Interlukin-6 antagonists improve outcomes in hospitalised COVID-19 patients

Claire Vale, one of the study's lead authors, reflects on  the largest ever prospective meta-analysis of treatments for COVID-19, and its implications for future evidence syntheses.

Results published yesterday in JAMA by the WHO REACT group have confirmed a benefit of the IL-6 antagonists, tocilizumab and sarilumab, in patients with severe COVID-19. 

Importantly, as well as being more effective overall, IL-6 antagonists were more effective in patients who were also receiving corticosteroids compared to those receiving corticosteroids alone (OR=0.78, 95% CI 0.69-0.88) That is important as since September 2020, WHO guidelines have recommended that hospitalised patients with severe or critical COVID-19 should be treated with corticosteroids.

These new results have informed a new WHO guideline (also published yesterday) which recommends giving tocilizumab and sarilumab alongside corticosteroids.

This is all clearly great news clinically and an important step forward towards reducing mortality in the wake of this pandemic.  But this is also an important step in the evolution of meta-analysis.  In particular, of prospective meta-analysis – or PMA.  Prospective meta-analysis of summary or aggregate data is a relatively new concept.

This PMA has been ground-breaking.  It has needed to be – we are in the midst of a global pandemic. But not only for its clinical importance.  For me, it has pushed the boundaries of what is possible in meta-analysis and potentially sets a new standard.  I explain here…

It has involved weekly team meetings – bringing together the principal investigators from all the key trials, representatives from trial sponsors – often from pharmaceutical industry, as well as from WHO.  I was lucky enough to be brought in, I think because of my previous experience of PMA and also of IPD meta-analysis.  The numbers on those calls swelled as more trials agreed to participate. 

In the end 27 trials contribute to the meta-analysis.  That’s 10930 patients from trials conducted across 28 countries. That’s around 95% of all patients randomised in all of the eligible trials

Because it was prospective all of those voices could contribute to the design of the meta-analysis. We sometimes painstakingly agreed what should the main comparisons be /what outcomes to address and how to define them? Ditto secondary outcomes, subgroups etc. These were all predefined with the aim of reducing potential biases and ultimately, ensuring results would be useful to clinicians and policy makers.

In all, we defined 10 outcomes for the meta-analysis. We planned to look at each outcome overall, and then stratified by whether the trial used tocilizumab or sarilumab, to ensure the effects of both drugs looked consistent.  But we didn’t stop there…

Trials had recruited patients before and since corticosteroids were recommended by WHO, and so we needed to assess the effects of these agents overall, and within patients receiving and not receiving steroids at the point of randomisation.

In the first WHO REACT group PMA, the effect corticosteroids appeared less marked in patients receiving invasive mechanical ventilation (ratio of ORs, 4.34 [95% CI, 1.46-12.91]; P = 0.008) and so in this PMA, we also planned to look at the effect in subgroups defined by different levels of ventilation. 

We also wanted to look at whether effects were consistent across standard subgroups like age and sex and also by important baseline characteristics like indicators of the inflammatory response or disease severity.

I can’t even tell you in total how many analyses that adds up to in total – I think it would shock me to know!  I do know that the data collection form for baseline data (which included extensive information used for risk of bias assessments) extended to 7 pages; and the outcome data was 10 tabs of an excel spreadsheet.

I assumed when trial teams saw all of that they would refuse to cooperate.  We were asking for so much data.  And these were busy trialists – treating patients in the midst of pandemic, as well as running and reporting their own trials.  But none did.  All supplied as much of the data as they could, answered queries to ensure that what went into analyses was as accurate as possible.

We tested our data collection forms with a couple of trial investigators, whose trials had already completed, during the second half of Feb 2021.  Once that test was completed and the forms tweaked, we started data collection proper on 27 the Feb. Through March and into early April, data kept coming in.  We also kept trying to contact investigators of trials (those we identified late or with whom we didn’t get responses immediately).

On 9th April we circulated preliminary analyses for the primary outcome to the collaborative group. We were still working on some of the subgroup analyses; cleaning and querying data and working on the remaining outcomes.  I was pretty anxious before that Teams call – but the results were generally well received by the investigators. It gave us chance to really understand the implications of the results and the concerns and queries of the investigators.  And if I’m honest, as someone new to the infectious diseases world – and to COVID research in particular - a chance to properly understand the trials and data we had been given charge of.

Over the following 5 weeks we finalised the data collection- 27 trials and 10,930 patients.  And ran analyses – and there were plenty!  My statistician colleagues, Pete and David – were rocks throughout.  They developed code to check data – making it easier to pick up inaccuracies and raise queries – not to mention to run the analyses. They even created code to automate the outputs – tables, graphs etc – into well thought through, clearly labelled and indexed results documents, that have been used as supplementary files for the publication.  

At the same time, a separate team, based at University of Bristol, undertook risk of bias assessments using the information we had collated, as well as papers and protocols from the individual trials.  When you remember that we’re talking 27 trials and 9 separate outcomes, that was a huge amount of work too!  

At the end of that 5 weeks – on 16th May 2021, we circulated a first version of the manuscript to the collaborative group.  After revisions and edits, the manuscript was submitted to JAMA on 21st May. That brings us up to date.  After peer review and editorial review, we were finally ready to publish our meta-analysis.  Alongside the WHO guideline that it informs.  No time wasted – results into guidelines and hopefully to the bedside and patients around the world who are still dying from this disease in huge numbers. 

It has only been possible through collaboration, dedication, and the good will of all concerned.  If I ever needed convincing that prospective planning and collaboration with trial teams was vital to great meta-analysis, this was all the proof I needed.  The entire experience was truly inspiring and humbling.

The evidence synthesis community needs to take note.  I think we have moved into a new era.  We can no longer wait for trial results to be reported before we think about starting a systematic review.  Things move too quickly – not only because we’re in a global pandemic. I have made my contribution to this work from a spare bedroom/ office in North London.  But this was not evidence synthesis in a vacuum.  I don’t believe we should ever again start a new systematic review at arm’s length from those who know the disease, the trials and the broader implications.  We owe it to the thousands of patients who take part in these trials and who will take part in the future.