Multiplying errors

There are two ways to estimate of how many people have been infected with pandemic influenza:

1. Actually measure how many people have been infected by examining their blood for evidence of antibodies against the virus (seroprevalence study).

or

2. Use a fantasy “multiplier” to derive a number from irrelevant datasets.

In a recent study, the CDC decided to go with option number 2.

From: Reed et al. (2009) Estimates of the Prevalence of  Pandemic (H1N1) 2009, United States, April–July 2009. Emerging Infectious Diseases.

To estimate the total number of cases of pandemic (H1N1) 2009, we built a probabilistic multiplier model that adjusts the count of laboratory-confirmed cases for each of the following steps: medical care seeking (A), specimen collection (B), submission of specimens for confirmation (C), laboratory detection of pandemic (H1N1) 2009 (D), and reporting of confirmed cases (E) (Figure).

[snip]

Using this approach, between April and July 2009, we estimate that the median multiplier of reported to estimated cases was 79; that is, every reported case of pandemic (H1N1) 2009 may represent 79 total cases, with a 90% probability range of 47–148, for a median estimate of 3.0 million (range 1.8–5.7 million) symptomatic cases of pandemic (H1N1) 2009 in the United States.

Wow, 79! What data did they rely on to get this magic number? Why, phone surveys of influenza-like illness of course!

From the carefully hidden “Technical” Appendex

In May 2009, after the identification of pandemic (H1N1) 2009 in the United States, a random-digit dialed telephone survey sampled similarly to the BRFSS [Behavioral Risk Factor Surveillance Survey] was conducted using only the ILI module from the 2007 BRFSS and some limited demographic information. Respondents were adults >18 years of age living in the same 9 states where the ILI module was included during the 2007 BRFSS plus New York State. Participants were asked the same set of questions included in the ILI module during the 2007 BRFSS, including ILI in the past month, care- seeking behavior, receipt of antiviral treatment, and influenza vaccination. Participants were also asked the same questions about all members of their household. A total of 1,788 adults responded during a 3-week period.

There is really only one problem with ILI phone surveys – they are worthless. As I wrote in a previous blog: Estimates of case fatality rates based on influenza-like illness are wrong. This is because many viruses, and even allergies, result in flu-like symptoms.

So a “probabilistic multiplier model” based on worthless data is also worthless. Garbage in, garbage out.

Why doesn’t the CDC simply do a seroprevalence study? They have the reagents. I think they don’t do it for the same reason that they are withholding data from CBS about the real number of cases that occurred in the Spring, they don’t want the public to know how lethal this virus is. The “probabilistic model” vastly overestimates the number of people who were infected because it likely includes many people who were not infected with pandemic H1N1. This results in a larger denominator when calculating case fatality rate. A seroprevalence study would likely show a much lower number of people infected in the Spring. This would naturally result in a much higher case fatality rate than the CDC is publicly promoting. I say publicly because they have internal data suggesting a very high case fatality rate.

Relying on the CDC for case fatality rate information may be dangerous to your health.

Advertisements

Leave a comment

Filed under public health, Science

Leave a Reply

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

WordPress.com Logo

You are commenting using your WordPress.com 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 )

Google+ photo

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

Connecting to %s