Vitamin C reduces mortality from sepsis with ARDS in CITRIS-ALI randomized trial ... ?
Vitamin C infusion has generated tremendous interest as an adjunctive treatment for patients with sepsis, since a widely publicized cohort study claimed vitamin C dramatically reduced sepsis-related mortality at a single institution.
The publicity, the plausible pathophysiologic mechanism, and the lack of any therapy for sepsis have led many intensivists to prescribe the so-called Marik cocktail (vitamin C, thiamine, and steroids), vexing colleagues who have argued randomized trials are essential before deploying the therapy.
Several such trials were designed, and the first results are in.
In CITRIS-ALI, the largest randomized trial to date testing vitamin C for sepsis, mortality was dramatically reduced among the patients receiving vitamin C -- but only as a secondary endpoint (among many tested), without a clear mechanism, and within an otherwise negative trial outcome. The plot thickens!
Authors randomized 167 adults at 7 centers diagnosed with sepsis and acute respiratory distress syndrome (ARDS) to receive ascorbic acid infusion (50 mg/kg) or placebo every 6 hours for 4 days. Patients had to develop ARDS within 24 hours of ICU admission. Patients were similarly vitamin C-deficient between groups.
Vitamin C infusion was delayed by design: patients who were known to have sepsis were not permitted to receive vitamin C until they also developed ARDS, often many hours later. Primary outcomes were organ failure and inflammation (a change in the modified mSOFA, C-reactive protein and thrombomodulin). Also analyzed were 43 prespecified secondary outcomes, mostly measures of inflammation and organ damage or recovery. Results were published in JAMA.
There was a remarkable reduction in 28-day all-cause mortality rate among the patients receiving vitamin C: 29.8% in the treatment group and 46.3% in the placebo group, which was statistically significant (P = .01) without adjusting for the multiple comparisons performed. There were also improvements in ICU-free days and days out of the hospital at day 60. (These tend to correlate with reductions in mortality.)
Interestingly, the primary outcomes were negative: there were no statistically significant reductions in organ failure, nor or in any of the secondary outcomes measuring organ failure and recovery.
Of the observed mortality reduction, authors warned,
These findings were based on analyses that did not account for multiple comparisons and therefore must be considered exploratory."
Vitamin C, Sepsis, and Multiple Comparisons
What are multiple comparisons, and why is adjustment for them considered advisable in a scientific study?
A P-value set at 0.05 provides a 95% confidence in a single comparison (i.e., rejecting a single null hypothesis). But if, say, 20 outcomes are tested, at least one is likely reach the P < 0.05 threshold by chance alone. When multiple comparisons are to be performed, statistical techniques can be used to “adjust” the alpha threshold for each hypothesis. The techniques vary in their stringency, but they all make it harder to achieve statistical significance.
This is particularly important in the age of genomic testing and other large datasets, in which hundreds or thousands of comparisons can be easily tested, generating “findings” that are just noise.
Differences in mortality are less likely to result from chance than multiple measurements of related physiological variables and organ failure scores. The P value for the mortality curve favoring vitamin C was 0.01, not borderline (i.e., 99% unlikely to result from chance, tested as a single outcome). But the authors did not design or power their study to test for a difference in mortality as a primary outcome (detecting a 15% absolute difference with high confidence would have required enrolling 324 patients). Thus, caution is indicated in interpreting the mortality benefit.
Since they're so important, why weren't multiple comparisons performed? Study authors did not explain why not, in the manuscript. The likely reason is that at n=167, the study was underpowered to detect differences with confidence after such adjustment.
They did take great pains to diminish any enthusiasm for their results, going so far as to describe the mortality reduction as one of the trial’s “limitations”:
Death and ICU graduation rates between the 2 groups were dissimilar, thus rendering the results susceptible to internal selection bias.”
But at the time of randomization, their groups weren’t dissimilar in any clear way that would influence the results. What internal selection bias?
In terms of learning whether vitamin C improves survival from sepsis, limitations of the trial were the delayed administration of vitamin C after sepsis onset, and the complicating requirement that patients also have ARDS, rather than "just" sepsis.
Multiple more trials testing vitamin C in sepsis are in process. A Chinese trial has been completed but its results have not yet been published. VICTAS, the largest of three U.S. based trials, announced an intended enrollment of 2,000 patients, but was recently reported to have enrolled 200 toward a goal of 500 patients. Its primary outcome is days free of vasopressors and ventilators, with mortality a secondary outcome.