Reducing Readmissions, Part 3: A Dashboard for Forecasting 30-Day Hospital Readmissions
Forecasting 30-day hospital readmissions on a proactive basis is the first step toward reducing the potential harm, unnecessary expense and potential financial penalties associated with re-hospitalization.
In a previous post, we shared information on identifying key data features impacting business intelligence and the importance of reducing data dimensionality. We also shared insight into how a Care Transition System developed by AAJ Technologies is being used to manage transitions between patient care settings.
Forecasting 30-Day Hospital Readmissions
In this post, Kevin Oppenheimer of KGO Consulting shares his expertise on building a dashboard to accurately forecast 30-day hospital readmissions:
Before we look at the dashboard, I just want to talk really quick about the data. First, this is not real patient data. This data was generated for the purposes of this demo. In the real world, you get this data from a number of potential sources: you can get them from 837 claim files, from HL7 integrations or even directly from your EHRs reporting system. KGO Consulting is currently integrating data from all three of these sources for multiple clients. For the purposes of this demo, we will limit the data to the seven procedures and DRGs that CMS looks at for 30-day re-admission outcomes.
A Dashboard for Forecasting 30-day Hospital Readmissions
Now let’s take a look at the dashboard. Our sample dashboard tells us a small piece of the story about readmissions.
First thing you’ll notice is that use of cell highlighting. It really draws your eyes to potential issues. It’s a real time-saver and it helps provide a guided approach to reporting. This dashboard has over 200 numbers on it and thanks to the use of cell highlighting, we can easily see where potential exists.
I’ll take a look at the individual charts. First chart here in the top official by Attending Physician as A Percent of Total Admission. Here we show each row has a different attending physician and we slice that data by initial condition or DRG. We can look at this and we can see that Dr. Tim Green has 4.3 percent of his patients that initially came in with CABG are re-admitted. Now when looking at this chart alone, we can see some information that might be useful. We can see the patients that initially came in with CABG are actually re-admitted more often and we can also see that the patients under Dr. Tim Green’s care are re-admitted more often than patients under his peer’s care.
Dashboards as a Precursor to Improving Care Transition
Let’s look at the second chart at the bottom left. This one shows Readmissions by Discharge Disposition as a Percent of Total Readmissions. So here we have a row for each of the discharges, each of the locations that patients are discharged to, and they’re grouped by their facility type. Again, we slice by the initial condition and we can immediately see by looking at this dashboard the patients that are discharged to Skilled Care One – a skilled nursing facility – are re-admitted more often than any other location that they’re discharged to. So, at this point, it would be time to go look and see why that might be. If we can improve the handoff to “Skilled Care 1” or potentially help them improve their care.
The third chart at the top right shows the Initial Condition Versus the Re-admit Condition. So, this is a percentage of total readmissions which is different than the other charts. So, we if we look at this, we can see that patients that initially came in with COPD when re-admitted 18.74% of the time are re-admitted with a heart attack. And patients that were initially were admitted with heart failure when re-admitted 17.92% of the time come back with pneumonia. So, this is kind of giving us an idea of what we need to look for when we’re discharging.
Fourth chart at the bottom right shows A Trend Over Time of Readmissions. So, the goal here would be to see all of these lines going downward and as care improves, then these lines will go down because readmissions will go down. But what we can deduce from this chart right now is that we can see that CABG generally has a higher percentage of readmissions than the other conditions. We can also see that here. So, because this was done in Power BI and it could also the same can be true for Tableau or any other most other dashboard tools. Here I can go in and I can click on “Dr. Tim Green” and it will actually filter the entire dashboard. Or I can click on “Skill Care 1” and filter the entire dashboard to that facility. Or I can control click on “Dr. Kimberly Yellow” and it filters it to both.
Forecasting 30-Day Hospital Readmissions via Self-Service BI
Now I’d like to show you one example of where self-service BI is heading. I’m going to edit this report. We’re going to go to a blank page and instead of creating calculations and dragging columns, I’m actually just going to ask it a question. I’m going to say” “Show me total re-admit counts by attending physician and by discharge location class where admit condition is pneumonia.” And it rather easily created a chart for me to show me exactly what I want to see. I didn’t have to do much work at all. This may be just a starting point or it may be exactly what you’re looking for. Either way we’re definitely making things easier. Let’s go back to the dashboard for just a moment. So, we can see here this dashboard that Dr. Tim Green definitely has a higher percentage of readmissions compared to his peers. But what’s important to remember is that even though he’s correlated with more readmissions, it doesn’t mean he’s the actual cause of them.
Correlation Does Not Mean Causation
For example, he may have a specialty that the other attending physicians don’t have; and so consequently certain patients are assigned always to him and those patients may actually have a greater chance of re-admission for some other reason. It’s just important to remember the correlation does not mean causation. Now all the charts on this dashboard show only two dimensions each. To do further analysis, you may need to look at other dimensions such as gender or age or any number of others. Every time you add a dimension you’re adding a level of complexity. When you go from two to three it gets a little harder to analyze. When you go from three to four it gets significantly harder. And it just continues from there. And this is where machine learning comes in and can really help.
See how Andrew Satz describes how Correlation Does Not Mean Causation in this post.
More About Eliminating or Reducing Readmissions
This post recaps the third of four presentations from our webinar on “Reducing Readmissions with BI & Analytics.” In the next and final post, we’ll be sharing information presented by Michele Russell of the Russell Consulting Group about different data types, and the best way to store them for secure, but easy accessibility.
For more information on reducing potential harm, unnecessary expense and financial penalties associated with 30-Day Hospital Readmissions, reach out to AAJ Technologies using the form below. We will share information on how your company can leverage BI and analytics to reduce unnecessary hospital readmissions.