March 11, 2015|Categories: Analytics and Research
The 2014-2015 flu season has been particularly severe, as indicated by data from the Centers for Disease Control and Prevention (CDC) and the athenahealth network. Tracking more than 20,000 primary care providers on athenahealth’s cloud-based network, national influenza-like illness (ILI) rates peaked at 4.5% of patient visits for the week ending Saturday, December 27, 2014, a level that’s on par with the severe 2012-2013 season. According to the CDC, this is mostly due to the predominance a mutated H3N2 strain of the flu virus, which rendered this flu season’s vaccine less effective.
As a result of the mutation, the CDC posted a health advisory on December 3, recommending that doctors prescribe antiviral medications to help protect high-risk patients. What effect did this guidance have on doctors’ prescribing activity? The athenaResearch team tapped into our network of providers to see.
At first glance, it looks as if the CDC advisory may have had an effect. Figure 1 shows the proportion of patients seen November 1, 2014 – January 15, 2015 who were diagnosed with the flu and given an antiviral prescription. There was a higher prescription rate almost every day following the December 3 CDC advisory. However, the prescription rate began increasing even before the announcement, suggesting that there may have been something else driving the change.
On closer inspection, the situation appears more complex. Figure 2 looks at the relationship between the prevalence of ILI and the likelihood of antiviral prescriptions. As shown in the “ILI Rate” graph, the CDC advisory was issued as the prevalence of ILI in physicians’ offices was already increasing. The “Rx Rate for ILI Visits” graph shows that as ILI became more common, physicians were more likely to prescribe antivirals to patients with diagnoses of ILI – even before the CDC advisory. Thus, the increase in the likelihood of ILI prescriptions may have resulted, in part, from higher ILI prevalence rather than just the advisory.
To get a more precise view of the impact of the CDC health advisory on antiviral prescribing patterns, it is important to distinguish between the effect of ILI rates and the effect of the advisory. We have attempted to shed light on this issue in Figure 3.
Each point represents a week during one of the past three flu seasons; the vertical axis indicates the prescription rate for that week, while the horizontal axis indicates the ILI rate.
The first thing worth noting: The 2012-2013 flu season (red dots) had lower prescription rates for any given ILI rate than the two more recent seasons, which look similar to one another. Therefore, in any analysis, we should treat the 2012-2013 flu season separately from the rest of the data.
The two curves in Figure 3 are statistical models of the relationship between ILI rate and Rx rate for ILI patient visits (a linear-log model) and tell us what prescription rate we could expect based on the ILI rate. The upper line uses the pre-announcement data (blue and green points) to find this relationship. How does this help us understand the effect of the CDC announcement? We can see whether doctors prescribed more antivirals after the CDC advisory then we would predict based on that relationship.
Specifically, the purple crosses (which represent the weeks following the CDC advisory) should be above the curve if the CDC’s guidance led to higher prescription rates. However, it is not clear that the post-CDC observations show a different trend from the pre-announcement points – some are above the line, some below.
Did the CDC’s advisory have an effect on antiviral prescribing behavior? In our opinion, there is no clear evidence that it did. As previously mentioned, the increased prescription volume could be attributed to increased ILI rates after the announcement.
We believe that this analysis is a good example of how the impact of public health advisories can be measured and improved with the use of cloud-based data. We will look for opportunities to conduct similar analyses and welcome your suggestions. We can be reached at firstname.lastname@example.org, or you can direct questions to my colleagues @IyueSung and @JoshGray_HIT on Twitter.