Steroids in septic shock: Four misconceptions and one truth

The utility of steroids in sepsis has been debated passionately for decades.  There is hope that steroids might improve mortality, but also fear that they could increase infectious complications.  Practice varies widely.  What does the data truly indicate?

This belief is based on the first modern RCT of stress-dose steroids in sepsis by Annane et al. 2002.  This study randomized 299 patients with severe vasopressor-refractory shock to placebo versus stress-dose steroids.  Patients were also divided into two subgroups depending on whether they responded to an ACTH stimulation test.  Their hypothesis was that patients who failed to respond normally to the ACTH stimulation test had inadequate adrenal function and would improve with stress dose steroids.  The primary outcome was 28-day mortality (additional details in TheBottomLine blog here). 

Analysis of the raw data reveals no mortality benefit for steroid among any of the patient subgroups (see the unadjusted p-values, which I have added to the table below on the right column).  Most notably, 28-day mortality was not significantly reduced among patients who didn’t respond to the ACTH stimulation test (p=0.11, highlighted red box). 

Here is where things get murky.  The authors went on to perform an adjusted mortality analysis using a regression model derived from a prior study.  The adjusted analysis yielded lower p-values, with a value of p=0.04 for the primary outcome (red box above).  The remainder of the paper, and nearly all of the subsequent medical literature, has focused exclusively on the p-values from the adjusted analysis. 

This may not be valid.  There is generally no need to perform an adjusted analysis of an RCT.  The entire concept of a RCT is that randomization will naturally account for random variation, making an adjusted analysis unnecessary.  An adjusted analysis may rarely be performed if there is a considerable imbalance between the groups at baseline, but in this study the groups appear fairly well matched at baseline.  There is no explanation in the manuscript as to why an adjusted analysis was performed. 

Currently the ACTH stimulation test in sepsis has been abandoned, partially because it is erratic.  Therefore, the non-responder subgroup analysis is no longer clinically relevant to us.  The data which are most relevant today are the results from all patients combined.  These results are negative regardless of which statistical analysis is used.  Therefore, from our perspective today there is unequivocally no mortality benefit.

As shown below, a meta-analysis failed to find any mortality benefit in anysubset from any study (1).  Note that this meta-analysis used the uncorrected data from the Annane 2002 study.

Thus, no study has convincingly shown a mortality benefit in any group of patients.  It is possible that a mortality benefit exists which is too small to detect in these studies.  It is possible that a yet-unidentified subgroup of patients experiences a mortality improvement.  However, no mortality benefit has ever been clearly proven.

Concerns regarding superinfection with stress-dose steroid are largely based on the CORTICUS trial, the second major RCT of steroids in sepsis.  This study randomized 499 patients with septic shock to receive stress dose steroids or placebo.  The primary endpoint was 28-day mortality, with no difference observed between groups (for additional details see TheBottomLine here).  The authors also evaluated the occurrence of 25 adverse events, most notably:

This data was reported in the paper as follows:

This description is misleading.  As shown in the table above, there were no statistically significant differences in the rate of superinfection, new sepsis, or new septic shock.  In order to overcome this lack of statistical significance, the authors seem to have created a post-hoc combined adverse outcome of “new episodes of sepsis or septic shock.”  The validity of creating this combined outcome is questionable, and the statistical analysis of this combination outcome remains unimpressive (2). 

As shown below, meta-analysis by Sligl 2009 failed to detect any change in the rate of superinfection among any study or the pooled results. 

Thus, no study or meta-analysis has shown an increase in superinfections.  This possibility certainly hasn't been excluded.  However, if this risk exists, it doesn't seem to be of substantial clinical significance.

Steroids consistently improve hemodynamic stability, thereby allowing earlier withdrawal of vasopressors and decreasing the duration of shock.  As shown in the meta-analysis below by Sligl 2009, this is reproducible across multiple studies and patient populations.

For example, the duration of septic shock among patients in the CORTICUS trial is shown below.  Among all patients, steroids decreased the median time to shock reversal from 5.8 days to 3.3 days (p<0.001). 

It is widely believed that the Annane et al. and CORTICUS trials conflict with each other.  Annane et al. is perceived as supporting the use of steroids, whereas CORTICUS does not. 

However, the actual datafrom these two studies are entirely consistent.  The raw data from both studies shows no mortality benefit with steroids, faster shock resolution, and no increase in superinfection.  The Forrest plots above illustrate that in every single case the point estimates from both studies overlap, indicating statistical agreement.  If these two studies had been statistically analyzed and written in a less imaginative way, they would look and sound nearly identical. 

Rather than the data itself, it is primarily the data interpretation and press surrounding these studies which is conflicting.  Annane et al. performed an adjusted analysis which seemed to show a mortality benefit, causing enormous excitement about steroids.  CORTICUS reported no mortality benefit and focused on some trends among the adverse events, splashing cold water on the enthusiasm generated by Annane et al.

This misconception is a direct consequence of the misconception that Annane et al. was a positive study whereas CORTICUS wasn't (misconception #3).  It is commonly believed that the reason for this "difference" is that patients in Annane et al. were sicker at enrollment (Annane et al. required patients to have vasopressor-refractory shock, whereas CORTICUS did not).  This has led to the common belief that steroids should be reserved for patients with vasopressor-refractory shock.

In fact, Annane et al. and CORTICUS revealed similar clinical benefits from steroids (faster shock reversal).  Within CORTICUS, steroids caused an improvement in hemodynamic stability and possibly renal function (more on this below).  Thus, benefits of steroids do not seem so be restricted to patients with vasopressor-refractory shock.  

Patients with more severe sepsis might benefit more from steroids than patients with milder sepsis.  As shown above the side-effects from steroids are probably roughly stable, whereas the benefit may be greater for patients with more severe disease.  However, the point at which benefits might outweigh risks is unknown.

Thus, steroids in sepsis are neither as awesome nor as scary as is often believed.  Steroids won't improve mortality, but neither will they lead to terrible superinfections.  The primary benefit of steroids may simply be to reduce the duration of shock.  This shouldn't be a huge surprise because stress dose steroids (200 mg/day hydrocortisone) are equivalent of 50 mg prednisone daily, a commonly used dose which is fairly safe in short courses.        

This may be similar to the role of steroids in patients with COPD exacerbation.  Steroids don’t improve mortality in COPD.  Compared to placebo, steroids reduced the average length of hospitalization by merely 1.2 days (Niewoehner 1999).  Nonetheless, steroids are an accepted treatment for COPD exacerbation.

The debate about steroids in sepsis has been focused on whether steroids effect mortality.  This might be the wrong question.  A more relevant and practical question may be whether it is worth using steroids in order to reduce the duration and severity of septic shock.

Reducing shock duration could have meaningful consequences.  Prompt resolution of shock could improve organ perfusion and function.  For example, a post-hoc analysis of CORTICUS showed that patients with renal failure who received steroids had a significantly higher likelihood of renal recovery (figure above; Moreno 2011).  As discussed previously here, renal recovery in septic shock appears to be critical for achieving good outcomes. 

The precise risk/benefit balance remains unclear.  Steroids are proven to improve hemodynamics and speed shock resolution, which might facilitate recovery among the most unstable patients.  However, steroids do have a number of potential side-effects including myopathy, hyperglycemia, and peptic ulcer disease.

The above approach may be reasonable.  Some patients have indications or contraindications for steroids, but mostly it may be a matter of clinical judgment.  Individual patient characteristics may tip the risk/benefit balance one way or the other.  In the absence of definitive evidence, judgment is needed. 

Unfortunately, clinical judgment is nebulous and leads to variation between practitioners.  However, this is the nature of critical care medicine.  The most that we can ask of ourselves and our colleagues is to carefully consider the situation and make our best call.