May 4, 2009 The Pump Handle 0Comment

by revere, cross-posted from Effect Measure

There have been questions in the comments about where the CDC estimate of 36,000 to 40,000 influenza related deaths a year comes from. It’s a figure I’ve used a number of times here to say generally that regular old seasonal influenza may be a mild disease for some but not for many others. Even if you don’t die of flu, it can be a miserable illness and lay you low for several weeks of acute illness and months of fatigue and malaise. Now the 36,000 deaths number is taking on a life of its own, so it’s time to explain exactly what it is and what it isn’t. There are more things it isn’t that it is.

First, of all, it isn’t a count of deaths each year from flu. We don’t know how many people die of seasonal influenza each year because there is no list we can use to count them. Why not just use the death certificate information? You can see what a death certificate looks like here. The Cause of Death section has two parts. Part I. asks for the immediate cause of death (first line) and any underlying conditions that brought it about (up to four subsequent lines). The underlying causes are the “due to” components. Suppose someone dies of gram negative sepsis, a total systemic failure usually caused by a bacterial infection. The lines below the immediate cause are supposed to be the links in the causal chain leading to the sepsis. Bacterial pneumonia would be a typical cause of sepsis, so we have an immediate cause of death of gram negative sepsis due to bacterial pneumonia. Secondary infection by bacteria is a common complication of a respiratory viral infection, so the next line might be viral pneumonia or, if known, influenza infection. Unfortunately, the cause of most deaths that may be related to influenza is not verified by any virological testing for infection by the virus. Many things besides influenza virus cause pneumonia. Most often there is no evidence of what virus or other agent was the cause of death, especially for viral pneumonias. If it’s during a flu outbreak, influenza may be listed as a probable underlying cause without any lab evidence. If influenza occurs at some other time it will likely not be listed anywhere.

There is also a second part of the Cause of Death to list any conditions that might have contributed to the immediate cause of death. Influenza could also go here, if in the judgment of the person who is filling it out influenza played some role. Suppose someone dies weeks or months after an influenza infection, say, from cardiac disease. Did flu cause or hasten their deaths? It’s not implausible, but the main question is untestable: would they have died at the same time if they had not had an influenza infection? So if there are any deaths of this type, it is quite likely they aren’t counted if mention on the death certificate is required.

So using the death certificate, then, you can get everything signed off as pneumonia, which would likely include influenza deaths where the actual agent wasn’t known, or those things where influenza was mentioned as the immediate, underlying or contributing cause of death. This would allow a week by week tally of all deaths in the US from pneumonia and influenza (P&I), which would include both influenza and non-influenza deaths, but also omit many deaths where influenza played part in the sense that the person wouldn’t have died if they hadn’t gotten the flu. In order to capture them you could use a wider cause of death category, say all respiratory and circulatory deaths, or even deaths from all causes. But obviously most deaths from any of these categories, including P&I, are not all influenza deaths. We are quite sure that if we only count death certificates where influenza is mentioned somewhere we will grossly underestimate the mortality from influenza. So what do we do?

There is no question that people die of flu during flu outbreaks. It is also true that mortality from all causes waxes and wanes throughout the year in a more or less regular wave-like pattern, worse in the winter than the summer. During a moderate or bad flu season, however, the regular wave will have a spike in it corresponding to a flu outbreak. Those are excess deaths that are occurring during a heavy flu season that don’t occur at other times of the year. One of the earliest approaches to estimating the mortality from flu, Serfling’s method, was to estimate what would be expected during the winter if the regular wave-like background were the only deaths and subtract that off from what is seen in a bad flu season (I am leaving out some technical details). In other words, the spike is excess mortality during a flu outbreak. We use this because when there is a recognized (by laboratory means) flu outbreak, the excess deaths during that period have a much higher probability of being the flu deaths.

This means that in a mild flu season, with no visible spike, the excess mortality would be zero by this method, even though we are pretty sure some people are dying of flu during flu season and perhaps outside of flu season as well. So we are underestimating the effect of flu. But since we don’t have any virological confirmation, maybe some of these spikes are due to other viruses. So that would over estimate the flu burden. There are also differences depending upon whether you use only underlying cause or both underlying and contributing cause and whether the measure is P&I, the broader respiratory and circulatory, or all causes. If we want to estimate the total burden of flu mortality we will use the all cause category, but that is the most imprecise and yields estimates with the most uncertainty.

There have been modifications and elaborations of Serflings method and in recent years some additional methods that make use of information about the prevalence of influenza positive cases of influenza-like illnesses provided on a weekly basis by the CDC influenza surveillance system. The new standard is due to Thompson et al. and was used by Jon Dushoff, Lone Simonsen and colleagues may be highly relevant to the current H1N1/2009 outbreak. In a paper in 2006 in the American Journal of Epidemiology they used regression analyses and subtype specific prevalence data to again make estimates of the excess mortality contributed by influenza. The results wereconsistent with the classical Serfling method, arrived at by a different means, an average excess mortality from flu in the years 1979 to 2001 of roughly 41,000. This number is not so informative, however, when we realize that the distribution of excess deaths year by year is probably multimodal, i.e., not some smooth and symmetrical bell-shaped curve with the average in the middle, but a jagged picture with several different peaks and valleys. That means that some years are far below 41,000 and others much above it, for all deaths from all causes related to influenza.

Table 2 in the Dushoff paper (p. 185) shows a very interesting tally of all estimated deaths for the two seasonal flu subtypes (H1N1, H3N2) and influenza B. Of the 41,000 deaths, H3N2 is the nastiest, contributing (annual average over the 23 years) 29,000 of the estimated 41,000 deaths. Influenza B comes next: 8500. Bringing up the rear is the seasonal flu subtype H1N1: just under 4000. This means that the seasonal flu subtype H1N1 is by far the least virulent, less than a seventh of the estimated mortality burden of the other seasonal flu subtype, H3N2, by this method. What there is about the seasonal H1N1 that makes it less virulent (or conversely, what there is about H3N2 that makes it nasty) we don’t know, but it appears that the current swine-origin H1N1 is more like its seasonal cousin. On the other hand, the 1918 virus was also H1N1.

Is there a take home lesson here? Flu viruses, even of the same subtype, can act very differently and we don’t know why. Could H1N1 change and become nasty like H3N2 or even the 1918 H1N1? No one knows. Could it reassort and pick up something from H3N2 or even H5N1 and become really nasty? No one knows. Seeing how quickly this one is spreading underlines the importance of investing in the science, so maybe we can know better when to worry and when not to worry. For the coming months, while we wait to see if the other shoe drops, investing in public health and social support systems would help us weather an assault from this bug or any other one that comes along.

Like the smartest of the Three Little Pigs we need a house built of bricks, not straw.

Leave a Reply

Your email address will not be published.

This site uses Akismet to reduce spam. Learn how your comment data is processed.