Should Glucocorticoid-Induced Hyperglycemia Be Treated in Patients With Septic Shock?

Critical illness, in particular severe sepsis, induces insulin resistance and hyperglycemia. Corticosteroids are often used for reversal of fluid- and vasopressor-resistant septic shock. Such an adjuvant treatment aggravates illness-induced hyperglycemia, even in a low-dose steroid regimen.¹ For glucocorticoid-induced hyperglycemia in noncritically ill patients, there is general agreement on treatment, because prolonged hyperglycemia causes cardiovascular and infectious complications.² Whether patients in septic shock in the intensive care unit (ICU) with glucocorticoid-induced aggravation of “diabetes of injury” should be treated is controversial. This debate is embedded in the overall controversy about whether to treat critically ill patients with hyperglycemia with insulin, and if so, to what blood glucose level.³

In this issue of JAMA, the Corticosteroids and Intensive Insulin Therapy for Septic Shock (COIITSS) study investigators consortium⁴ report the results of a multicenter randomized controlled trial that addresses this relevant question. The primary hypothesis the authors aimed to test was whether targeting blood glucose levels to a range of 80 to 110 mg/dL (to convert mg/dL to mmol/L, multiply by 0.0555) improves outcome of patients with hydrocortisone-treated septic shock compared with usual care. The overall conclusions of this trial were that among patients with hydrocortisone-treated septic shock, intensive insulin therapy targeting normoglycemia compared with usual care did not improve in-hospital mortality and that compared with use of hydrocortisone alone, the addition of fludrocortisone did not improve in-hospital mortality.

A decade ago, usual care was not to treat critically ill patients with hyperglycemia unless glycemia exceeded the renal threshold of approximately 200 to 215 mg/dL (with glucosuria and hypovolemia as a consequence) on the assumption that hyperglycemia was a potentially beneficial adaptation to critical illness—an appropriate stress response. However, this treatment strategy changed after studies found that lowering blood glucose levels to between 80 and 110 mg/dL decreased mortality by an absolute 3% in surgical ICU patients⁵ and reduced morbidity by prevention of secondary complications, such as severe infections,⁵ kidney failure,^{6 - 7} and intrahepatic cholestasis and biliary sludge,⁸ as well as critical illness polyneuropathy,^{9 - 10} dyslipidemia,¹¹ and endothelial dysfunction.¹² This intervention was labeled “intensive insulin therapy,” and although the proof-of-concept study was performed in one center in Leuven,⁵ blood glucose lowering was rapidly adopted in ICUs worldwide, despite different settings and patient populations. Perhaps surprisingly, expert guidelines also advised treatment of hyperglycemia in ICU patients, especially those with sepsis.¹³ Although the proof-of-concept study concluded that normoglycemia was superior to not treating unless glycemia exceeded the renal threshold, the guidelines advised targeting a blood glucose level to an intermediate level of less than 150 mg/dL.¹³

As a consequence of this change in clinical practice, the “usual care” in the COIITSS study targeted blood glucose levels less than 150 mg/dL, consistent with the guidelines.¹³ In the intervention group, the target glucose level was between 80 and 110 mg/dL, although the achieved mean levels were between approximately 120 and 130 mg/dL and at times were higher. In the control group, the mean blood glucose levels were approximately 145 mg/dL. The range of the actual glucose levels achieved was broad, and there was considerable overlap between the levels in the 2 study groups. Thus, the lack of difference in outcome could be because normoglycemia apparently was difficult to achieve in a multicenter setting and the actual blood glucose levels were not substantially different from a strategy that targeted an intermediate glucose level of less than 150 mg/dL. Importantly, patients in both study groups required considerable insulin to lower the pronounced baseline hyperglycemia of 215 mg/dL, presumably induced by the combined diabetic stress of sepsis and hydrocortisone treatment. Although the authors consider the control strategy to be comparable with the Leuven study, this was not the case; a small proportion (39%) of patients in the control group in the Leuven study received insulin, with a median daily insulin dose of zero for the total control group⁵ compared with a mean daily insulin dose of 46 units in the COIITSS study control group.⁴

The COIITSS trial results might thus suggest that clinicians should titrate insulin to a blood glucose level of approximately 145 mg/dL in patients with hydrocortisone–treated septic shock, thereby avoiding hypoglycemia. However, such a conclusion requires careful consideration of the statistical power of the trial. In the COIITSS trial, the authors hypothesized that based on the 32% relative risk reduction in the first Leuven trial,⁵ intensive insulin therapy would be expected to result in a 25% relative risk reduction from a baseline 50% mortality, with a hypothesized absolute risk reduction of 12.5%. For an intervention designed to prevent complications—rather than cure the underlying disease—the estimated absolute mortality risk reduction should be used to calculate the sample size. Indeed, only the excess deaths attributable to those complications, in this case hyperglycemia-induced, can be prevented by the intervention. With intensive insulin therapy, an absolute mortality reduction of 3% is what could be expected based on the Leuven trial,⁵ and thus the 12.5% absolute risk reduction was unrealistically high.

Moreover, in retrospect, the hypothesized 12.5% absolute risk reduction appears to be even more of an overestimate considering that the usual care group received more insulin and had lower blood glucose levels than those in the Leuven study. To place these estimates in perspective, if the absolute expected mortality reduction with targeting normoglycemia might have been only 1.5%, half of that obtained in the Leuven trial,¹⁴ the required sample size would have been 70 000 patients. The 509 patients studied in the COIITSS trial are not sufficient to confidently conclude equivalence between the 2 compared blood glucose targets. Thus, clinicians caring for patients with septic shock treated with hydrocortisone will still be left with uncertainty as to whether insulin should be given and to what level the blood glucose should be lowered, adding to the uncertainty of whether to treat with hydrocortisone in the first place.¹⁵

Such a situation seems, on the surface, somewhat hopeless: the COITTSS trial investigators executed a difficult multicenter trial in very complex, seriously ill patients, and yet clinicians can only conclude from their efforts that there is still uncertainty about how to do things differently. Another conclusion might be that often the only robust evidence that clinicians can use comes from megatrials, the very execution of which could be plagued by ongoing changes in usual-care practices.¹⁶ Yet, syndromes such as septic shock continue to portend a grave threat to life, the provision of ICU care for these patients is exceedingly costly, and a lack of adequately robust evidence on how best to provide that care is a glaring deficiency. Rather than tolerate a climate of clinical uncertainty, it seems imperative for funding agencies and researchers invested in care of critically ill patients to conduct adequately powered trials, even if these trials might be far larger than those of the past. To ensure that such studies can be completed in a timely fashion, the cooperation of national and international trials groups, and their funding sources, will likely be necessary. Precedents for large-scale international cooperation exist in oncology and cardiology. Given the huge global burden of conditions such as septic shock, which causes hundreds of thousands of deaths in the United States alone each year, such international collaboration should and must be achievable.