More Than Meets the Eye: The Importance of Risk Adjustment

Share on FacebookTweet about this on TwitterPin on PinterestShare on Google+Share on LinkedInEmail this to someone

mpirica-risk-adjustment-apples-to-orangesAs a self professed data geek, I have a special love for brain teasers. So I’d like to start today’s post with this one:

If two hospitals had the same number of patients undergoing a surgery this year, and 5 people died during surgery in Hospital A, while 10 people died in Hospital B, which hospital should you choose for your next surgery? (This is not a trick question.)

If you answer Hospital A, you are not alone. A random, not-remotely-statistically-significant, sampling of what many would consider “smart” people (mostly engineers and scientists) chose Hospital A 12 out of 15 times. Seems logical to choose the hospital that killed the fewest people, but were they right?

IT DEPENDS.

The Importance of Context

Consider this: if you knew that Hospital A is a specialty hospital with a really healthy patient population and Hospital B is a trauma center that served the sickest patients, would that change your mind?

Whether it’s a doctor or a hospital, when you evaluate a provider’s outcomes, the first thing you should understand is the composition of the patient population. Because you don’t want to compare apples with oranges, you need to make sure that a death, readmit or complication (collectively refer to as “adverse events”) from one hospital is comparable to another. That task is non-trivial; in other words, it’s really really hard for most people, and pretty hard even for statisticians.

Risk Adjustment

To deal with the problem of comparing apples and oranges, statisticians adjust results for risks. In the context of healthcare quality, “risk” is any condition that could cause bad outcomes that is beyond the control of the physician. There are different approaches to risk adjustment, and not all are created equal.

The simplest way to account for risk is to look at patients’ health profiles and assign correction factors, then adjust quality score either by weighing differently or adding/subtracting points. While that is a perfectly valid way to take risk into account, MPIRICA Quality Scores incorporate risk adjustments through the use of a computational statistical model. This might sound like magic because it’s so complicated, but I assure you it’s not. I will talk more about what a model is in the next section.

We felt strongly that a doctor or a hospital should not be punished for taking sicker patients or more complex cases, and we wanted to provide the most rigorous risk adjustment possible.

The Nuts and Bolts

Without going into a lot of technical mumbo jumbo that will put everyone to sleep, here’s what we do: a team of physicians identify over 500 risk factors (including the patient’s demographic information, health state such as diagnostic results or /pre-existing conditions, and procedural history) that reflect the patient’s profile. Then statisticians, guided by physicians, create models that reflects medical reality. The process of modeling is simply a shorthand way of saying that we take all the information we have about patients, feed them into a computer to make a number-based picture of reality.

The computer then points out any problems that patients might experience based on the picture we have. These predictions of problems are the “expected outcomes,” and when you compare that to the actual outcome, you can tell whether a doctor or hospital is doing awesome or awful.

A Picture is Worth A Thousand Words (Well, Maybe Only 175)

Example of risk adjustment - MPIRICA Health

Example of risk adjustment – MPIRICA Health

To understand how this plays out in the real world, let’s go back to our brain teaser situation and add an illustration. Back to Hospital A, which had a lot of healthy patients (shown in grey); the model predicts that they should see 3 deaths, but they ended up with 5, which is 2 MORE than expected. By comparison, Hospital B had a lot of sick patients (shown in blue); the model predicted 15 deaths, but they only had 10, which is 5 LESS than expected.

In this context, it’s clear that Hospital B is the higher performing hospital. This goes to show, if you just have the raw number of complications or deaths, it doesn’t give you the full picture, and doesn’t actually help you make good decisions.

That’s why the MPIRICA Quality Score is so important. We take into account all the critical information that you don’t even know to ask, and show you a single, simple FICO-like score. So now you can make your major medical decisions with ease and confidence.

To learn more about how MPIRICA creates our Quality Scores, feel free to visit our Quality Score summary.