26 August 2010

New CDC flu numbers: This may not go well

(Constant readers: Apologies for the slow blogging. Casa Superbug's little medical crisis from a week ago has recurred, and things are a bit distracting. Back to normal soon, I hope.)

In public health, one of the numbers you hear most often — and especially so the past few years — is 36,000. That's the number of deaths that the CDC estimates occur in an average year from influenza.

Or rather, estimated. Because today, in its weekly bulletin MMWR and also in a teleconference for the press, the CDC announced that it is discarding that widely used number, in favor of newer numbers from newer studies that take into account the wide variation in illness and death from one flu season to the next.

The new estimate is: 23,607. Or, a range that goes from 3,349 to 48,614. Or, in the language recommended by a CDC scientist and a communications specialist in the press call, "tens of thousands of people [who] may die each year in an average flu season."

If that sounds difficult to communicate in a concise manner, well, the reporters on the CDC call today clearly thought so too. And while reporting study results forthrightly is transparent, and more precise numbers are almost always better, I can't help but wonder whether this attempt at precision and transparency will not be received well. After all, we are only a few months (or a few weeks, depending whose end date you accept) away from the dribbling conclusion of a worldwide pandemic that was taken so not-seriously by the public that, in the US, 71 million doses of H1N1 vaccine went unused — and in Europe, some public representatives alleged that the entire emergency was a concoction by pharmaceutical companies.

Given that history, putting out a public message that flu kills fewer people than we thought — but is, still, a serious disease that should be planned for and vaccinated against — sounds like a hard sell.

Here's how today's new numbers came about:

The mortality rate from flu has always been difficult to assess: People die of influenza directly, but they also die of underlying conditions — heart disease or chronic obstructive pulmonary disease, among others — that might not kill the person if influenza were not putting an extra strain on the system. In either case, but especially in the latter, the death may not be attributed to flu, particularly if the victim has not been tested for the presence of the flu virus.

So, to arrive at an estimate, the CDC has used a statistical model. As explained in the briefing today by Dr. David Shay of the CDC's Influenza Division:
We have two categories that we look at... One is death certificates that have an underlying diagnosis of pneumonia or influenza. 99% of those deaths are actually coded as pneumonia. So, that's to make an estimate of deaths in a particular season from pneumonia that are associated with flu. And typically, that's about 8.5% of deaths over the time period that we looked at...  The broader category of respiratory and circulatory deaths we think encompasses the full picture of influenza-associated deaths, including things such as people who might die because of worsening chronic obstructive pulmonary disease or worsening congestive heart failure that results in death after an infection. And we estimate that about 2% of that broader category in any typical year is associated with influenza.
One other factor affects flu mortality: Which flu strain type is dominant in the season being measures. Flu is generally taken to cause the most severe disease, and the greatest number of deaths, in the elderly; but some strains cause more severe disease than others, and some (H1N1 "swine" flu, for instance) attack the young, who are healthier and less likely to die, more than they do the old. Again, Shay:
[I]t's important to keep in context, which we don't really describe in this article because of space, that there's at least four factors that affect sort of flu mortality in any particular year, and those four would be the specific strain or influenza strains that are in circulation, sort of the length of the season or how long influenza is circulating in the united states, how many people get sick, because of course, the more people get sick, there is more likely to be more serious outcomes, and finally, who gets sick.
In the study released today, the CDC did two things: It broadened the range of flu seasons from which it took data to feed into the statistical model, and it took a second look at the years on which the previous model, the one that produced the 36,000-death estimate, was based.

When the range of years was broadened to 31 flu seasons (1976-77 to 2006-07), here's what shook out:
  • For deaths from influenza and pneumonia: from 961 in 1986-87 to 14,715 in 2003-04, an average of 6,309
  • For deaths from respiratory and circulatory complications: from 3,349 in 1986-87 to 48,614 in 2003-04, an average of 23,607.
(When asked which number should be used for shorthand, Shay said: "The broader category of respiratory and circulatory deaths we think encompasses the full picture of influenza-associated deaths.")

When the 36,000-death estimate was re-examined, Shay said:
The 36,000 number that's often used pertains to a very specific time period from 1990 to 1999. And in that decade, where we had prominent circulation of H3N2 viruses, they were prominent in eight of the nine seasons that are contained within the data that were used to make that estimate, and those are, as you know, typically more severe seasons. We had a high mortality for that nine-year period.
According to the MMWR analysis, mortality rates in the H3N2 years were 2.7 times higher than in years when other types were dominant.

So that's the rationale behind today's dialed-down numbers. Here's the potential problem with it: It just took me about 1,000 words to (somewhat talkily) explain. It requires patience and detail to impart, which in the current media environment are in very short supply. As one of the participants on the call said today:
I'm really scratching my head here wondering what I'm going to use, because we really don't have a lot of time ... to present a lot of numbers, and I think in a sense to say that the range is 3,000 or 3,300 to 49,000 raises a lot of questions, and I think we don't have time to answer those questions in every report. And I also wonder if it's not a bit misleading to use 3,300 as the bottom number since it's been 20 years since it was that low, and even in the last 20 years, the mortality has never been much below 12,000.
You see the problem.

To repeat: This is an effort at transparency and accountability; those are worth applauding. But it's also a nuanced and difficult health-communication message, launched into a zeitgeist already tuned toward conspiracy theories and a media marketplace with little time or expertise to counter them.

Pessimistically, I wonder how long it will be before this message gets transformed into something like, "See? I told you so. Flu isn't that big a deal after all." I hope the CDC is prepared when it does.

(Here's today's MMWR article, the transcript of the press briefing, and a Q&A on the new calculation. The cite is: Morbidity and Mortality Weekly Report, "Estimates of Deaths Associated with Seasonal Influenza --- United States, 1976--2007." August 27, 2010. 59(33);1057-1062)

1 comment:

Julie said...

Interesting analysis of how the reporting of numbers can skew our view of a problem. You're right- the difference between 3,000 and 49,000 is huge! How do you correctly report that in a way that gives the big picture?