Flu season has started in earnest, even though it’s not “officially” flu season until week 40 (first week in October this year). How do we know it’s flu season if we don’t test everyone and can’t count flu? We use a surveillance system. The flu surveillance system has lots of moving parts and five or six components (or as many as nine, depending on how you count). None of them tell us exactly what we want and putting the different pieces together can sometimes be like the blind men and the elephant. But the system does work better than you’d think and it’s undergoing modifications and improvements as new sources of information and tools become available. So what is it telling us that allows me to say flu season has started?
Let’s just look at one component, the Outpatient Influenza-like Illness Surveillance Network (ILINet). Look at the name. These are outpatients, that is, people who aren’t in hospitals. People in the community, who, when they don’t feel well go to a “health care provider” (the phrase is revealing of the difference in medicine today compared to when I was in medical school; then we would just have said “your doctor”). The “Network” part refers to a designated set of about 2400 health care providers in all 50 states who see patients in their offices or clinics (about 1300 sites for the 2400 providers). They see these patients for all sorts of the usual outpatient illnesses (heart problems, asthma, arthritis, urinary tract infections, etc.). That’s a lot of visits over the space of a year, somewhere around 16 million. A certain number of these visits are for something formally called an “influenza-like illness” or ILI. It’s not necessarily influenza. It’s an illness that’s like influenza and it has a formal definition: temperature of 100°F [37.8°C] or greater; and a cough and/or a sore throat in the absence of a known cause other than influenza. Two things about this. A significant proportion of people infected with influenza are asymptomatic or have symptoms other than fever, so they get missed by this definition of ILI. And a lot of other viruses can cause the same symptoms, so they get counted as ILI even though they aren’t influenza.
Throughout the year there is a steady baseline drumbeat of the proportion of visits to health care providers for ILI, but the proportion shoots up during “flu season” (weeks 40 of one calendar year to week 20 of the next year). We know from many studies that when the proportion of visits for ILI shoots up during what CDC has designated the flu season, a lot of the reason (but not the entire reason) is an upsurge in influenza, a seasonal disease very prevalent during those months. There is influenza infection outside the official flu season window, too, although until very recently we haven’t been looking very hard at how much. The assumption has been that however much flu might be there, it is at a low level. As we begin to look harder throughout the year we’ll probably find there was a lot more than we thought. But the surveillance system works by assuming that most of the ILI outside flu season isn’t flu, and if the object is to find out when we have a sudden increase in flu, it still works as a surveillance indicator.
Sudden increase over what? Increase over the baseline for the nation or for a region (more on that shortly). The baseline for the nation as a whole and for each of the ten different CDC geographic regions is determined by taking an average for the two previous seasons during the months outside of flu season for each week. That’s not quite all there is to it. The number for each week and region bounces around quite a lot during this time period so you get a whole distribution of percentages of ILI. It might look something like this, although for each week there are two points (one for each of the previous two “non-flu” seasons; this illustration wasn’t specifically made to illustrate flu surveillance but one I picked off of Wikipedia to illustrate the concept of the spread of proportions one might get):
Now scrunch all the points up (say push them all to the right). You’ll get a pile of points, with most in the middle around the average, trailing off to fewer on the low end and the high end. For simplicity, think of that distribution as being like your standard “bell shaped curve” (also called the normal or Gaussian distribution):
The center (peak) of the curve is the “average” proportion of ILI’s seen by the providers during the non-flu part of the year, but there is another measure, one that determines how peaked the bell is. It is a measure of spread of the distribution called the standard deviation. For a normal distribution, one standard deviation from the center on either side always includes a little more than 68% of the values and two standard deviations a bit more than 95%. So the baseline is set at two standard deviations above the average. You can see in the graph above that would include almost 98% of the values in the “off season” distribution. So the baseline is set at the upper 2% of what is seen throughout the non flu weeks. By doing it in this way, when the proportion of ILI visits exceeds the baseline it is higher than almost any ILI proportion we’d see during the non-flu season (except for maybe 2% or so of the time). Since under this probability model, it could happen a couple of times out of every hundred non-flu weeks that the proportion got as high as that baseline, CDC doesn’t declare a flu season to have started until the baseline is exceeded three weeks in a row. So it’s a pretty stringent standard.
There is one more complication in all this. Influenza is a notoriously patchy and uneven disease geographically and temporally for reasons we don’t understand. So in addition to having one big summary baseline for the nation, CDC also calculates separate baselines for each of the ten regions. They can be very different. For example in Region 1 (CT, ME, MA, NH, RI, VT) the baseline figure is 1.5% while for Region 4 (AL, FL, GA, KY, MS, NC, SC, TN) it is 2.2%. The national baseline is 2.4%.
With all of that background we can now see that ILINet is telling us flu season has started:
The dotted (horizontal) line is the national baseline and you can see two things. We never sank down to the proportion of visits during the spring and summer this year that we did in the previous two seasons (and one of those seasons, 2007 – 2008 was a particularly bad flu year). That’s swine flu, or at least it’s the increase in visits for ILI that coincided with the advent and extensive media coverage of swine flu. It’s possible that the public concern led many people with an ILI to seek medical care that wouldn’t have in previous years, so that’s something to keep in mind. But look at what’s happening starting in week 34 when we hit the baseline earlier than we ever have, and then shot above it for three successive weeks. It’s now at 4.6% of visits to the ILINet providers, almost twice the baseline (which itself is the upper 2% of the off season proportion).
But that’s the picture for the nation as a whole. If you look at the individual regions (which you can do by clicking on the link below the graph at the CDC site which is labeled “View ILINet Regional Charts“) you will get a map of the US showing the regions. If you click on any one of them you will get the same kind of chart we showed above for the nation but now confined just to the region. Let’s do this for Region 1 and Region 4. Here’s Region 1:
Region 1 (New England) has been below baseline and is barely making it up to baseline even now. But look at Region 4 (South Atlantic) during the same time period:
This is a clear example of the patchiness of flu in time and space. New England will likely quickly catch up to Region 4, but by the time that happens, Region 4 may already be on the downswing.
Or maybe not. Any prediction about what flu is going to do is hazardous. And this is only one view of the elephant. There are other arts of the surveillance system (deaths from pneumonia and influenza in 122 cities or pediatric hospitalizations for lab confirmed flu in selected localities) that reveal other parts of the story. They don’t all move together. So far pneumonia and influenza deaths are not above what is expected at this time of year. That measure usually comes after ILINet starts to register a signal. It is also measuring something different, the really severe cases, while ILINet looks at community cases (outpatients).
Newspaper reports, official press releases and blog posts are all telling us little bits and pieces of a very complex picture. If it seems that the public is confused or that news reports are contradictory it’s because the situation itself is confusing and contradictory.