Living With Uncertainty in the Intensive Care Unit: Title and subTitle BreakShould Patients With Sepsis Be Treated With Steroids?

Physicians who practiced medicine 25 years ago likely remember the use of high-dose intravenous steroids, such as 1-g boluses of methylprednisolone, for septic shock. The practice was fueled by anecdotes of miraculous recoveries, but large randomized trials failed to demonstrate benefit and raised the specter of possible harm. Steroids were abandoned and lay dormant for years, until recent studies suggested that septic shock may be exacerbated by relative adrenal insufficiency and that smaller dose of steroids for longer periods of time may be beneficial. In 2002, in a multicenter French study, Annane et al¹ reported lower mortality among patients in septic shock who were randomized to a 1-week course of low-dose steroids. This study ignited such rapid interest in steroids that the United States temporarily ran out of hydrocortisone. For those who lament the usually slow adoption of evidence into clinical practice, this event was a marked departure from the norm. Subsequent meta-analyses, systematic reviews, and professional society guidelines all echoed the notion that vasopressor-dependent septic shock should be treated with corticosteroids.

However, the story was not over. In an attempt to validate the findings from the French study, Sprung et al² and a European consortium conducted a second large multicenter trial, the Corticosteroid Therapy of Septic Shock (CORTICUS) study, which failed to demonstrate an effect of steroids in reducing mortality in patients with septic shock. These findings dampened enthusiasm for steroids, as exemplified in the 2008 clinical practice guidelines from the Surviving Sepsis Campaign (SSC).³ Although the 2004 guidelines from the same group had advocated strongly for steroids, the latest guidelines (to which both authors of this editorial contributed) made only a weak recommendation for steroids and suggested that use of steroids be restricted to patients with shock that is least responsive to supportive measures.³

In this issue of JAMA, Annane and colleagues⁴ report a meta-analysis of the entire set of trials studying steroids for sepsis. This study, conducted under the auspices of the Cochrane Collaboration, was performed and written with the full cooperation of the authors of the primary studies and with access to unpublished information. The authors report that there have now been 12 randomized trials testing the more recent strategy of low-dose steroids for a week or more and suggest an impressive overall reduction in mortality (risk ratio, 0.84; 95% confidence interval, 0.72-0.97; P = .02), even when accounting for the CORTICUS study. They conclude that steroids are indicated for all patients in septic shock, despite the findings of this trial, and in contrast with the position articulated in the SSC guidelines.

So it seems that clinicians treating patients with sepsis have 3 choices regarding steroids: no use, limited use, or broad use. Steroid use could be abandoned if the largest and latest trial, CORTICUS, is thought to effectively trump all prior studies. Steroids could be used in a limited set of patients and initiated only after it has been demonstrated that these patients are not responding to conventional measures, based on the SSC guidelines. Or steroids could be used broadly in septic shock, and possibly even in all severe sepsis (ie, any infection complicated by acute organ dysfunction), based on the results of the meta-analysis by Annane et al.⁴

Putting it mildly, this is a messy situation. Corticosteroids are inexpensive and widely available. Severe sepsis and septic shock are substantial public health problems worldwide, accounting for hundreds of thousands of deaths every year. A short, inexpensive course of steroids, if it reduces mortality, would seem to be a wonderful and prudent treatment option. But corticosteroids are no panacea, and the list of potential harmful effects is long. While there are concerns about whether corticosteroids actually could help these patients, there are also concerns about whether they actually could be harmful. Thus, it seems important to have the story right, which is why there is a rich body of clinical trials. Yet despite the large number of trials, the answer certainly is not clear. Further trials are under way, which may help tease out subsets of patients who should or should not receive steroids. But while awaiting those results, how can clinicians make sense of the available literature and of the diverging expert opinions and determine what to do now?

First, there are some technical considerations regarding the report by Annane et al.⁴ Their current meta-analysis includes information from some trials relevant to the study question that was not available to the SSC guidelines committee.³ Although the updated information had only a small effect on the risk ratio point estimate, the confidence intervals around this risk ratio narrowed such that the upper bound fell below 1 and the P value nudged below the conventional threshold of .05. Faced with such data, it is conceivable that the SSC guidelines committee may have offered a stronger recommendation for steroids.

Second, septic shock is a syndrome consisting of different clinical presentations, and differences in entry criteria among previous trials may explain discrepant results. For example, the French trial¹ may have targeted a sicker population of patients than that recruited in CORTICUS,² as evidenced by entry criteria that differ in determining how responsive blood pressure was to supportive measures and differences in control group mortality rates. Thus, it is possible that benefit is restricted to the more severely ill. When postulating that a drug works differently in a particular subgroup, there are some methodological principles to apply.⁵ The proposed difference in effect should be an a priori hypothesis with a specified direction, a credible biological rationale, and little likelihood that the difference in effect occurred by chance. That steroids should work better in patients with more severe shock was 1 of only 3 a priori hypotheses in the meta-analysis⁴ and does appear to satisfy these other conditions. However, confidence in this finding is lessened because this observation is based on between-study differences; an individual patient data meta-analysis would be required to more definitively establish or refute the hypothesis.⁶

But will further examination of the existing data reveal some convincing truth to settle the debate? No. There is residual uncertainty, and the question therefore becomes one of how to embrace that uncertainty, both in developing overarching practice guidelines and in making individual patient management decisions.

Practice guidelines are least controversial when based on high-quality evidence showing consistent and large beneficial treatment effects with minimal undesirable consequences. However, recommendations must often rely either on lower-quality evidence (due, for example, to deficiencies of methods or inconsistencies in results) or on evidence that, even though robust, suggests a close balance between the desirable and undesirable consequences of alternative management options. These 2 different types of uncertainty are crucial to distinguish: in the latter case, there is some reassurance that alternative treatment options will likely yield similar outcomes, whereas, in the former, there are concerns that favoring one option over another could have important consequences, and yet the evidence is insufficient to strongly favor either direction.

The authors of the current meta-analysis and the authors of the SSC guidelines³ used the Grading of Recommendations Assessment, Development, and Evaluation (GRADE) framework to assess the evidence, which considers these 2 sources of uncertainty separately,⁷ and both sets of authors acknowledge that the evidence is an important source of uncertainty. Presumably because of this uncertainty, both groups shy from declarative statements: Annane et al⁴ use the term “should be considered” and the SSC guidelines committee³ categorized their steroid recommendations as “weak,” using the wording “we suggest,” rather than “we recommend.” In other words, the major difference between the SSC guidelines and the current meta-analysis seems to be a difference of judgment and opinion in the face of inconclusive evidence. That means that the final decision rests squarely on those at the bedside.

It is difficult for physicians to admit to patients, families, colleagues, and, sometimes, themselves that there is uncertainty surrounding some decisions. In some working environments and specific clinical encounters, uncertainty may undermine confidence, question competence, and provoke anxiety in otherwise already emotionally charged situations. At the same time, it is easy and sometimes seductive to become emotionally invested in ideas, despite the potential for polarization and potentially acrimonious debates. And yet uncertainty is prominent in medical decision making, such as current debates regarding long-acting β-agonists in asthma⁸ or proton pump inhibitors in patients taking aspirin and clopidogrel.⁹

When physicians can embrace uncertainty, conflicts diminish and shared decision making (at least on guideline panels if not at the bedside) becomes easier.¹⁰ Embracing uncertainty helps defeat unwarranted dogma, makes the environment receptive to the conduct of clinical trials designed to diminish uncertainty, and facilitates acceptance of the results of more definitive trials that contradict existing guidelines.^{3 ,11} Embracing uncertainty also facilitates explicit communication with colleagues and students and, more importantly, with those most interested—patients and their families. Based on current evidence, physicians can and should counsel the family of a patient with septic shock that the decision to use corticosteroids (or, for that matter, to use activated protein C or right heart catheter) is not black and white and that reasonable people may reach different conclusions about exactly what is the correct decision. Admitting this uncertainty does not necessarily burden patients and families with the final decision, but they deserve to know that their physicians are not sure.