Yesterday (today as I am writing this) the British Medical Journal published another Cochrane meta-analysis on the efficacy of neurimminidase inhibitor antivirals (the only two in use now, being oseltamivir [Tamiflu] and zanimivir [Relenza]). Their conclusions have made the news, so I guess I should cast my baleful eye on their handiwork. I think there is less here than meets the eye, but first let’s look at what meets the eye.
This is a meta-analysis, that is, an analysis of other analyses, the other analyses in this case being drug trials of Tamiflu or Relanza in children. So it’s an observational study of experimental studies, if you follow me. “Experimental” in this context means studies on children where the investigators control the treatment assignment (an antiviral or a placebo or no treatment). In essence, it is a review article of other studies where the authors employ various qualitative and quantitative techniques to combine results to come up with a summary of the studies, taken together. It’s more complicated than that, but that’s the general idea. It’s potentially better than the kind of reviews of the literature commonly done because it is very explicit about what kinds of analyses will be included and the methods of weighting the results are also explicit and often quite sophisticated. I say potentially better because it takes a very narrow view of the subject, only allowing data from scientific studies of a particular kind. The Cochrane consortium last looked at this question in 2005 and this article looks at what has been learned in the interim, clearly with an eye on the swine flu pandemic now underway. All of the analyses they looked at were for seasonal flu, not pandemic flu. How generalizable this is to pandemic flu is unclear, but it is reasonable to think it has relevance.
There were only four studies meeting their strict eligibility criteria in this time period using antivirals to treat flu in children, two for each antiviral. The total number of children in the treatment studies was just under 1800, about two thirds of whom had virologically confirmed flu. Two thirds of those had influenza A. The rest had flu B. There was no differentiation of flu A subtypes. Three more studies used antivirals (two Relenza, one Tamiflu) as post exposure prophylaxis (just under 900 children). The studies were randomized trials involving children less than 12 who had flu (either clinically suspected or virologically confirmed) but not sick enough to be hospitalized. Thus the conclusions say nothing about the efficacy of these drugs for really sick children, although this is not clear from the news reports (example from the BBC: “The antiviral drugs being used to treat swine flu do not appear to work well in children, say UK researchers.” This also fails to mention this was not a study of the use of these drugs for swine flu.)
I’ve never been much of a fan of antivirals as a way to deal with pandemic flu, but that’s because I’m a public health person, not a clinician. My interest is in things that work on the population level, not applied at the level of the individual. In addition I was never impressed by how well these drugs worked. At times I even wondered if they worked at all were we to take a hard look at them (many of the studies were financed by the drug companies and we now know this is often a tainted source). Antivirals aren’t like antibiotics, where you can sometimes work almost miraculous cures. If they work, they work at the margins. With those prejudices of mine as background, I can state that one clear message of the 2005 review and this one is that both of these drugs work to some extent. You wouldn’t get that idea at all if all you read was the news headlines (the BBC again: Flu drugs ‘unhelpful’ in children). They not only work, but they work on every single measure examined: reduction of time with symptoms (.5 days to 1.5 days); time of illness (.4 days to 1.5 days), where this was defined as resolution of all symptoms and resolution of fever and return to school or normal activities; reduction in cough or fever (1.3 days for Tamiflu, “reduced incidence of moderate or severe cough” at day five with Relenza); reduction in asthma exacerbation; improvement in pulmonary function; reduction in antibiotic use; reduction in otitis media (middle ear infection) at day 10; and in confirmed secondary cases with an infected index case (reduction in transmission of about 8%).
In some of these outcomes the differences were not statistically significant, but in no instance — not a single one I was able to see in this paper — was there an outcome that wasn’t consistent with the drugs working. Remember that the authors were surveying less than 3000 children divided up into seven studies, each of varying quality and different outcomes. The most plausible explanation for a lack of statistical significance given the consistent pattern and the biology is that there was insufficient statistical power. It is certainly false and absolutely incorrect to say that the outcomes that were not statistically significant showed no effect. In every instance they showed an effect of the drug in the right direction. In some instances the alternative explanation of random variation remained on the table along with a real effect.
In trying to make the jump from this review to the current swine flu situation there are other issues to contend with. Anywhere from 2% to 20% of the children in these trials had gotten seasonal flu vaccination, which could well have modulated the severity of their illness and the response to antivirals. This is especially pertinent in trying to extrapolate to populations of children in the initial stages of this pandemic, where none of them will have been vaccinated. Second, as already noted, none of these children were sick enough to be hospitalized. The marginal improvement in these community residing patients may look very different in very sick children who are on the margin between needing vents or no vents or even a fatal outcome. Nor do we know much about children with serious underlying medical conditions. That is work that remains to be done and some of it is happening now. Third, we know little about the efficacy of these drugs across flu A and flu B or between subtypes of flu A. Are there significant differences? We know that the neuriminidase protein on the virus looks different in different subtypes (there are at least two main NA subgroups) but we don’t know much about the effect of those differences on antiviral efficacy. Moreover it is likely that many of these children were infected with viruses resistant to Tamiflu. So far the swine flu virus remains sensitive.
While we fear this paper will lead to misinterpretation and we think it has been oversold, it does raise some very important questions. It presents clear evidence these drugs work, but do they work well enough? In the Big Picture, for most patients the improvement is not very significant (although for individuals a day or more gained without symptoms is not just a tunafish sandwich). Tamiflu has some adverse effects, most commonly nausea and vomiting. That’s not fun, either. When told of this, my daughter declined the Tamiflu she was offered when she came down with the virus.
Finally, there is the prospect that unwise use will promote resistance and deprive us of one of the few therapies we have for swine flu. This is not a simple issue. To have the greatest chance of success these drugs should be used relatively early in the course of the illness, often before anyone knows which direction it is headed. If we wait to use them only when the patient goes sour, they will have lost much of their potential efficacy. On the other hand, suppose we do lose their use because the virus develops resistance. The current seasonal H1N1 is almost entirely resistant to Tamiflu (but not Relenza) and we have managed quite well without it. That’s the silver lining to the fact these are not miracle drugs. If you lose them, you aren’t losing the ability to work a miracle. It’s the good news in the bad news.
The real take home lesson in the BMJ paper is that for seasonal flu these drugs work for children but not well enough to rely on them for much. We don’t know whether this is also true for swine flu and for very sick children, but studies now underway which will likely provide some data. How much of what this paper provides will turn out to be of any use remains to be seen.
Meanwhile, caveat lector.