The big news out of the UK this week is the RECOVERY trial. We just may have found a simple therapeutic intervention that can reduce the mortality in the COVID epidemic. This is fantastic news. So I am going to break down the trial and get past the media hype and press releases to see what this trial really means for front line medicine..
DISCLAIMER: this is a preliminary preprinted version of the trial. There is still analysis and data to be crunched… but we have the big picture now, not just a press release: so I think it is ready for practice.
RECOVERY. Is an adaptive platform trial ( or series of trials really ) run in the NHS in the UK. There are a number of pragmatic therapeutic trials going on, and this is just one of the interventions that are being tested.
This part of RECOVERYwas an open-label, randomised controlled trial that used 6mg of dexamethasone (oral or IV) vs. usual hospital care.
To get into the trial patients needed to be admitted to hospital and with either lab-confirmed or “clinically suspected” SARS-CoV-2. They were excluded if the treating clinical team thought that the patient had a clear contraindication or possibly a prior indication for dexamethasone (though I can’t think of many absolute indications??).
11,000 screened. 9355 eligible but of that 6425 were included in analysis… we don’t know what the exclusions were for or who they were… so important to know before we are confident. Having said that it’s a big representative cohort of the sort of folk we would expect in this population.
This was a pragmatic trial which means that the treatment was allowed to be as natural and reflective of real-world practice as possible. In theory this design should increase the external validity of the trial outcomes.
Clearly dexamethasone is a widely available medicine, so it did not require special trial dispensation and approvals. We already use dexamethasone or other steroids in a range of conditions similar to SARS eg. Sepsis, pneumonia, ARDS in many patients. More on this later…
Population was initially adults but expanded to include children and pregnant or breast feeding patients which is rare in trials. COVID has made us change the way we do research. This is good as we may be able to apply this data to more folk out there.
Patients were consented, enrolled and randomised 2:1 to either group. They were stratified by the level of ventilatory support at the time of randomisation [ no oxygen requirement, those needing supplemental oxygen ( including NIV) and those needing invasive ventilation].
The average age was 66 years, and there was a slight but “statistically significant” difference between the groups with those in the Dex group being a year older on average. This may disadvantage that group given the big effect of age on COVID- Mortality. The trial team actually retrospectively corrected for this and it made no difference to the actual outcome.
The intervention was simple really 6 mg of dexamethasone a day. This could be continued up to 10 days in total. In reality – 95% of patients allocated to dexamethasone group received at least 1 dose and the median number of days of treatment was 6 days.
7% of the “usual care” group received dexamethasone. The data was analysed according to the intention-to-treat principle. This 7% cross over would tend to reduce the potential benefit of dexamethasone.
The timing of the intervention relative to disease onset has been an issue with a lot of the Covid trials. In this trial the least sick group started Dex about a week after symptoms, 9 days for the O2-only group and 13 days for the ventilated patients. This was basically the same in both arms. About a quarter of the patients were diabetic, we don’t know about glycemic control data, but it didn’t seem to kill them if you accept the improved mortality.
The two groups were otherwise very similar at the outset with all the usual comorbid conditions. 81-82% were swab positive, about 10% negative and the rest “pending” at recruitment.
Now let’s get to the outcomes.
We are trying to wean off of p- values and “significance”. So I will talk about the point estimates and we can save the p-value to explore the Bayesian weight of the data produced here..
The primary outcome is all-cause mortality at day 28 post randomisation.
Dex group = 21.6%, usual care = 24.6% Absolute risk reduction = 3% which gives an overall NNT of 33 for all comers.
The numbers that were being spruiked in the press release were about the relatively larger benefits seen in those patients who were at the more severe end of the disease spectrum. Although this is true… it should not really come as a surprise, spectrum bias is certainly a thing and with this disease ad a relatively ‘safe’ intervention we would expect to see a larger gap at the bad end of the spectrum. If you crunch the numbers the NNT gets better as you get sicker: it was 28 for those needing O2 or NIV and 8.5 for those on a ventilator.
You could argue for possible harm in the least sick group, though it is a hard call. This group still had high mortality – in the mid teens. It is unclear as to how many of them were not offered higher level care as a result of their pre-existing disease or how many had very mild disease that really would get better with no doctors involved.
This group were randomised 3-6 days earlier than the other groups in the course of their disease. As we know COVID can follow a lot of different trajectories it is very likely that this group includes folk who went home well a day later and folk who got tubed a few days later… very heterogeneous group at outset… we need to know more individual-level data to see what this group actually represents.
To be clear, we need to recognise that there was no benefit for those who were not receiving supplemental oxygen – AT THE TIMEOF RANDOMISATION. What this trial cannot tell us is the benefits of dexamethasone once folk get sicker and need more support. But, yes, it does seem to follow logically that they should get the same benefit if they were to subsequently need NIV or intubation.
There was a secondary analysis of the effect of dexamethasone on “disease progression”. The risk of progression to invasive mechanical ventilation was lower among those allocated dexamethasone vs. usual care (risk ratio 0.76 [95% CI 0.61 to 0.96]; p=0.021) – this is a secondary outcome, so needs to be replicated in a dedicated cohort.
So if it helps stop progression yet might harm the mildly afflicted then it becomes a question of timing – choosing when to pull the trigger on the dex gun…. somewhere after a week? after the first need for oxygen supplementation? when the inflammatory markers peak? We do not know….
Now, if yo have ben listening to the podcast recently you will be aware of the statistical process of converting p-values into Bayes factors and trying to make sense of new data as it applies to our prior understanding of an intervention.
So let’s take the primary outcome from RECOVERY. The p-value is quite potent – p< 0.001. This converts to a maximum Bayes factor of 53. That is a strong magnet to swing our needle of doubt in this case. What that means is that even if your prior belief in the efficacy of dexamethasone was just 5% then your posterior probability would jump up to 73% – well past 50/50.
Given that we already have some evidence (not too strong) for steroids in ARDS, pneumonia and sepsis then we could probably be more generous with our prior probability of efficacy – if we use 20% as a guess-timation then our posterior belief would be 93%. That is as close to a good thing as science gets in medicine.
So that just leaves bias to muddy to the waters. If there is significant bias ten this with throw a spanner into the RECOVERY works and no number of zeros prior to the p-value will be useful. So was there bias?
This is an open-label trial – so yes, there will be bias. The outcome was measured using an online form that was presumably completed by staff who were potentially aware of allocation. This can produce bias in favour of the intervention. But yes, it is hard to fake death! It is the ultimate patient-oriented, binary outcome. However, we can still worry about the other secondary outcomes for this bias effect.
This is a state-funded trial of a cheap and widespread drug therefore it seems less likely that we will see sponsorship bias. It was pre-registered and stuck to randomisation and protocol so that is good.
One more potential bias is what I call “desperation bias” this is what we have seen in recent months – we desperately want our therapy to work and we have seen a lot of swings and misses… this trial however is well conducted and minimally biased with strong outcomes so I don’t believe that desperation bias is in play here!
SUMMARY: Dexamethsone is beneficial in hospitalised patients who are in need of supplemental oxygen, or ventilatory support.
I am a GP working in Broome, NW of Western Australia. I work as a hospital DMO (District Med Officer) doing Emergency, Anaesthestics, some Obstetrics and a lot of miscellaneous primary care. Also on the web as @broomedocs | + Casey Parker | Contact