As someone who does data visualization for the Infection Prevention office at a large academic medical center, I deal with a lot of infection data. One of the main types of infections that I work with is Surgical Site Infections. Please note that all data shown below is faked due to confidentiality issues.
Surgical Site Infections are infections that are associated with surgical procedures. To maybe state the obvious, they’re very bad. My department works throughout our hospital system to help clinical staff reduce infection rates by applying best practices from the medical literature.
Let’s start simple
To start our analysis, let’s just consider a single type of surgery. A basic way to display the data is to show the number of infections over time. This is a good way to see how we are doing compared to the past – are we getting better, or are we getting worse?
It starts to get a little more interesting when we compare different types of surgeries to each other. To do this, I created a small multiple in Tableau. This view allows our epidemiologists to see at a glance where the biggest issues are occurring. You can see the interactive dashboard here. I’ll delve more deeply into the technical setup in a later blog entry.
From numbers to rates
The issue with showing the number of infections is that it doesn’t account for the number of surgeries done. If you had two infections, it makes a difference if that was out of five surgeries or five hundred! To account for this issue, I created a second small multiple to show infection rate (per 100 surgeries).
From rates to risk-adjusted ratios
This is starting to offer some pretty good insights! One issue, however, is that not all types of surgeries have the same risk for infection. For instance, a colon surgery is inherently riskier than a hip surgery, simply because a colon is dirtier than a hip. To account for this difference, it’s helpful to use a risk-adjusted metric. Luckily, the CDC (Centers for Disease Control and Prevention) developed the Standardized Infection Ratio (SIR) for just this purpose. Basically, they assess the risk of each individual surgery, which lets you compare your predicted number of infections to your observed number of infections. SIR = sum(observed) / sum(predicted). If your SIR is under 1, you’re doing well. It it’s over 1, not so much.
In order to have a valid SIR, there must be at least 1.0 predicted infections. This helps reduce meaningless extreme values. For instance, if there are 0.1 predicted infections and you have 1 infection, your SIR would be 10. This seems too harsh, and so anything with too small of a denominator is excluded from the SIR. Because of this, however, it is problematic to report SIRs at the quarterly level for surgery types that have small denominators. For this reason, we display our Surgical Infection SIRs by year instead of by quarter.
From risk-adjusted ratios back to numbers, sort of
The SIR is a great metric, but it’s a bit abstract. Some clinical staff aren’t very comfortable with data (it’s okay, you wouldn’t want me to put an IV in!), so keeping things relatively straightforward is generally a good idea. At the same time, I don’t want to completely throw out the SIR we just built. A happy compromise (for us, at least) has turned out to be showing the number of infections, with the number of predicted infections as a benchmark.
This lets you compare your performance without losing track of what we’re really talking about – people. Too much abstraction makes people forget that at the end of the day each of these data points represents a person with thoughts, feelings, and desires. Plus, it let me build a ‘living legend,’ which was a fun exercise!
Click here to see the dashboard in action! I’ll provide a more detailed technical breakdown of how I actually did this in a later post.