In the wake of the much-awaited ADRENAL trial, APROCCHS hasn't received much attention. Maybe because of its name (“APROCCHSS”? Seriously??). Regardless, APROCCHSS does add some useful insights to the steroid puzzle. The essentials of this trial have been reviewed already by TheBottomLine, so this will be a short post focusing on implications.
The elephant in the room: mortality benefit
ADRENAL and APROCCHSS were both designed with mortality as a primary endpoint. They reached opposite conclusions: steroid had no effect on mortality in ADRENAL, whereas it improved mortality in APROCCHSS. Why?
The optimistic viewpoint
There are several reasons why APROCCHSS may have detected a mortality benefit, which was missed in the ADRENAL trial:
- Patients in APROCCHSS were enrolled earlier than patients in ADRENAL. The time from shock onset to randomization was 20 +/- 90 hours in ADRENAL, versus < 24 hours for all patients in APROCCHSS). Never mind the golden hour – some patients in ADRENAL didn't receive steroid until an entire golden day had passed since the onset of shock.
- Likewise: Patients in APROCCHSS were bolused with steroid (50 mg IV q6hr), whereas patients in ADRENAL were started on a continuous infusion without any loading dose. Starting therapy with a bolus dose in APROCCHSS could allow earlier achievement of therapeutic levels.
- Patients in APROCCHSS were sicker than those in the ADRENAL trial. For example, on average patients in APROCCHSS were on ~70 mcg/min norepinephrine, whereas patients in the ADRENAL trial were on ~30 mcg/min.
Perhaps early administration of steroid to the sickest patients reduces mortality. It's possible that this effect was missed in ADRENAL, simply because patients were less sick and received steroid too late.
The pessimistic viewpoint
The mortality benefit in APROCCHS is fragile, with a borderline p-value (0.03) and a fragility index of 3. This effect could occur simply due to random chance. The fragility of this effect is evident in the fact that there was no mortality benefit observed at day 28 (p=0.06). A large, robust mortality improvement should be observable at both time-points. The fact that the p-value dances around 0.05 over time shows that the mortality benefit lacks robust statistical significance (1).
Recall that the interim analysis of the ADRENAL trial suggested a benefit from steroid. However, recruitment of further patients eventually showed no benefit (suggesting that early signs of benefit were a statistical fluke). This emphasizes the possibility that APROCCHSS could show benefit merely due to random chance. If APROCCHSS had recruited additional patients, perhaps it too would have been negative.
I don’t think it’s currently possible to know whether steroids affect mortality in septic shock (or some subset of these patients). As discussed recently, it's nearly impossible to unequivocally prove mortality benefit in critical care trials. Good arguments can be made on both sides of this debate. Ultimately, this matter will remain unresolved for now. Meta-analyses will be performed, but they won't be helpful (meta-analysis cannot fix problems due to heterogeneity between different study populations and different interventions).
What can we learn from the secondary endpoints in APROCCHSS?
The use of secondary endpoints has been hotly debated, particularly with regards to the ADRENAL trial. When a large robust trial is performed and the primary endpoint is negative, it seems wasteful to ignore the rest of the data. However, secondary endpoints do create certain statistical pitfalls. My blog last week explored one approach to secondary endpoints, which will be used here. In short, this strategy focuses on 1-2 major secondary endpoints, and evaluates them with a more stringent p-value threshold (p<0.025-0.01).
A reasonable set of major secondary endpoints in APROCCHSS might be vasopressor-free days and ventilator-free days. Both of these endpoints are clinically meaningful, patient-centered endpoints. Since they are more proximally related to steroid administration than 90-day mortality, these endpoints should be easier to test definitively. To avoid type-1 error, a cutoff value of p<0.01 should be used when evaluating these two secondary endpoints. The results of these endpoints are as follows:
Unfortunately, intervention-free days were analyzed in terms of mean and standard deviation. Intervention-free days form a profoundly skewed, non-normal distribution, which makes measurement of standard deviation inappropriate (figure below). Since the mean is highly susceptible to outliers, this metric isn't optimal either. Overall, evaluation of intervention-free days using mean and standard deviation is a crude approach, which could miss significant effects (over-estimate the p-value). Specifically, steroid reduced the number of ventilator-free days with p=0.07 using this analysis, so one would expect to find a lower p-value using a more sensitive analysis (2).
The observation that steroid accelerates weaning off vasopressors has been replicated consistently across numerous prior trials of septic shock (e.g. ADRENAL, CORTICUS, Annane 2002). At this point, it seems safe to conclude that this effect is real.
The effect of steroid on weaning from the ventilator has previously been shown in the ADRENAL trial. Likewise, steroid has been shown to avoid the need for intubation among patients with community-acquired pneumonia (Siemieniuk 2015). Although this result isn’t quite as definitive as vasopressor weaning, this appears to be a real and replicable effect.
Safety endpoints were also similar to the ADRENAL trial. Steroid increased the rate of hyperglycemia, but didn’t appear to cause more serious side-effects. The total rate of serious “adverse events” was actually higher in the placebo group than in the steroid group!
This study doesn’t change my approach to steroid
My current approach is to initiate stress-dose hydrocortisone (50 mg IV q6hr) for any septic patient on vasopressors without contraindication to steroid. The rationale for this is to accelerate recovery (less time on ventilation, less time with a central line, less time in the ICU). Accelerated recovery is a meaningful patient-centered outcome. This is identical to the use of steroid in numerous other conditions, where steroid hastens recovery without affecting mortality (e.g. COPD exacerbation). Perhaps steroid even reduces mortality in a subset of patients – if so, that’s an added bonus. Given the proven safety and efficacy of steroid, mortality benefit isn’t necessary to justify its use.
Details of steroid selection & administration
Hydrocortisone infusion vs. bolus
ADRENAL used a continuous hydrocortisone infusion, which is different from usual practice in the United States. There are several reasons that using a hydrocortisone infusion is illogical:
- Starting an infusion without a loading dose means that the patient won't receive a therapeutic dose of hydrocortisone for several hours.
- Setting up hydrocortisone as a continuous infusion is logistically challenging (this may tie up an intravenous line and require infusion of additional fluid as a vehicle for the infusion).
- The rationale for administering any drug as a continuous infusion is that it has a short half-life (e.g. vasopressors, esmolol). Hydrocortisone has a biological half-life of ~8-12 hours, so it makes little pharmacological sense to give it as a continuous infusion. In no other situation is hydrocortisone given as a continuous infusion.
APROCCHSS proves that giving hydrocortisone as an intermittent bolus (50 mg IV q6hr) is successful at accelerating clinical improvement. If you believe the mortality benefit, then APROCCHSS might even show that intermittent dosing is superior to a continuous infusion. Overall this supports current practices in the United States of giving hydrocortisone as an intermittent bolus (3).
It is doubtful that fludrocortisone is doing anything here, for a few reasons:
- 50 micrograms of fludrocortisone daily isn’t really all that much. In one-third of septic patients, this doesn’t even generate a measurable serum level of fludrocortisone (Annane 2016).
- Hydrocortisone has both glucocorticoid and mineralocorticoid activity. It’s questionable how much fludrocortisone adds to the mineralocorticoid activity of hydrocortisone and endogenous aldosterone.
- The COIITSS trial investigated the additive effect of fludrocortisone on top of hydrocortisone in 509 patients with septic shock. There was no detectable benefit from fludrocortisone on any outcome.
Some may argue that the addition of fludrocortisone is why both Annane 2002 and APROCCHSS showed mortality benefit. I disagree. It’s a common myth that Annane 2002 showed mortality benefit, but this is not what the actual data demonstrates. Mortality benefit was only shown in a sub-group, adjusted analysis – not in the analysis of the entire patient cohort (more discussion of this here).
My opinion is that the available evidence doesn’t support the use of fludrocortisone. However, it’s not unreasonable to do so.
- It is unclear whether steroid can reduce mortality in a subgroup of patients with septic shock. Given inconsistencies between ADRENAL and APROCCHSS, this question cannot be answered definitively using the available evidence. Meta-analyses are coming soon, but they won't be valid or helpful.
- Steroid administration hastens clinical resolution, with accelerated liberation from the vasopressors and mechanical ventilation. This corroborates the results found in the ADRENAL trial.
- Hydrocortisone 50 mg IV q6hr was effective in this trial, demonstrating that a continuous infusion of steroid is unnecessary.
- Steroid use in sepsis remains a contentious issue. Early initiation of hydrocortisone in septic shock is safe, accelerates recovery, and might improve mortality. This is a sensible approach to patients without a contraindication to steroid.
- My opinion on steroid in sepsis
Acknowledgement: Thanks to Dr. Gilman Allen for thoughtful comments on this post.
- Furthermore, statistical significance at a level of p<0.05 isn't that robust anyway. More on this to come the next post.
- The same mistake was made in the ADRENAL trial when analyzing the number of vent-free days. Analysis using a mean +/- standard deviation yielded a borderline significant result (p=0.06). However, analysis of time to weaning via time-to-event analysis using a hazard ratio showed that steroid caused significantly faster liberation from the ventilator (p=0.0007). This was explored in the prior blog on ADRENAL trial here.
- If I started ordering hydrocortisone infusions, I would almost certainly get assassinated. The only question is whether it would be at the hands of the MICU pharmacist or the nurses.