. Defining AI this way also groups patients who have primary, secondary and tertiary AI together and does not isolate the effect etomidate might have on RAI and mortality. In addition, Hamrahian et al. [22] have shown that using total cortisol to diagnose AI in this way leads to many false positive diagnoses of AI.

Recently, many authors have suggested that etomidate should not be used for critically ill patients, especially for those in septic shock [84, 86, 88, 157]. Indeed, Annane [84] has recommended that etomidate should not be used for intubating critically ill patients on the basis of a subgroup analysis from his study published in 2002 that supports the theory that RAI has deleterious effects in septic shock patients. Treating this RAI with physiologic doses of hydrocortisone improved survival and seemed to reverse all the negative effects of etomidate on steroid synthesis. In this study [85], 68 of the 72 subjects (94%) exposed to etomidate developed RAI. Those subjects having received etomidate without corticosteroid supplementation had a 21% greater absolute risk of death compared with subjects who had received etomidate with corticosteroid supplementation (p=0.03) [82, 85]. Although the absolute number of subjects in this study was small, these results have prompted many other investigators to study the relationship between etomidate and RAI in the setting of critically ill patients. A recent study has indeed confirmed Annane’s conclusions linking etomidate to an increase in the rate of RAI [81]. Unfortunately, this study did not include a large group of brain trauma victims. Recent calls to abandon the use of etomidate due to its suppression of adrenal function have raised many questions about the advantage of using etomidate over other induction agents during intubation of the critically ill patient [91].

The theoretical model underlying this study is described in Figure 1. The graphic concept underlying this figure was adapted from a similar one in Cooper et al. [7]. Three different clinical scenarios are depicted: the activity of the hypothalamo-pituitary-adrenal axis under normal conditions (Panel A), during the appropriate response to stress (Panel B) and during the inappropriate response to traumatic brain injury (Panel C).

CRH: corticotropin releasing-hormone; ACTH: corticotropin; CBG: corticosteroid-binding globulin; a plus sign indicates stimulatory effect; a minus sign indicates an inhibitory effect

Specifically, this study was trying to isolate the role etomidate has in inducing an inappropriate adrenal response (Panel C) to the stress of traumatic brain injury. This inappropriate adrenal response has many possible causes. RAI induced by etomidate is one of these causes. RAI is actually a form of primary adrenal insufficiency because it is caused by a primary failure of the adrenal gland to produce cortisol appropriate to the level of stress. Other causes of RAI in the context of trauma are cytokine production, hemorrhagic shock, traumatic adrenal hemorrhages and, rarely, infection. In TBI, other causes of inappropriate adrenal response are secondary and tertiary adrenal insufficiency caused by corticosteroid use, and trauma to the hypothalamus and to the pituitary gland. To isolate etomidate’s role in this inappropriate response to TBI, ACTH tests were performed in a cohort of TBI victims. A comparison of the adrenal responses of subjects having received etomidate to the subjects having received other induction agents was then performed.

A sample of patients was formed based on the following inclusion and exclusion criteria: patients admitted to EJH’s ICU and intubated directly in the ED; patients intubated in the ED of another referring center and then transferred to EJH; isolated head injury or multiple organ injuries with a concomitant head injury; patients with moderate or severe TBI intubated with or without the use of etomidate. A moderate TBI was defined by a Glasgow Coma Scale (GCS) score of 9 to 12 or any trauma patient with evidence of brain injury on CT scan as read at admission by the radiologist on call (with any of the following signs: intra or extra parenchymal bleeding, petechiae, cerebral contusions, diffuse axonal injury, basilar skull fracture, depressed skull fracture, traumatic sub-arachnoid hemorrhage, compression of the basilar cisterns, deviation of the midline over 5 mm and traumatic hygromes). A severe TBI victim was defined as a patient having evidence of TBI (see above CT scan criteria) and a GCS score between 3 and 8. Subjects had to be aged 16 years and over. They were excluded if they presented an allergy to Cortrosyn, were known to have pre-existing adrenal insufficiency, had received systemic steroids in the last 6 months, had received any steroids during the first 24 hours of the study, had surgery done to their pituitary stalk, presented with cerebral death criteria in the first 24 hours of the study, presented with septic shock at the moment of intubation, were taking ketoconazole, were pregnant or were seropositive for HIV.

As this was an observational cohort study, etomidate and other induction agents were given according to availability and emergency physician preferences. Every patient arriving at EJH for trauma was screened for inclusion in this study. After consent was obtained from the patients’ family members, participants underwent three ACTH stimulation tests at three different times: 24, 48 and 168 hours. These tests were required to respect a particular schedule so as to ensure comparability. This schedule was followed for the first 24 and 48 hours. Tests that were supposed to be done at 168 hours were permitted to vary by a margin of +/- 24 hours since some of the subjects were not in the ICU anymore and nurse staffing problems were easier to manage with a flexible schedule. Serum cortisol levels were determined at baseline before every stimulation test and then 30 and 60 minutes after administering 250 mcg of ACTH. ACTH was administered after a powder form was reconstituted with a diluent and then injected as a bolus in any available intravenous line. The registered nurse (RN) administering the ACTH had to fill out a special laboratory requisition form on which she had to specify the time of administration and the time the blood samples were collected. Specific blood sample tubes were only used for the purpose of this study. Blood samples were sent immediately to the biochemistry laboratory and were centrifuged and frozen. Samples were analyzed as much as possible in batches so as to decrease any inter-assay variability. Physicians caring for the subjects included in this study did not receive the results of these tests. As a consequence, when a treating physician requested that the test results be available for the care of the patient, other samples had to be drawn at the same time. Physicians in the ICU caring for these subjects were not blinded to the agent used in the ED.

Results for cortisol assays were collected by the biochemist (LN) who was blinded to the induction agent used by the emergency physician. The presence of RAI was determined by measuring the difference between the maximal increase in cortisol at either 30 or 60 minutes compared to baseline. If this difference (delta cortisol) was inferior to 248.4 nmol/l (9 mcg/dl) then the subject was deemed to have RAI. Baseline demographic data that were collected prospectively for all patients screened for inclusion by a clinical member of the research team (PA, FL, AR, DM) were: age, gender, GCS, agent used for intubation and reason for exclusion (if applicable). Other baseline characteristics (Injury Severity Score(ISS), New Injury Severity Score (NISS), transfer status, injury mechanism, presence of ethanol in patient’s blood, drug intoxication) and endpoints concerning mortality and morbidity (rates of pneumonia, rates of severe sepsis, length of stay, FIM, orientation at discharge) were collected retrospectively once the whole study was finished using the Quebec Trauma Registry (QTR), which was compiled by medical archive specialists blinded to the agent used. This data collection was done at hospital discharge, after death or whenever a patient was transferred to another center. Some variables were not available in this database: initial vital signs recorded at triage, baseline comorbidities, APACHE II score, total amount of resuscitation fluid and rate of clinically significant AI. This information was extracted from medical charts with a structured data collection form by a research nurse, a research assistant and PA. Absolute blinding from the induction agent used was not possible at this stage. When data were missing, no assumptions were made and data were left missing. When data from referring centers were missing, the PI obtained the information from the medical archives of the referring center. All compiled data were entered into an Excel spreadsheet and then into a master database using the SAS software for statistical analysis. The distribution of variables was verified and any missing or outlying data were checked as to insure the quality of data collection.

The primary outcome measure of this study was the cumulative incidence of RAI (at 24, 48 and 168 h). It was defined as the proportion of subjects presenting RAI, as determined by the peak cortisol increase between the baseline cortisol level and the highest cortisol level at 30 min or 60 min after the ACTH stimulation test inferior to 248.4 nmol/l (9mcg/dl). The variable delta cortisol (defined as the difference between the maximal increase in cortisol and the baseline cortisol level) was also considered as a continuous variable, to allow a thorough assessment of any relationship etomidate could have on cortisol synthesis after an ACTH stimulation test.

Secondary outcomes were defined a priori. The cumulative incidence of clinically relevant AI was defined as the proportion of subjects treated for RAI or any other form of adrenal insufficiency (secondary or tertiary) without having laboratory confirmation of the diagnosis. Subjects had to be treated with hydrocortisone for the diagnosis to be deemed clinically significant. In addition, the final diagnosis of clinically significant AI did not have to be supported by an independent ACTH stimulation test. The objective of this secondary endpoint was to assess how often clinicians feel AI could be causing a clinically relevant impact on their patient. This outcome is obviously fraught with great limitations and subject to a potential Hawthorne effect [158] and will have to be interpreted with great caution.

Other clinically relevant and patient-centered outcomes were assessed as well. In addition to the data collected for the subjects having given consent to undergo ACTH stimulation tests, data from the QTR were used to determine the clinically relevant outcomes for all the subjects who were eligible for this study. Mortality was assessed 24 hours, 48 hours, 7 days and 28 days after intubation. As well, total hospital mortality was analyzed after each patient had either died, had been discharged from EJH or had been transferred to another institution. This dichotomous variable was extracted from the QTR. Length of stay in the ICU was also extracted from the QTR. This variable, measured in days, was defined as the length of hospital stay in the ICU from the moment of admission until discharge from the unit or death. Orientation of the patient at discharge from the hospital was also analyzed. This categorical variable was determined using the QTR, and it had the following values: return home without help, return home with homecare support, transfer to a long term care facility and transfer to a rehabilitation center. The functional status of subjects at discharge was also analyzed. This information was available in the QTR since every patient is systematically assessed by an occupational therapist with the FIM before discharge. This instrument has been validated and found useful to assess the evolution of the functional status of trauma patients [144, 145].

Other secondary outcomes were also studied. The QTR documents all complications and coexisting medical diagnoses if the discharging physician includes this diagnosis in his discharge summary. We extracted data from this registry to assess the rates of pneumonia and severe sepsis.

Since this study was not randomized, many different potential confounding variables had to be accounted for and used to adjust the estimates of the strength of the associations examined. A priori, we defined three variables in that regard: age, gender and the ISS score. After analyzing our data at the end of our study and in light of new research that had been published since our study started, we also considered other confounding variables such as respiratory rate, GCS, mean arterial pressure (MAP), transfer status, transfer time, underlying chronic illness (chronic hypertension, coronary heart disease, chronic respiratory illness, chronic liver disease), surgery in first 24 hours of admission to EJH, mechanism of injury, presence of ethanol in patient’s blood, illicit drug intoxication, NISS [147, 148], baseline cortisol level and albumin level. It was hypothesized that these variables could influence the risk of developing RAI without being involved in the causal pathway between etomidate and the outcomes.

Since our main hypothesis was that etomidate induces RAI, we wanted to test the hypothesis that the effect lasted more than 24 hours and to determine the importance of etomidate on the cumulative incidence of RAI. To do so, we had to consider the potential confounding variables and to adjust for the actual confounders. Baseline demographic variables were compared using the chi-square test when appropriate. When the normal approximation was not holding, Fisher’s exact test was used. All continuous data were tested for normality with the Shapiro-Wilk test. If the data were normally distributed, a Student t-test was used to compare the two groups. All non-normal data were compared using the Mann-Whitney test. Once the subjects were described and the baseline characteristics of the exposure groups were compared, we proceeded to analyze the crude association between exposure to etomidate and the cumulative incidence of RAI at 24, 48 and 168 hours in bivariate analyses. Fisher’s exact test was used to compare the two groups on the cumulative incidence of RAI. We also compared in bivariate analyses the continuous variable delta cortisol, to assess any effect etomidate could have on adrenal reserve at 24, 48 and 168 hours since by considering these data dichotomously, important information was lost. Multiple logistic regression analyses were then performed to analyze etomidate’s effect on RAI (dichotomized) while taking confounding effects into account. Multiple linear regression models were performed with delta cortisol as the dependent variable to assess its relation with etomidate while adjusting for different confounding variables. All the previously stated potential confounding variables were assessed one by one in these analyses. Variables that changed the regression coefficient of exposure to etomidate by 10% or more were kept in the final models. To test the robustness of these models, an analysis of the outlying residuals was performed. The level of significance retained for all analyses was 0.05.

The other dependent variables (mortality at 28 days, total hospital mortality, incidence of clinically diagnosed AI, length of ICU stay, FIM, rate of pneumonia and rate of severe sepsis) were also analyzed in the same fashion to assess any relationship they may have had with the use of etomidate. To increase the statistical power of this study, these analyses were performed on all eligible subjects, including those who had declined participation in the ACTH stimulation testing. Since information on these variables was accessible through the QTR and did not involve any input from the subjects, permission to use this information was requested and obtained from the IRB. Logistic regression was performed to obtain adjusted estimates of the strength of the association between etomidate use and the dichotomous outcome measures. Linear regression analyses were performed to assess the strength of association between etomidate and continuous outcome measures. For the FIM, a Rasch analysis [152] was performed to permit the statistical analysis of this variable. All continuous variables (delta cortisol, length of stay and FIM) were adjusted for confounding variables using an analysis of covariance. A sample size assessment for a future, larger, randomized controlled study was also done using the results of this study.

In order to verify the presence of a selection bias influencing the outcomes of our study, we performed certain sensitivity analyses to see if the conclusions of our study would change.

One potential source of bias was a referral or selection bias. Since EJH is a tertiary neurosurgical reference center, it is possible that only the most severe TBI patients were referred. This referral bias would be important only for centers not having sent us all their intubated TBI patients. In the Province of Quebec, all intubated TBI patients are supposed to be transferred automatically to a tertiary neurosurgical reference center even if a neurosurgeon is present at the referring hospital. It is possible that specific centers would not have complied with this rule all the time and retained certain less severe TBI patients. This possibility would have been most likely in centers were a neurosurgeon was present. For certain specific cases, the neurosurgeon could feel that the severity of the TBI would not have warranted a transfer to the tertiary referral center. This would have potentially biased the inclusion of patients from that center since only the patients having the most severe TBI may have been referred to the study hospital. A sensitivity analysis excluding the subjects referred from these centers was performed to assess the impact of this potential bias. The results of this sensitivity analysis will be presented in the results section (Tables 11 and 12, section 3.3).

Since this cohort study was not able to include all potential subjects for the ACTH stimulation test, a sensitivity analysis was performed to determine the effect of the potential selection bias on the outcomes of these ACTH tests if these subjects had been included. The impact of this selection bias was assessed on the primary outcomes: the risk of developing RAI and the delta cortisol level at 24 hours. For the first outcome (risk of RAI), we assumed two different scenarios. The first scenario assumed that all subjects having received etomidate that were not included for ACTH presented the same abnormal rate of RAI as the included subjects who had received etomidate. The second scenario assumed that all the non-included patients who had received etomidate presented the same normal rate of RAI as all the patients not exposed to etomidate. These two assumptions helped us assess the impact of the potential selection bias on the main outcome. The results of these two scenarios will be presented in the results section (Table 10-A, section 3.3).

For the variable delta cortisol at 24 hours, we performed the same sensitivity analyses with two different scenarios. In the first scenario, we assumed that the non-included patients having received etomidate would present the same abnormal delta cortisol as the mean delta cortisol in the included etomidate subjects. In the second scenario, we assumed that the non-included patients having received etomidate would present the same normal delta cortisol as the mean delta cortisol for the included subjects who had not received etomidate. The results for these other sensitivity analyses will also be presented in table format in the results section (Table 10-B, section 3.3).

Among 185 patients considered, 94 were eligible to participate. The study included 40 subjects (40/94: 43% participation), who underwent ACTH stimulation tests (15 in the etomidate group) and were included in analyses on RAI. Fifty-four other eligible subjects who did not have ACTH stimulation tests performed were included in the analyses on secondary outcome measures only. The other 91 patients were not eligible.

Table 1 presents the baseline characteristics of the subjects who underwent ACTH stimulation testing. It is noteworthy that subjects having received etomidate were younger and were more likely to have been intubated at EJH (without transfer). Overall, although differences were not statistically significant, subjects having received etomidate presented markers of more severe injury. Indeed, subjects in the etomidate group tended to present to hospital with a lower mean arterial pressure (MAP) (84 vs. 90 mm Hg). This was still true even when considering all eligible subjects (Table 2).

Table 3 compares the characteristics of the participants who underwent ACTH stimulation testing (n=40) and those patients who were eligible for ACTH stimulation testing who were missed or did not consent to the tests (n=54). It shows that participants had a lower arterial blood pressure and a lower GCS than eligible subjects who did not participate to ACTH testing, and that participants were also less often transferred to EJH from another hospital. Participants to ACTH testing were thus more severe cases.

§ Appropriate tests were used: Mann-Whitney, Chi-square, Fisher exact test, Student t-test with or without Satterthwaite correction

{n}Number of missing values for this variable

§ Appropriate tests were used: Mann-Whitney, Chi-square, Fisher exact test, Student t-test with or without Satterthwaite correction

{n} Number of missing values for this variable

§ Appropriate tests were used: Mann-Whitney, chi-square, Fisher’s exact test, Student t-test with or without Satterthwaite correction.

{n} Number of missing values for this variable

Results of bivariate analyses conducted for the primary outcome, cumulative incidence of RAI at 24 hours, are presented in Table 4. There were neither statistically nor clinically significant differences between the exposure groups at 48 and 168 hours. At 24 hours, the only statistically and clinically significant difference between the groups was for delta cortisol measured continuously, where the etomidate group presented a lower delta cortisol (consistent with adrenal suppression) that was however beyond the 248.4 nmol/l cut-off chosen for dichotomization. These crude results are presented graphically in Figure 3.

Comparison of the baseline cortisol and the increases in cortisol at 48 and 168 hours did not reveal any statistical differences

Mann-Whitney test with normal approximation

†† Fisher exact test

§§ 3 subjects did not have an ACTH performed (mortality = 1, hydrocortisone given = 2) and 1 missing lab result

** 2 subjects did not have an ACTH performed (mortality = 1, hydrocortisone given = 1) and 1 missing lab result

4 subjects did not have an ACTH performed (mortality = 3, hydrocortisone given = 1)

4 subjects did not have an ACTH performed (mortality = 1, hydrocortisone given = 1, withdrawal of consent = 1) and 1 missing lab result

Yellow curve: normal distribution curve

Red line: delta cortisol cut-off criteria for RAI (< 248.4 nmol/l)

Results for primary outcomes adjusted for confounding variables are presented in Tables 5 and 6. The main observations that can be made were that there were no differences in rates of RAI between etomidate and other agents when RAI was defined with a delta cortisol inferior to 248.4 at 24 hours after adjusting the results for all confounding variables. However, when delta cortisol was considered as a continuous variable adjusted for confounders, etomidate blunted the rise in cortisol following an ACTH stimulation test done at 24 hours after the induction agent’s use. This difference was still significant even after extreme values were excluded from the regression model. This difference disappeared at 48 and 168 hours after the use of the induction agent (data presented in Annex I) .

∂ The p value was calculated for the model adjusting the rank transformation of the maximum cortisol post ACTH stimulation for the basal cortisol level. This produced a more conservative p value. Delta cortisol was adjusted for the following confounding variables: age, gender, ISS, transfer time. The hypotheses of normality of the residuals and homoscedasticity were met. * Excluding the 2 outliers in these analyses did not change the conclusions (p=0.002).

Crude and adjusted results for our secondary outcome measures (mortality rate, rate of pneumonia, rate of severe sepsis, length of stay in the ICU and FIM at hospital discharge) for all of the 94 subjects eligible for our study, are shown in Tables 7, 8 and 9. Interestingly, none of the conclusions from the bivariate analyses changed after adjusting for confounding variables. Subjects who had received etomidate had more pneumonia at 28 days (p=0.03) and had a lower motor FIM score at discharge (34 versus 55, p=0.01). While no other comparisons were statistically significant, the vast majority of the other outcome measures were worse in the etomidate group, including total hospital mortality (22.2% vs. 10.5%), with differences that are clinically not trivial.

§ Appropriate tests were used: chi-square, Fisher’s exact test, Student t-test with or without Satterthwaite correction; Δ number of subjects admitted directly to intermediate care unit;

© This clinical outcome was assessed retrospectively only on the 40 subjects who had the ACTH testing performed. This explains why there are 12 missing data in the etomidate group and 42 missing data in the other group. The motor and cognitive FIM scores are presented after a Rasch transformation and the p value was calculated using a parametric analysis of variance. This outcome was only adequately assessed in the 40 subjects having undergone ACTH testing. However, there was one other case of clinically relevant adrenal insufficiency associated with hypopituitarism diagnosed on day 4 in the etomidate group from the total cohort of eligible subjects. Since this outcome was only assessed formally in the cohort of 40 ACTH-tested subjects, this subject was not considered in the analyses.

§ LOS in the ICU is adjusted for the following confounding variables: age, gender, NISS, initial GCS, transfer status. {n} number of subjects admitted directly to intermediate care unit; ¤ The hypothesis for normality was not met. For this reason the p-value has been calculated using a non-parametric rank transformation analysis. Motor and cognitive scores for the FIM were adjusted for the following confounding variables after a Rasch transformation analysis: age, gender, NISS, initial GCS, alcohol present in blood test. The p value was calculated using a parametric analysis of variance after the Rasch transformation.

Table 10 shows the results of analyses that were conducted to estimate the impact that eligible subjects who did not participate to the analyses on RAI had on the conclusions. One can see that whatever the scenario considered, the results are always pointing towards the same conclusions, that of etomidate having a negative impact on the production of cortisol after an ACTH test. However, the strength of the associations found between the use of etomidate and a lower delta cortisol is weaker in these simulations. This does not change our conclusions because even in the most conservative scenarios etomidate seems to negatively impact adrenal production of cortisol.

P value calculated with the Fisher’s exact test

§ Only the crude analyses were performed because the sensitivity analysis of the adjusted model would have been very difficult to interpret with such a small sample size. With the small sample size, imputing simulated results to this dichotomous outcome variable (RAI) would have produced many different statistical interactions depending on the confounding variables in the adjusted model and the imputed value given to each subject in the simulation.

Note: p-values were calculated using the rank transformation

For the sensitivity analysis concerning the potential bias induced by the referral of patients from peripheral hospitals with neurosurgeons, as explained in section 2.10, one can see in Tables 11 and 12 that when all three patients referred to EJH from the hospital in Saguenay (CSSSC) are removed from the analyses, our conclusions do not change albeit the statistical significance of the results does change.

◊ Delta cortisol was adjusted for the following confounding variables: age, gender, ISS, delay for transfer. ¤ The ICU length of stay was adjusted for the following confounding variables: age, gender, NISS, initial GCS, transfer status. {n}: number of missing data for this variable

In the following paragraphs, the different limitations of this study will be discussed. Each limitation will be assessed in terms of how it impacts the conclusions of this study. The internal validity was affected by three different possible selection biases: a selective referral bias induced by selective referral of patients from certain peripheral hospitals, a participation bias was possibly introduced by a low participation rate in the ACTH testing phase of the study and finally, a potential lost to follow-up bias will also be presented. The internal validity was also affected by an information bias and different confounding variables that will be analyzed in terms of their impact on the outcomes. Limitations to the external validity and the impacts of the small cohort on the statistical power of this study will also be discussed.

As presented in section 3.3, sensitivity analyses were performed to assess the potential bias induced by selective referral of sicker and more injured subjects coming from hospitals that retain less injured trauma cases in their center. The hospital in Saguenay (CSSSC) was an example of this potential bias since a neurosurgeon was on staff there. As explained in section 2.10, having a neurosurgeon in this secondary trauma center could have created a situation were only the most injured brain trauma patients were being transferred to the tertiary referral center. Our study included three subjects referred from this hospital. Two of them were included in our ACTH testing cohort and both of them had received etomidate as an induction agent. One of these subjects had proven RAI. The other patient who was not included in our ACTH testing cohort but included in the larger cohort of 94 subjects, received another induction agent and was not included because the research nurse was not contacted in time for inclusion. We could not verify how many patients in total from the hospital in Saguenay (CSSSC) were not transferred to the EJH. In order to eliminate any potential bias caused by including subjects from this hospital, a sensitivity analysis to see if our conclusions would change by excluding these subjects was performed. As presented in Tables 11 and 12, only the level of statistical significance associated with the delta cortisol at 24 hours changed from 0.02 to 0.07 without changing the direction of etomidate’s negative impact on the adrenal response to an ACTH test. The small change in statistical significance for the variable delta cortisol at 24 hours does raise the question about the significance of this potential referral bias. But, when all the outcomes were analyzed closely, the statistically significant association between the use of etomidate and an increased risk of pneumonia did not change and all the trends demonstrating the potential harm of etomidate remained. Hence, in our view, this potential selective referral bias does not negate the associations demonstrated in this study. This change in statistical significance also has to be interpreted in the context of the small sample size of this study.

When we compared the subjects included for ACTH testing and the subjects eligible for testing who did not participate, we saw that more subjects with higher ISS scores and higher NISS scores were included for ACTH testing. In order to assess for any potential selection bias related to participation, another sensitivity analysis was performed. The results of this analysis, available in Table 10-B show that, in the best case scenario (scenario 2), etomidate would still negatively blunt the response to an ACTH test at 24 hours, but this association would then lose its statistical significance. This raises the possibility of a selection bias affecting the validity of the findings concerning the negative impact of etomidate on delta cortisol. Even the 70.1 nmol/l difference between the etomidate group and the control group does not seem clinically significant either.

However, it is important to consider that this is presenting only the best case scenario which is actually favoring the null hypothesis. In the context of this study, which is trying to determine if a medication commonly used is safe to administer to patients, we feel it is safer to privilege the worse case scenario so that no possibility of harm is left possible. For this reason, refuting the conclusions of our study on the basis of this possible selection bias is not a safe option. It is also important to remember that this potential selection bias related to participation does not affect any of the clinically relevant outcomes since all subjects were included in the analyses of these outcomes.

No differential misclassification is likely to have influenced our primary outcomes. ACTH testing results were interpreted by the medical biochemist without any knowledge of the induction agent used. Most of our clinically relevant secondary outcomes (mortality and length of stay) were assessed blinded to the induction agent used. Clinicians making the diagnoses of pneumonia and severe sepsis were not necessarily blinded to the induction agent used. We do not think that they were influenced by this information at the time of the study (2003-2004) because many of the negative studies concerning etomidate were only published in 2005 [84-86, 88, 90, 91, 157, 163]. In fact at the time, the general thought was that a single bolus dose of etomidate did no harm [31, 36, 46-49, 56].

Since clinicians were not necessarily blinded to the induction agent used by the emergency physician, this information could have biased the clinical decision to give subjects hydrocortisone to treat adrenal insufficiency. Subjects exposed to etomidate could have been perceived at a higher risk of adrenal insufficiency. They would have then received more hydrocortisone than subjects of the control group. Within the 40 subjects tested, 5 patients were diagnosed clinically with adrenal insufficiency needing hydrocortisone treatment. In the cohort of 54 subjects not tested, only 1 additional subject was treated with hydrocortisone leading us to believe that a Hawthorne effect might be present. Of the 5 subjects treated in the ACTH testing cohort, 3 had received etomidate as the induction agent.

Logistic and linear regression models were elaborated for each outcome. Each model integrated a series of confounding variables that had been determined a priori to produce adjusted results: age, gender and ISS. Other variables determined a posteriori were also considered as potential sources of bias: baseline cortisol, transfer time, transfer status, NISS, presence of alcohol in patient’s blood and mechanism of injury. These variables had to be considered linked to the use of etomidate and the outcome without being considered part of the causal chain.

The confounding variables also had to change the adjustment model by at least 10% to be kept in the final multivariate model. Others were considered but were not retained because of this criterion: initial GCS, intoxication with an illicit drug and undergoing surgery in the first 24 hours of hospitalisation. The presence of hemorrhagic shock was also assessed as a confounding variable using different markers of hemorrhagic shock: MAP on arrival in the ED, the presence of at least one hypotensive episode during the whole stay in the ED, the lowest measured systolic blood pressure during the ED stay, the amount of fluid resuscitation administered in the ED, the number of units of packed red blood cells transfused during the ED stay and the use of vasopressors. None of these surrogate markers for hemorrhagic shock was retained in any of our final adjustment models.

Certain sources of confounding biases were not well evaluated by this study because the data had not been collected and were not available in the QTR. A physiology-based severity score like the Revised Trauma Score (RTS) would have been interesting to integrate into our models, but unfortunately this information was not consistently collected in the QTR. The APACHE II score calculated in the first 24 hours of ICU admission was not helpful either because it was determined after etomidate was given. This score cannot be used as a potential confounding variable because its measure could actually be influenced by the use of etomidate. The albumin level would have been an interesting variable to consider in light of the Hamrahian et al. study [22]. We did try to adjust for the first level of albumin available in the first 48 hours of each subjects’ admission to the ICU. Unfortunately, no albumin level had been measured in 14 subjects (out of a total of 40). A sensitivity analysis was performed giving a normal albumin value to all the subjects without an albumin level present in their charts. All this analysis did was strengthen the associations already demonstrated between the use of etomidate and a decreased adrenal response to ACTH stimulation. Since the validity of this sensitivity analysis was weak and that it did not change our conclusions, we decided to exclude this variable from our final adjustment models.

Even though most sources of confounding were considered, some residual sources of bias were not: the presence of prehospital hypoxemia, the presence of abnormal pupillary reactions and the CT scan findings. Instead, we used the NISS and ISS scores as surrogate markers of injury severity. We think that the NISS and ISS, anatomical-based severity scores that predict mortality well in brain trauma victims, permitted us to adjust for the same information provided by pupillary abnormalities and CT scan findings. Since the presence of hypotension is a more powerful prognostic factor in predicting mortality in brain trauma than prehospital hypoxemia, we considered that adjusting for the presence of hypotension is probably sufficient to compensate for not having the data to adjust for prehospital hypoxemia.

The presence of adrenal haematomas and adrenal traumatic lesions are another potential bias that we did not control. The prevalence of unilateral haematomas has been reported between 0.86% and 4% [176-179]. For bilateral haematomas the prevalence is lower: 0.04 to 0.2% [176, 178, 179]. Two autopsy series report adrenal haemorrhage prevalence rates of 7.8% to 28% [181, 182]. This higher prevalence in autopsy series suggests that diagnostic imaging modalities presently in use are missing cases.

After an explicit review of all the subjects’ discharge summaries, we found only one patient that had a unilateral adrenal haematoma. However, we can’t rule out that other adrenal haematomas were missed in our cohort. If we apply the highest prevalence of reported adrenal haematomas to our cohort, we conclude that we could have missed up to 3 cases of unilateral adrenal haematomas. It is unlikely that this potential bias would have influenced our results especially considering that the clinical significance of such haematomas is still debated [183-186].

The level of cytokine production is another potential source of bias that we did not control either. Other authors have demonstrated an association between the levels of IL-6 and RAI [92, 96]. If subjects receiving etomidate had a more severe inflammatory reaction as measured by IL-6, this could have biased our results. Other cytokines, like TNF-alpha have also been involved as a cause of RAI [187]. We believe that all the different physiology-based markers of severity that we already considered in our models have adjusted for the same factors associated with cytokine levels.

Even though the rate of etomidate use (29%) in this study was generally lower than reported in other studies like Cohan et al. [99] and the NEAR 2 database [50] where etomidate use was much higher (80% and 69% respectively), there is no reason why the results of this study could not be generalized to all TBI victims aged 16 years and over. Since our study included only three 16 year-old subjects and only one 17 year-old and excluded 24 subjects aged less than 16 years, it will be important that this study be repeated in the paediatric population.

Since ACTH testing was performed in only 40 subjects, external validity could be decreased since these patients were generally sicker than patients who did not undergo ACTH testing. Indeed, these patients had higher ISS, higher NISS, lower GCS scores and lower MAP on arrival. Since many other studies done in less sick elective surgical populations [40, 41, 54, 57, 60, 74, 75] found the same results as our study in terms of etomidate’s effect on adrenal reserve, we think that our results can be generalized to a larger group of intubated brain trauma victims.

Even if our study excluded all patients receiving steroids, we think that the patients who received steroids were at higher risk of RAI and bad outcomes. These patients were mostly patients receiving dexamethasone for severe TBI as was common practice before the CRASH study was published [140]. Since the publication of CRASH, we think that the results of our study are even more important to consider, because this subgroup of TBI victims will now be at higher risk of developing RAI and possible harm from the use of etomidate.

One of the main limitations of this study was the small sample size. This limitation should be viewed in the context of the ethical difficulties encountered. For many reasons, family members of potential subjects were unwilling to participate. About 40% of all eligible subjects were included for ACTH testing. Initially, the sample size was calculated to detect a difference of 40% in the risk of RAI determined at 24 hours between both exposure groups with a bilateral hypothesis, statistical power determined at 80% and an alpha set at 5%. The effective power of our study was only 10%. A sample size of 329 subjects in each group would have been needed to detect a difference of 8% risk of RAI at 24 hours with an alpha at 5%, a bilateral hypothesis and the statistical power set at 80%. The estimates used in planning this study were based on the risk of RAI related to the use of etomidate reported by Absalom et al. [55], the only study at the time that had studied the risk of RAI after etomidate use in critically ill patients. It is possible that the risk of RAI in moderate and severe TBI could be lower than the risk of RAI after etomidate use in the population of unstable patients undergoing urgent laparatomies. Nevertheless, the definition used by Absalom et al. to define RAI led probably to more diagnoses of RAI than in our study. They measured delta cortisol only at 30 minutes after the ACTH stimulation test, unlike our study that measured it at 60 minutes. To define RAI, they also used a cut-off point of only 200 nmol/l. This could have led to a higher risk of RAI than in our study.

In order to deal with the lack of statistical power to detect a difference in RAI, we analyzed the variable delta cortisol, which is the variable used to define RAI. The analysis of this continuous variable permitted us to better detect any residual effects etomidate had on adrenal reserve measured by the ACTH stimulation test. With this approach, the effective power of our study was 60%. Unfortunately, the small sample size precluded us from drawing decisive conclusions about the effect of etomidate at 48 and 168 hours, because some of our subjects were excluded from analysis at these times due to mortality or because they were being treated with hydrocortisone.

The impact of etomidate on the clinically relevant secondary outcomes was analyzed with the objective of calculating the sample size of a larger study. Nonetheless, we were surprised to find many worrisome trends in terms of mortality and a significant difference in the risk of pneumonia. The power of this study to detect a difference in risk of pneumonia at 28 days was 57% with an alpha at 5% and a bilateral hypothesis. If the increased mortality trend would have continued, a significant difference would have been detected with only 156 subjects in each exposure group.

The risk of RAI found in other studies depends on the definition used, the population studied, the exposure to etomidate and the moment the ACTH test was done. Within the limits of such a comparison, the risk of RAI found in this study is relatively similar to the risk found in other brain trauma studies: Bernard et al. (25%) [97], Dimopoulou et al. [96] (15%) and Price et al. [93] (41%). Price et al. only included brain trauma victims exposed to etomidate. Conversely, Bernard et al. did not have any subjects exposed to etomidate and the exposure to etomidate status was not known for Dimopoulou et al. [96].

However, in populations with severe hemorrhagic shock [92], septic shock [11, 12] and critically ill surgical patients [55], the risk of RAI after etomidate use seems much higher than found in this study. Hence, there must be other risk factors that interact with etomidate to increase the risk of RAI. Hoen et al. [92] had found a 47% risk of RAI related to hemorrhagic shock. Even though the severity of injuries measured by the ISS in this study (29) was similar to the ISS in our study (35), we had less hemorrhagic shock patients. As described earlier, Annane et al. [12] reported a very high risk of RAI in septic shock patients after the use of etomidate (94%). Some authors are actually blaming etomidate, not septic shock, for causing the RAI found in the Annane study [85, 162] raising many questions about the true role of RAI in septic shock. Even after considering the limitations previously expressed about the Absalom et al. study [55], they also found a much higher rate (88%) of RAI 24 hours after the use of etomidate in critically ill surgical patients undergoing urgent laparotomies.

Considering the limits of this study previously stated, we are concerned that etomidate continues to be used outside of research settings without very good evidence of absence of harm. Theoretically, etomidate could be a contributing factor of increased rates of pneumonia, severe sepsis and eventually of increased mortality by inducing RAI and blunting the normal adrenal stress response, making patients less capable of mounting an appropriate physiological, hemodynamic and immunological response. At the very least, this study has not ruled out this possibility. For this reason, we think etomidate’s short-lived advantages in terms of hemodynamic stability and neuroprotective properties might actually be outweighed by its longer-acting negative side-effects. Could these side-effects be completely cancelled out by replacement doses of hydrocortisone for 24 to 48 hours to supplement the blunting effect on adrenal reserve? Stuttman et al. have shown this in the past [79]. Another important question to investigate with prospective and randomized studies is if other agents like ketamine could be used safely in brain trauma victims. It is clear that episodes of hypotension remain the strongest negative predictor of mortality in TBI victims. Physicians involved in intubating these patients must avoid hypotension at all costs. Etomidate for this reason is a very attractive agent. The capacity to maintain normal cerebral perfusion pressure (CPP) in the first days after TBI is also important. A normal reactivity to vasopressors is important in situations were these agents are needed to maintain good mean arterial pressures and cerebral perfusion pressures. Does the adrenal blunting induced by etomidate actually reduce the effect of vasopressors significantly? More extensive investigation is needed to explore etomidate’s initial potential for neuroprotection and its future potential to hinder the normal vasoreactivity and thus making CPP-targeted therapy more difficult. Will future studies looking at this question reproduce the findings in our study that cognitive FIM scores were more preserved than the motor scores in the etomidate group, indicating a potential neuroprotective advantage to the use of etomidate?

The essential question that must now be resolved is to determine if all the negative outcomes found in this study were related to a selection bias that we could not prevent with the type of study design used or were they really caused by the use of etomidate. In other words, is etomidate only associated with other unknown factors that are really the cause of these negative outcomes or is etomidate the only causal factor?

In the context of resuscitation, it is very difficult to enroll subjects in a clinical trial. Ideally, a larger randomized controlled trial comparing etomidate to another induction agent would have been the ideal study design to possibly answer these questions. Unfortunately, many ethical concerns prevented us from randomizing subjects to one induction agent versus another without getting consent from the next of kin. Emergency physicians decide very quickly what induction agent will be used and proceed thereafter to intubation. Time constraints prevent any real consent process to take place. For this reason, we designed our study using an observational prospective cohort with the limitations this entails, knowing that we chose the most ethical design to answer our questions.

In summary, etomidate was associated with a blunting of the adrenal response to an ACTH test up to at least 24 hours after its use. The use of etomidate was also associated in this study with a significant increase of the risk of pneumonia at 28 days, lower motor FIM scores and a worrisome trend to increased total hospital mortality and severe sepsis at 28 days.

© Patrick Archambault, 2007