High-flow nasal cannula (HFNC) has become popular for the hypoxemic respiratory failure, driven partially by the FLORALI trial.1 Recently, the HIGH trial evaluated the ability of HFNC to reduce mortality among immunocompromised patients with acute hypoxemic respiratory failure.2 HFNC failed to improve mortality or significantly reduce intubation rates.
Popular narrative about this trial
A common viewpoint of this trial is explored in an accompanying editorial by Dugan et al.3 These authors opine that FLORALI led to premature adoption of HFNC, despite weak evidentiary support. Furthermore, they feel that the HIGH trial contradicts the FLORALI trial, indicating that HFNC shouldn’t be used among patients with immunocompromise:
The remainder of this post will argue against this position.
There is extensive evidence to support HFNC
HFNC isn’t a passing fad. In addition to the FLORALI trial, two other RCTs have shown that HFNC can prevent re-intubation following extubation.4 5 Although secondary prevention of intubation isn’t the same thing as primary prevention, the physiology is the same.
Aside from these RCTs, there are numerous other studies showing that HFNC has the capacity to reduce work of breathing, alleviate tachypnea, and improve oxygenation.6 The mechanisms underlying the efficacy of HFNC are extremely plausible (CO2 washout and low levels of PEEP).
Results from FLORALI and HIGH are entirely consistent
The popular narrative is that FLORALI is a “positive” trial, whereas HIGH is a “negative” trial. Therefore, it would seem that these trials contradict one another. However, if we examine the actual data from the trials, there is no disagreement. Let’s focus on three outcomes: intubation, mortality, and ventilator-free days.
I will use free meta-analysis software produced by the Cochrane Collaborative (REVMAN5) to combine and compare data from the two studies. This obviously isn’t a formal meta-analysis. However, this provides a simple way to directly compare raw data from the two studies.
Intubation rates
There was no statistically significant reduction in intubation rate in either trial (comparing HFNC versus conventional oxygen treatment). Both studies show a trend towards reduced intubation with HFNC. If these results are combined, the net result almost shows a statistically significant reduction in intubation (p=0.06).
Please note that the HIGH trial does not exclude a reduction in intubation rates. Within this study the hazard ratio for intubation has a very wide 95% confidence interval (ranging from 0.68 to 1.06). If, for example, the hazard ratio were actually 0.68, that would constitute a clinically useful reduction in intubation rates. This demonstrates that the HIGH trial is underpowered to exclude a reduction in intubation rate.
Mortality rates
FLORALI reported a 2.55 hazard ratio for mortality when comparing standard oxygen to high-flow nasal cannula. However, the 95% confidence interval of this hazard ratio comes extremely close to one (1.07-6.08) with a marginal p-value of 0.046. If you re-run these numbers using a Fisher’s Exact test, this difference isn’t statistically significant at all (p = 0.16). This sort of discrepancy between different statistical tests is surprisingly common. It highlights the folly of strictly adhering to a p=0.05 cutoff.
HIGH didn’t find a mortality benefit either. So, these studies agree: there is no mortality benefit from HFNC.
Lack of a mortality benefit from HFNC shouldn’t be surprising. First, mortality is a very difficult endpoint to ever affect in a clinical trial (as explored here). Second, the effect of HFNC on intubation rates isn’t huge. Intubation probably does cause a small increase in mortality (e.g. the intubation procedure itself carries a small mortality rate). However, given that the increase in intubation rates is small, and that the mortality increment from intubation is small, it shouldn’t be surprising that HFNC doesn’t cause a statistically measurable mortality benefit.
Ventilator-free days?
The number of ventilator-free days over the first month is a standard endpoint for these studies.7 FLORALI showed an increase in ventilator-free days with the use of HFNC.
HIGH initially listed ventilator-free days as a secondary endpoint on clinicaltrials.gov and also in their study protocol (available at JAMA). Unfortunately, this endpoint mysteriously vanished from both the final manuscript and supplemental data. Post-hoc abandonment of this endpoint is a substantial methodological flaw. Ventilator-free days is arguably a more clinically useful endpoint here than mortality, given the high likelihood that a mortality endpoint would be negative.
Immunosuppressed patients with hypoxemic respiratory failure are a challenging cohort
The above analysis shows that FLORALI and HIGH are in considerable agreement. However, it must also be noted that they did evaluate different patient populations. HIGH involved immunocompromised patients, most of whom had malignancy, poor performance status, and ongoing chemotherapy:
These are an extremely challenging cohort of patients. Based on the severity of their underlying disease, it’s likely that many of these patients would do poorly regardless of therapeutic strategy. For example, during the trial 22% of patients placed limitations on the aggressiveness of their treatment and these patients eventually suffered a 79% mortality rate. Any intervention tested in this cohort will tend to look unimpressive.
Clinical bottom line & risk/benefit analysis
For patients with hypoxemic respiratory failure and marked dyspnea, we would like to offer them some supportive therapy to avert exhaustion and thereby avoid intubation. Historically BiPAP was often used here, but FLORALI showed harm from BiPAP. Therefore, currently our choice is HFNC versus low-flow oxygen.
HFNC is notable for a truly impressive safety profile. No study has found any significant risk from HFNC (neither FLORALI, nor HIGH, nor other RCTs of HFNC). I have yet to encounter any serious adverse effect from HFNC, despite using this on a daily basis in very sick patients.
The benefits from HFNC may be less impressive than commonly perceived. There doesn’t seem to be a mortality benefit from HFNC. However, available evidence does suggest that HFNC reduces intubation rate. There is a very strong physiological rationale that HFNC improves work of breathing and therefore should reduce intubation rates. FLORALI and HIGH both found trends in this direction which, when combined, nearly reach statistical significance (figure above). This is further supported by two RCTs showing that HFNC reduces the re-intubation risk following extubation.4 5
Utility = Benefit / Risk
Overall it appears that the benefits of HFNC outweigh the risks of HFNC (which are essentially nil). I will continue to use HFNC for patients with severe hypoxemic respiratory failure in the ICU, even patients with immunosuppression. Of course, HFNC should be used with high-level monitoring and intubation should be provided promptly for patients failing treatment.
Parting shot: Biases involved in publication
There are two biases at play here, which are worth mentioning. Most medical writers are guilty of these, myself included.
Punchline bias
Publishers and readers of a study are looking for a catchy punchline. Is HFNC good or bad? Yes or no? Writers may be unconsciously pressured to write in such a way that the messaging is consistent. In a “positive study” (such as FLORALI), this could pressure authors to massage the data to make it look more consistently positive. In a “negative study” (such as HIGH), this could pressure authors to present things in a more pessimistic light. This may overall cause authors to present data in a black/white manner, overlooking subtle inconsistencies in the data.
Recency bias
There is a tendency to assume that the newest, hottest study is the correct one, whereas an older study is wrong. For example, the HIGH trial just came out, so it must be more accurate than FLORALI. This is often true, but certainly not always. Overall recency bias creates a dangerous tendency to focus solely on the newest study, while ignoring prior work on the topic.
- Conventional wisdom is that FLORALI is a positive trial, whereas HIGH is a negative trial. As such, these trials are portrayed as contradicting each other.
- The actual data from FLORALI and HIGH are consistent. FLORALI isn’t quite as optimistic as it is commonly perceived, whereas HIGH isn’t quite as dreary as it has been described. The truth probably lies in between.
- Evidence doesn’t show a mortality benefit from HFNC. However, combining data from HIGH and FLORALI does suggest a non-significant reduction in intubation rates. Such a reduction has been shown in two RCTs following extubation.4 5
- HFNC is remarkably safe. None of these trials report any worrisome complications from HFNC.
- Currently, benefit/risk analysis continues to favor the use of HFNC for hypoxemic respiratory failure. HFNC isn’t a magic bullet for hypoxemic respiratory failure, but it may help some patients and it is extremely safe.
Related
- HFNC to avoid reintubation
- HFNC for early ventilator weaning
- HFNC for pneumonia & FLORALI study
- mastering dark arts of HFNC & BiPAP
References
- PulmCrit Blogitorial – Use of ECGs for management of (sub)massive PE - March 24, 2024
- PulmCrit Wee: Propofol induced eyelid opening apraxia – the struggle is real - March 20, 2024
- PulmCrit wee: Why I like central lines for GI bleed resuscitation - March 13, 2024
Great analysis of the trial. I know it’s anecdotal but I think we’ve all seen the struggling patient who gets put on HFNC and looks like a million bucks. The high trial won’t change what I do.
Yep, one nice thing about HFNC is that it’s easy to try it on a patient with basically zero risk.
Dear Josh, thanks for another great article! In the post hoc outcomes the authors write : “In the overall population, vasopressors and renal replacement therapy were needed in 153 patients (19.7%) randomized to high-flow oxygen therapy and 31 patients (4%) randomized to standard oxygen therapy, with no statistical difference between groups. ” Not to imply causality here, but what did you make of this? Need for vasopressors at time of randomization was one of the things the stratified randomization was trying to equalize. Limitations to treatment seem to be equal (both at randomization and during the course of treatment) too.… Read more »
Thanks Josh, very interesting read and good principles that can be extrapolated to the simple life (ie ED)
Whenever you see a large effect size without a significant p-value, that means there is a lack of statistical power.
This is a secondary endpoint of little interest so I wouldn’t put much weight in it. I pay attention to secondary endpoints, but only when they are things that I would previously have predicted would be relevant/central to the outcome. https://emcrit.org/pulmcrit/secondary-endpoints/
Please describe the oxygen delivery system you are using to apply HFNC oxygenation therapy to your patients. My understanding of this type of oxygen delivery device consists of an air/oxygen blender, an active humidifier, a single heated circuit, and a nasal cannula. It delivers adequately heated and humidified medical gas at up to 60 L/min of flow and is considered to have a number of physiological effects: reduction of anatomical dead space, PEEP effect, constant fraction of inspired oxygen, and good humidification. Others in my organization have indicated to me that HFNC oxygen therapy is delivered through a simple nasal… Read more »
Recently published systemic review by B. Rochwerg in ICM (2019) confirm much of your points…Tq