background on IV vitamin C
We live in divided times. Paul Marik’s study on using vitamin C in sepsis was perhaps the most polarizing publication in recent memory (aside from perhaps the Mueller report). It was lauded by some, yet derided just as strongly by others. Unfortunately, the kerfuffle surrounding this study overshadowed the larger picture of vitamin C in critical illness (which is supported by a reasonable body of basic science evidence and some small clinical studies).
Currently, numerous RCTs are underway evaluating the role of vitamin C in septic shock. CITRIS-ALI is the first major RCT to reach press, with others chasing closely at its heels. It’s tempting to simply wait a couple years until all the studies come out, the carnage subsides, and some consensus emerges from the rubble. However, our critically ill patients can’t wait that long – so we are forced to parse each new study for bits of truth that could help our patients now.
background on the CITRIS-ALI trial
A bit of background will help explain the design of this study (and, frankly, add some dramatic irony). For years, mortality was a preferred endpoint in studies of ARDS. This led to numerous “negative” studies, for two reasons:
- ARDS is a multi-factorial disease that is exceptionally hard to treat.
- Mortality endpoints are extraordinarily difficult to achieve in a modern critical care setting (explained further here).
Eventually, the NIH got tired of funding studies which used mortality as a primary endpoint. Consequently, they designed a specific grant to study a therapy for ARDS which included the following stipulations:
- Mortality could not be the primary endpoint.
- Co-primary endpoints should include both a clinical endpoint and a biomarker endpoint.
And thus, CITRIS-ALI was born. This is pure speculation, but my guess is that the CRP and thrombomodulin endpoints were added primarily to obtain the NIH grant (they’re unusual primary endpoints to choose in a multi-center clinical trial).
This NIH grant may have created another wrinkle in the story. Fowler's research team had previously performed a pilot study involving IV vitamin C in patients with severe sepsis.1 However, this grant was focused on ARDS, so research objectives were shifted slightly towards use of vitamin C for ARDS due to sepsis. This design change between the pilot study and the multi-center RCT is subtle, but potentially problematic.
design of the CITRIS-ALI trial
This was a multi-center, placebo-controlled, double-blind trial of IV vitamin C among patients with ARDS due to sepsis. Key inclusion criteria were as follows:
- Suspected or proven infection.
- ARDS, defined as follows (the diagnostic criteria of ARDS and ALI actually changed during the study period, so “Acute Lung Injury” is no longer preferred diagnostic terminology!)
- Acute onset hypoxemic respiratory failure.
- Bilateral opacities not explained by chronic lung disease or heart failure.
- PaO2/FiO2 ratio <300 mm Hg.
- PEEP of at least 5 cm water via endotracheal tube.
- ARDS onset must have occurred within <48 hours of study inclusion.
- Absence of active kidney stones.
Patients were randomized to placebo versus IV vitamin C 50 mg/kg IV every six hours for up to four days (e.g. 3.5 grams IV q6hr for the prototypical 70-kg patient). This is a bit more than twice the dose used in Paul Marik’s trial and in most ongoing studies of vitamin C (which are using a fixed dose of 1.5 grams IV every six hours). The rationale for using this higher dose is a pilot study by Fowler’s group which suggested greater benefit from higher doses of IV vitamin C.1
The co-primary endpoints were:
- Change in SOFA score at 96 hours compared to baseline.
- Levels of C-reactive protein (CRP) and thrombomodulin.
These primary endpoints were based on a pilot study of severe sepsis wherein IV vitamin C was provided earlier in the disease course (within <48 hours of ICU admission).1 They were applied to this study, even though IV vitamin C was potentially administered later on in the disease course (within <48 hours of ARDS onset).
Secondary endpoints included the following:
- All-cause mortality at day 28.
- ICU-free days at day 28, ventilator-free days at day 28, and hospital-free days at day 60.
- Indices of oxygenation and ventilation efficiency.
- Vitamin C levels.
- Additional biomarkers (procalcitonin, receptor for advanced glycation endpoints, and tissue factor pathway inhibitor).
- SOFA score sub-components (e.g. creatinine, urine output, platelet count etc.).
Patients were well matched at baseline. Most patients had pneumonia as the source of sepsis and ARDS. Patients were fairly ill (~70% of patients were on vasopressors, the average PaO2/FiO2 ratio was ~200, and about a quarter of patients had acute kidney injury). 65% of patients were on corticosteroid upon study entry.
There was no difference at all in any of the primary endpoints (change in SOFA score, CRP levels, or thrombomodulin levels). So based on the primary endpoints, this is a solidly “negative” study.
But it’s not that simple. Patients in the Vitamin C group had a lower mortality. This difference is most striking during the first 96 hours while on IV vitamin C (during which mortality was ~23% in the placebo group vs. ~4% in the vitamin C group):
Patients in the Vitamin C group also experienced more ICU-free days (median of 11 vs 0, p=0.03) and more hospital-free days (median of 22 vs 0, p=0.04). These differences are largely driven by differences in mortality (patients in the control group with zero hospital-free days and zero ICU-free days mostly represent those who died):
There was no evidence of harm from IV vitamin C. In particular, patients in the Vitamin C group didn’t have increased risk of kidney injury (based on the kidney component of the SOFA score).
mortality benefit with no improvement in △SOFA score?
Now we get to the crux of the paper. During the first four days:
- The mortality in the vitamin C group was dramatically lower than the control group (~4% versus ~23%).
- The average SOFA score improved less in the vitamin C group compared to the control group (3-point improvement in the vitamin C group vs. 3.5-point improvement in the control group, p=0.9)
These trends are running in opposite directions! The vitamin C group had a much lower mortality, but their SOFA scores improved less. That just doesn't make any sense.
How can we reconcile these stark differences?
The answer may be survivorship bias (a form of selection bias which results from focusing on survivors, while overlooking those who died). Increased mortality in the control group selectively removes the sickest patients. This will paradoxically make the control group look better on average!
In fairness, the study did use the the change in SOFA score over 96 hours as their primary endpoint (△SOFA). Using the △SOFA will tend to reduce survivorship bias, because each patient is compared to their own baseline. However, △SOFA still remains susceptible to survivorship bias, because the sickest patients will generally recover slowest (so eliminating the sickest patients will still tend to improve the average △SOFA).
One way to conceptualize is as two teams which are running a race against each other. The distance covered by each person within 96 hours is analogous to their △SOFA. The control group has a much higher mortality, thereby eliminating their sickest patients (who will be least competitive in this race). This will tend to improve the average performance among the control group – even though fewer patients remain in the race.
Thus, it’s possible that vitamin C may be a victim of its own success. By keeping patients alive, Vitamin C retained the sickest patients within its sample. These extremely ill patients tend to recover very slowly – so they have a low △SOFA, preventing Vitamin C from demonstrating efficacy.
Overall, this suggests that the unexpectedly large mortality difference may compromise the validity of the primary endpoint (△SOFA). The ability of a secondary endpoint to stage a statistical coup d’état which invalidates the primary endpoint is controversial, to say the least. Some statisticians will doubtless comment below that such a coup is impossible, but as a rebel I’d like to believe that it is.
is the difference in mortality clinically meaningful?
This is the key question. Strong arguments can be made on both sides of this debate, as follows:
Reasons the mortality difference isn’t meaningful:
- Mortality is a secondary endpoint. According to traditional rules for interpreting clinical trials, if the primary endpoint is negative, then secondary endpoints can be used only for hypothesis generation.
- The difference in mortality isn’t incredibly robust. A Wilcoxon test based on the entire survival curve analysis yields p=0.01. However, Chi-squared analysis using only the final mortality rates after 28 days yields p=0.03. As discussed previously, although these p-values are below 0.05, this doesn’t exclude the possibility that they could be due to chance.
Reasons why the mortality difference is meaningful:
- The primary endpoint is negative, but the mortality differences may actually undermine the validity of this primary endpoint (staging a statistical coup d’état as discussed above).
- Mortality is such a critical endpoint that it’s always important (regardless of whether it’s technically the primary or secondary endpoint).
- Imagine, for a moment, that Vitamin C caused an increase in mortality. That would be extremely concerning – and would certainly attract a lot of attention. Thus, it’s only fair to make note of the fact that Vitamin C appeared to reduce the mortality. If we are willing to accept only negative data, but not positive data – that’s bias.
Currently this debate defies any definitive answer.
where does this leave us now?
Opinion on this trial will doubtless be sharply divided along partisan lines. Vitamin C skeptics will view this as a negative trial (which it technically is). Vitamin C enthusiasts will view this as a positive trial (based on evidence of safety and mortality benefit). Ultimately, the study probably won’t sway many hearts or minds.
I view this study as moderately positive. Vitamin C was given very late in the illness course (up to 48 hours after meeting diagnostic criteria for ARDS). This is far beyond the “golden hour” or even the “golden days” of resuscitation. The possibility that Vitamin C could have any impact when provided this late is impressive. On the whole, evidence of some benefits (especially mortality) combined with lack of harm moves the needle towards the administration of vitamin C.
This study obviously isn’t definitive. We eagerly await upcoming RCTs to clarify the topic further. In the interim, Vitamin C seems to be a reasonable therapy which is safe and potentially beneficial.
- CITRIS-ALI is a multi-center RCT investigating IV vitamin C in patients with sepsis-induced ARDS. Vitamin C was initiated relatively late in the disease course (up to 48 hours after the development of end-organ failure meeting ARDS criteria).
- The primary endpoints of this trial were negative (including change in SOFA score over 96 hours and two biomarkers). However, some secondary endpoints suggested benefit from Vitamin C (including improvements in mortality and ICU-free days).
- It’s possible that mortality differences could have biased the primary endpoint (SOFA score), due to selective elimination of sicker patients from the control group. Thus, it's conceivable that the secondary endpoint staged a coup d’état, overthrowing the primary endpoint.
- No evidence of harm from IV vitamin C was found (despite use of larger doses than used in Paul Marik’s study).
- CITRIS-ALI manuscript on JAMA site.
- Discussion of statistical methods in CITRIS led by statistician Dr. Andrew Althouse
- Evidence for metabolic resuscitation (PulmCrit)
- Metabolic resuscitation, and Questions/Answers (iSepsis, by Paul Marik)
- YouTube video by Dr. Fowler discussing CITRIS-ALI (section on CITRIS starts ~38 minutes into the presentation):
- 1.Fowler A, Syed A, Knowlson S, et al. Phase I safety trial of intravenous ascorbic acid in patients with severe sepsis. J Transl Med. 2014;12:32. https://www.ncbi.nlm.nih.gov/pubmed/24484547.