And If So, What’s that Worth?
No one has the ability to capture and analyze data from the future. However, there is a way to predict the future using data from the past. With the evolution of big data in other industries, healthcare has embraced the concept of using patterns of behavior to predict future outcomes. Patterns of payors’ denials can allow us to better predict when something occurs that will trigger a denial, so that it can be “corrected” before claim submission.
Next, we have to ask the two most important questions that any manager must be able to answer, about reliability and effectiveness:
- How reliable are the predictions? Are they accurate and timely; and
- What is the inherent efficiency to be achieved, and the cost to collect on that efficiency?
[Another way to asking that second question is, does this save us money or just cost money, and what’s the Return on Investment (ROI) if we invest in such a program?]
HUGE questions, right? If you go to your CEO with a proposal to implement such a program, you can bet s/he will ask at least those two questions, so you better have answers!
With those questions in mind then, let’s look at what we’re talking about.
What Can Predictive Analytics Do for Us?
Predictive analytics are a source of many healthcare initiatives. Consider population health management and the need to deploy resources in the most cost efficient fashion to gain preventative benefits and avoid high cost admissions. We see the same analytics come forth in a not so beneficial fashion with the government and commercial payor claims databases that scrub hundreds of thousands of claims for patterns of errors that a MAC or RAC or other auditing entity might pursue. With the advent of these changes for the payors, facilities and providers are also beginning to utilize software analytics and analytic firms to prepare models to “predict” their weaknesses and deploy resources in the most efficient way to achieve the highest efficiency.
In the past couple of years, predicting hospital readmissions has been a huge topic, and appears to be only gathering speed. The growth of articles and publications about it is a good barometer.
Source: PDF from HealthCatalyst.com
The underlying reason for this interest is pretty obvious: providers have two primary motivations – (1) improving outcomes, and (2) avoiding financial penalties by the insurers, as spearheaded by our government.
Adopting the use of predictive analytics is, however, not a trivial matter.
How Can You Put Predictive Analytics to Use at Your Site?
Consider this list of what you would need to start, grow and run a Predictive Analytics group at your campus:
- Learn how to use data more effectively and glean valuable analytical insights,
- Manage and coordinate data, people, and technology at an enterprise level,
- Understand and support what analytical leaders do,
- Evaluate and choose realistic targets for analytical activity, and
- Recruit, hire, and manage analysts.
Ok, so you can start breathing again, now… but it should be obvious that no one can take this on lightly, and it would need major budgetary support, which of course means the C-Suite would have to be completely sold on the value of it and how it could have tangible (meaning financial) benefits.
Let’s first consider the costs of acquiring the output of predictive analytics so you can garner the benefits.
Buy, Rent or Build-Your-Own?
There are three methods of developing a new process (for a business model, a new product line, or a new revenue stream) to add to a company: (1) buy something off-the-shelf (already made) that mostly fits what you want to do; (2) modify an existing system/product and add some enhancement and/or value to it; (3) build your own system from scratch. In my experience, we would start with (1), then move on to (2), and then, whatever made the most sense, either kill it, stick with (2), or move on to (3).
By far, the least expensive way to begin is to “buy” or license an off-the-shelf product that will do what you want, or something close.
So, the question is, what do you—as a manager, not an analyst—really need to know in order to interpret provided results, make better decisions, and derive tangible benefits? Also, because someone may ask you to give a simplified answer to another question, you must know this: how do these data scientists do what they do? (Not to worry, it’s not that difficult to answer.) You only need to understand a few key factors to be able to adequately explain and communicate the results and recommendations from predictive analytics.
What Do We Really Need to Know About It?
Quantitative analysis isn’t magic—but it is normally done with a lot of past data, a little statistical wizardry, and some important assumptions. Let’s talk about each of these: Data, Statistics, and Assumptions. Basically, we’ll use those three elements to look for patterns among random data collected and use those patterns to predict the likelihood that a future pattern will also occur.
Inability to gain transparency into operations through data is the most common reason seeking to employ predictive analytics. Hospitals, however, have an incredible asset for predictive analytics using their claims and charge data. Decision support teams within the enterprise (a system or facility) can then massage the data into meaningful insight for the facility.
Regression analysis in its various forms is the primary tool that organizations use for predictive analytics. The analyst hypothesizes that a set of independent variables (say, gender, payer, visits to ED, many more) are statistically correlated with claims denied by some payer for a sample of patients. The analyst performs a regression analysis to demonstrate patterns between random variables to provide transparency into processes. While this usually requires some iteration to find the right combination of variables and the best model. Let’s say that the analyst succeeds and finds that each variable in the model is important in explaining the payments or denials, and together the variables explain a lot of variation in the payer’s history. Using that regression equation, the analyst can then use the regression coefficients—the degree to which each variable affects the payer’s behavior—to create a score predicting the likelihood of denial.
Voila! You have created a predictive model for other patients who weren’t in the sample. All you have to do is compute their score, and flag their claims if their score exceeds a certain level. It’s quite likely that the high scoring claims will be denied—assuming the analyst did the statistical work well and that the data were of good quality.
Now, this is an important point! This does NOT mean you have necessarily identified a billing error! Often, but this is not always the case, all you have likely done is predicted the RISK of some payer denying the claim, based on history with that payer. I’m hedging here a little because you may have discovered an error in billing or coding that needs to be fixed, but technically, the statistical analysis may not say anything about the claim being improperly coded or billed or documented. Therefore, you need to talk with your legal advisor(s) about how to handle such a determination, in order to avoid a “False Claims Act” violation. While you might want to revisit the coding, billing or documentation, this score rather assumes that the claim is documented, coded and billed appropriately. Is that an acceptable assumption? That brings us to the third key factor of these models, Assumptions.
The third key factor in any predictive model, and perhaps the most dangerous—is the set of assumptions that underlie it. Every model has them, and it’s important to know what they are and monitor whether they are still true. The big assumption in predictive analytics is that the future will continue to be like the past. Sometimes, however, as all my readers know, the rules change, and/or the policies of the payers change. If there is one thing we can count on in our industry, it’s that the payment policies will change, with or without notice, and sometimes even retroactively. So, the models that are used to predict denials may no longer be valid, month to month, much less year to year.
What makes assumptions invalid? The most common reason is time. If a model was created even a single year ago, it may no longer accurately predict current payer behavior. In many industries, the greater the elapsed time, the more likely someone’s behavior has changed. In our industry, times frames are arguably much shorter.
Another reason a predictive model’s assumptions may no longer be valid is if the analyst didn’t include a key variable in the model, and that variable has changed substantially since the model was created. A great—and scary—example, from outside our industry, is the financial crisis of 2008-9, caused largely by invalid models predicting how likely mortgage customers were to repay their loans. The models didn’t include the possibility that housing prices might stop rising, and even that they might fall. When they did start falling, it turned out that the models became poor predictors of mortgage repayment. In essence, the fact that housing prices would always rise was a hidden assumption in the models. Anyway, you can see how such assumptions might also be hidden in a model about payer denials.
Since faulty or obsolete assumptions can clearly bring down whole banks and even (nearly!) whole economies, it’s pretty important that they be carefully examined. Managers should always ask analysts what the key assumptions are, and what would have to happen for them to no longer be valid. Both managers and analysts should continually monitor the world to see if key factors involved in assumptions might have changed over time.
Who You Gonna Call?
So, let’s get back to what you can do as a manager to acquire and use predictive analytics, in particular to try to predict denials, even denials by specific payers. As I said above, you probably want to start out just licensing some kind of tool from some vendor on a short time frame – sort of a proof of concept trial run, so to speak, if you can get one. Before you pick one to use, there are some questions I would recommend you ask, to get a feel for how their tool or system works.
Because of the relatively short time frames we work with in healthcare reimbursement, the underlying assumptions are perhaps not the most complicated or numerous, but if I were considering hiring someone to do such analyses, I’d at least want to ask them some questions, but you can always just watch for the answers to these questions while you watch a demonstration or listen to their pitch on what they offer.
Key Considerations to Pick a Tool or Vendor
Remember, we just went over the three keys, Data, Statistics and Assumptions. So, here is what I’d be thinking about and/or asking:
- Can you tell me something about the source of data you use in your analyses?
— Be wary of sources that cannot be documented or re-created…
- Are you sure the sample data are representative of the population, my population?
— Was the universe of random elements included ?
- Are there any outliers in your data distribution?
— How did they affect the results?
- How many measures do you track/consider?
— What are the top two or three influencers?
- Is there some benefit to me tracking just those two or three?
— This might allow you to “stick your toe in the water” to see what you can learn…
- What is the benefit of being able to track all the measures that you track?
— There must be a reason to do all that, right?
- What assumptions are behind your analysis?
— Think of what assumptions you make yourself…
- Are there any conditions that would make your assumptions invalid?
— Healthcare reform is a moving target, so think of what changes could happen…
- Can you tell me the ROI I might be able to expect, using your analyses/predictions?
— This is a really tough question, but there should be some kind of answer available…
- In your experience, are there other uses of these analyses, outside denial prediction?
— For example, contract negotiations, etc.
I’d be very, very confident to go before a CEO if I knew the answers to these questions beforehand. I’d want to walk in with all these answers, especially for 6, 9 and 10. Additionally, ROI is a relative matter and like many other business opportunities, must be vetted and tested.
Vendors are typically prone to show two things about such solution(s):
- Show that if the software or model is not used, some adverse event will happen to the client (which is often true), and
- Show ROI that is based on a data sample, to prove the value of the solution. (Again, often true.)
While some observers may complain that the sample data used for such displays is not “complete” or is “cherry-picked” to show the solution in the best light, one should keep in mind that a model is, by definition, a simplified version of reality, by necessity. No model can or even should reflect reality completely. If it does, then less can be learned, assuming it can even be understood in the first place. Indeed, the reason for a model is to simplify a system, in order to see what makes it work. So don’t expect a sample to be “reality.”
The best advice is to examine a vendor or even your own staff’s claims with a critical eye – meaning, don’t gloss over the key factors. Ask questions. And remember, if it looks too good to be true, it likely is. Ultimately you must show your CEO not how the solution will answer an analytic need, but how the solution, using the current staffing and skill sets, will provide transparency about what’s really happening in your operations, and therefore, your revenue integrity process.
So can we make denial predictions in a profitable way? Absolutely, with a little work. Personally, I’ve been using predictive analytics ever since I was working on my Bachelor’s degree in Archaeology, which was mostly about studying and analyzing the past. (Long story.) But, it’s still pretty amazing to me that we can even use analytics to predict the future. All we need is the right data, the right statistical model, and to be careful about our assumptions. And the right tools.
As I said at the outset, no one can see and capture data from the future, but we can easily show the likely events that will come to pass in the near future, in at least some processes. The likelihood of payments and denials is one of those processes, and it is readily available to revenue cycle professionals.
Join us for an informative hour on the February 12th 2016 edition of Finally Friday! when we look at how a Certified Medical Auditor would use predictive analytics in a provider campus to predict and fix denials before they even occur.