Drs. Noble and Badgett raise important concerns about the findings of our meta-analysis. In response to Dr. Noble, a properly performed meta-analysis is similar to a clinical trial in that they are both prospective studies with inclusion criteria, end points, and analytic strategies defined before the beginning of the study. Furthermore, in a meta-analysis with significant heterogeneity, it is essential to partition the data set into homogeneous groups to explain the heterogeneity of the data (1). This is not considered an analysis of multiple comparisons but rather a fundamental technique of meta-analysis. In our study, the significantly heterogeneous effects of steroids on survival were explained by partitioning the trials by year of publication. On the basis of this division, the study by Bennett and coworkers (2) was included in the set of trials published before 1989. Subsequent descriptions of the trials were based on differences between the sets of trials, and sensitivity analyses were performed within each of these sets. Finally, the use of etomidate in the trial by Annane and coworkers (3) is concerning. However, low-dose steroids remain significantly beneficial when this trial is removed (relative survival benefit, 1.36 [95% CI, 1.04 to 1.79]; P = 0.03). Moreover, recently available data from the trial by Chawla and associates, as reported by Annane and coworkers (4) also demonstrate a consistent beneficial effect of steroids on survival (relative survival benefit, 1.82 [CI, 0.80 to 4.00]; P > 0.2).

In response to Dr. Badgett's concern of heterogeneity in mortality data among responders, the 3 trials he mentions are not different according to traditionally used significance cutoffs (P = 0.15). However, even if the studies are not combined because of the concerning I² value, there are no consistent trend toward harm in the trials and no significantly different treatment effects of steroids in responders and nonresponders within the individual trials. Categorization of septic patients as responders and nonresponders to corticotropin stimulation testing is based on a classification system that has not been validated. Moreover, the validity of these subgroups needs to be reassessed on the basis of recent data suggesting that total cortisol levels may not accurately reflect adrenal function in critically ill patients (5).

Withholding steroids may potentially harm 1 of every 9 eligible patients. Low-dose steroids have demonstrated a consistent beneficial effect on survival in 5 trials and a significant improvement in survival in our meta-analysis and in another (4). This beneficial effect remains when the trial by Annane and coworkers is excluded (3). Furthermore, no increase in adverse events was reported with steroid therapy in these trials or meta-analyses. The currently available data indicate that low-dose steroids should be considered for all patients with vasopressor-dependent septic shock. Treatment decisions should be based on individual-patient risk–benefit profiles and not on a categorization system that has not been validated, that uses total cortisol levels of questionable value, and that has produced inconsistent and nonsignificant results in clinical trials.