Estimates of case fatality rates based on influenza-like illness are wrong

Case fatality rate (CFR) is calculated by dividing the number of people who have died of a disease by the total number infected. How do you determine how many people have been infected? There are several ways to do this. However, only one method is being used in most estimates of CFR for the new H1N1 virus – surveys of people with “influenza-like” symptoms.

Here is one example that appeared prominently in the media recently:

From Reuters, September 16, 2009

The death rate from the pandemic H1N1 swine flu is likely lower than earlier estimates, an expert in infectious diseases said on Wednesday.

New estimates suggest that the death rate compares to a moderate year of seasonal influenza, said Dr Marc Lipsitch of Harvard University.


Lipsitch took information from around the world on how many people had reported they had influenza-like illness, which may or may not actually be influenza; government reports of actual hospitalizations and confirmed deaths.

He came up with a range of mortality from swine flu ranging from 0.007 percent to 0.045 percent.

Although Dr. Lipsitch does not provide the details of his approach, similar types of studies involved calling people up and asking if they had symptoms associated with flu: fever, coughing, etc. Since these surveys are intended to detect mild cases, they typically have a very low bar for what is considered “influenza-like”.

It would appear obvious that people with mild respiratory symptoms could have many other conditions. However, it would be helpful to be able to quantify this. A study conducted in Sweden allows us to do exactly that.


Follin et al. (2009) A variety of respiratory viruses found in symptomatic travellers returning from countries with ongoing spread of the new influenza A(H1N1)v virus strain. Euro Surveill. 14:. pii: 19242.

This report includes samples of patients who, during the period from 24 April to 10 June 2009 presented with influenza-like symptoms and a history of recent travel to the United States or Mexico, and therefore were recommended for examination and sampling.


In total, samples from 79 patients were tested (42 males, 37 females; median age 30 years, range 1-75 years), with between 10 and 16 samples on average each week and most of them taken from patients with respiratory symptoms and a history of recent travel to North America (Figure). Four cases with the new influenza A (H1N1)v variant were diagnosed. Interestingly, in 56% of the cases, other aetiologies were identified (Table 2).

The most common finding was rhinovirus, observed in 28 of 82 cases (34%) and three of these patients also had a second viral infection (enterovirus, metapneumovirus and adenovirus). The frequent identification of rhinovirus and other viruses demonstrates that the criteria for suspected cases of influenza A(H1N1)v are relevant as indicators of a viral infection, but not specific for influenza A.

In this study, only 4 out of 79 patients with influenza-like illness (ILI) actually had been infected with the new H1N1. That means that 95% of the ILI cases did *not* have H1N1. Note, many of these people did have something, rhinovirus (the common cold) being the most common. They just didn’t have H1N1.

If we “correct” Dr. Lipsitch’s figures by multiplying them by 20, we get CFRs of 0.14% t0 0.9%, numbers much closer to the CFRs obtained when real-time PCR is used to determine whether or not people are infected.

Proponents of symptom surveys could argue that they could correct their estimates with some sort of fudge factor. However, as the Swedish study indicates, any correction would have to be very large. To further complicate matters, the many different viruses that cause respiratory symptoms may have their own seasons and vary in unpredictable ways. Finally, allergies oftentimes cause respiratory symptoms. Since these vary by individual, by month, and by geographical region, it would be nearly impossible to control for this variable. To summarise, the number of variables that can confound symptom survey studies renders them close to useless, especially when trying to identify people with “mild” symptoms.

There are other methods of determining whether people have been infected with the new H1N1 virus. Real-time PCR, the method used in the Swedish study, can be done on hundreds of samples at a time. Antibodies can be used in seroprevalence studies. Both of these approaches are far more accurate than symptom surveys.

Which raises a question:

Why is the only method being used to estimate CFR by the public health establishment the one that is deeply flawed and likely to give grossly overoptimistic estimates of the outcome of this pandemic?


Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )


Connecting to %s