An aerial perspective: Percent of visits with CAC codes

June 17th, 2016 / By Clarissa George

Have you ever flown over the Midwestern United States? If you have, you know that as you look down from the airplane windows you see a ton of squares with crops covering much of the land. It’s amazing how the squares are so visible from the air, but, when you drive through these states, the endless rows of corn, wheat and other plants don’t look like they are separated by anything so orderly. This aerial perspective of the Midwest allows us to see something we don’t when we are down right next to the crops and landscape. Of course, we can’t see the same details from the airplane as we can from the road, however both views are important and offer additional insights. The percentage of visits with CAC codes provides this same insight into the overall view of coding.

Percentage of visits with CAC codes is the number of visits that generated auto-suggested CAC codes out of the total number of visits. Thus, if we want to calculate a hospital’s inpatient percentage of CAC codes for May 2016, we would take the number of inpatient visits with CAC codes and divide it by the total number of inpatient visits in the month of May. In recent blog posts, I have discussed coder acceptance methods and annotations. Jason Mark has talked about precision and recall. Today I want to show how percentage of visits with CAC codes interacts and adds additional perspective to these measurements.

Before I dive more deeply into this metric, let me explain why this is important using a non-CAC system example. When I was younger, I had the opportunity to figure skate competitively. I was no Olympic figure skater, but enjoyed learning and competing in local competitions. I had several amazing coaches throughout my ice skating career and the coach that worked with me the longest (over six years) would always tell me “Clarissa, if you’re going to do something you might as well do it right.” This idea can be applied to measuring visits with CAC codes. Making sure that the engine can see as many visits as possible and auto-suggest for them is making sure that we are maximizing the engine’s opportunity to have as much of an impact as possible.

There are a number of reasons why a hospital’s percentage of visits with CAC codes could be low — these include (1) not all documents are being sent to the engine, (2) data is stored in formats that are not easily auto-coded (e.g. handwritten documents), and (3) configuration issues that prevent the engine from considering some documents and evidence. Now, if documents for some visits are handwritten, there isn’t much a CAC system can do to help with these visits. However, if not all documents are being sent to the engine, or if the data is stored in formats that are not easily auto-coded, there are some areas that can be improved to help more visits get auto-suggested codes. These include making sure correct interfaces are in place so that documents can be sent and keeping the CAC system up-to-date when changes in documentation types (often called “regioning and sectioning”) have occurred. If the engine hasn’t been updated for certain visit types since install, but your internal records and document types have changed, the engine needs to know!

Application to Precision, Recall and Coder Acceptance

How does percentage of visits with CAC codes interact with precision, recall and coder acceptance? Understanding how many visits have auto-suggested codes helps to make the other performance metrics even more insightful. Consider the two hospitals in the table below:

Clarissa CAC chart

Here we can see each hospital’s precision, recall, percentage of visits with CAC codes and the CA (coder acceptance) for CAC methods. Let’s take a look at Hospital A. Hospital A has a precision of 69 percent and a recall of 66 percent. Additionally, their percent of visits with CAC codes is 95 percent and their coder acceptance for CAC Methods is 50 percent. Since the percentage of Hospital A’s visits with CAC codes is close to 100 percent, we can see that as this hospital increases their usage of CAC methods, the impact will influence most of their records (we know they can increase because their CAC methods is lower than recall -see my blog post about coder acceptance for additional explanation). This means the increase in coder acceptance, or any other measurement, will be across their entire hospital.

Hospital B, on the other hand, has a precision of 72 percent, recall of 66 percent and a 76 percent of visits with CAC codes. Plus, their use of coder acceptance CAC Methods is 40 percent. Hospital B still has room to improve, however. Since 76 percent of their visits are already getting CAC codes, as the coders increase their use of CAC methods, it will only be on those visits with CAC codes. Thus, increasing recall and coder acceptance will still have a large impact on the overall total final billed codes, but these changes will not affect 24 percent of the visits.

When you see percentage of visits with CAC codes below whatever your organization is comfortable with, it is important to check how many documents are hand written, whether regioning and sectioning are up-to-date and if the engine is receiving all the necessary documents. Please note some organizations use more handwritten documents than others. If Hospital B in the example above has about 20 percent or more of their documents handwritten, for example, they would probably not be worried by a percentage of visits with CAC codes of 76 percent.

Understanding and applying percentage of visits with CAC codes gives us greater insight into multiple metrics, the impact of changing certain metrics and what it will look like on a holistic level.

Clarissa George is a business intelligence specialist at 3M Health Information Systems.