Lately I have seen a great deal of discussion concerning whether extrapolation, when used to determine overpayment amounts, is of benefit or detriment to any healthcare organization.
The alternative, from the payer’s perspective, is to conduct a 100 percent audit. What does that mean? It means that instead of using a random sample and gauging what likely would have happened if all claims were reviewed, all claims are, in fact, reviewed. That sounds like an overwhelming task, and it is. Imagine this: you have a practice that has filed 10,000 claims during the last year, and each claim represents an average of 2.2 claim lines (meaning that in a 100 percent audit, the payer would have to review 22,000 individual procedures to determine if those 10,000 claims were paid appropriately).
The Payer Perspective
Let’s say that an auditor can handle five claim lines per hour (not for coding, but for the purpose of the audit). That equals 4,400 hours of work, or the equivalent of just more than one year’s worth of effort by a full-time equivalent employee. Obviously, an auditor would want the audit completed long before that, so let’s say that they budgeted three months for completion. That would require eight full-time reviewers. Let’s say that each reviewer earns $45,000 per year, and now we are talking about the payer investing $360,000 to conduct an audit that may end up returning significantly less than that on the initial investment.
The Provider Perspective
Let’s look at this from the provider’s perspective. At first glance it seems like a better deal for the provider and a worse deal for the payer. In fact, on many occasions I have advised a practice to go for a 100 percent review because I knew (based on workload) that a payer instead would opt out for a real-time assessment – meaning it would estimate overpayments only on audited claims. But what happens if a payer does agree, which has happened in the past? Think about the workload involved for the provider, having to pull and make copies of the documentation required to support 10,000 claims. This includes basic charts, post-operative notes, lab and imaging studies, prescriptions, etc. This can be an overwhelming task. So why would a provider ever decide to go this route? It’s a good question and it deserves a thoughtful answer, so let’s take a look at what I call the “audit ROI.”
The Audit ROI
First I want to comment on whether extrapolation is ever appropriate. There are two opposing schools of thought on this. In the “world according to Frank,” extrapolation, if used properly, is a very appropriate and often very effective method for inferring an outcome for a larger set of data. It is often the only option available, particularly when studying an entire universe of data is impractical. Consider, for example, political polling. Let’s say there are two candidates running against each other in an election and I want to know which is likely to win; suppose this is a national election, so it involves the entire voting population of the United States, or about 207 million people. Now, we could just go out and try to survey all 207 million registered voters, but for obvious reasons that would be impractical. Even if you tried to do this, you always would end up missing someone somewhere, or some people would not give you an honest response. In any event, you might get very close, but could never achieve 100 percent certainty. Another option, however, is to poll a random sample of the voting population and then estimate, using their responses, how the entire voting population would vote. This is extrapolation.
One of the main issues that arise when conducting an extrapolation analysis is the idea of sample error, which is a bit more than we can address in this article. In general, however, this means that for every sample I take representing a certain group of people, the overall ratio of how many of them would vote for a certain candidate would vary – so instead of predicting an actual value, I need to predict a range, accounting for that potential variation. For these types of cases, the smaller the sample size, the larger the potential error. For example, if I was satisfied with a 5 percent margin of error – meaning that if I estimated that 50 percent of the population would vote for a certain candidate, it would yield a range of 45 to 55 percent – I would need to survey 783 likely voters. If I wanted to keep the sample error down to 3 percent, I would need to survey 2,178 likely voters. If I wanted to keep sample error to 1 percent, I would need to survey 19,620 likely voters. The ROI is determined by weighing the cost of the sample (or audit, in our case) to the criticality of the outcome (overpayment amounts).
The Probe Audit
So, if given the option, how can a practice determine which route to take? One way is to perform your own probe audit. Pull a random sample of claims (say, 30) and conduct your own review to determine what the damage would be if you were to be subject to an outside audit. If it’s a huge amount, you may not have a lot to lose in a 100 percent review. If it is a smaller amount (or financially survivable), extrapolation is likely your best bet. In either case, you already will have determined the range in which the overpayment estimate should fall, giving you the ability to assess the accuracy of a payer’s audit.
The Essential Issue
One very critical point to remember here, and I can’t emphasize this enough: how accurate an extrapolation audit turns out to be is almost completely dependent on how random a sample was. If the sample is not random, then all bets are off. Remember, extrapolation amplifies the degree of error within a sample – so in our example here, if an estimate for a sample is off by $10 per claim, multiply this by 10,000 claims and it means that someone just made a $100,000 mistake (and in my experience this type of mistake rarely, if ever, happens in favor of the provider).
You must ensure that samples are statistically valid random samples, and if you don’t know how to determine this, get help from someone who does. Remember, your financial survival may depend on it.
About the Author
Frank Cohen is the senior analyst for The Frank Cohen Group, LLC. He is a healthcare consultant who specializes in data mining, applied statistics, practice analytics, decision support and process improvement. Mr. Cohen is a member of the RACMonitorEnews editorial board and makes frequent appearances on Monitor Monday podcasts.
Contact the Author
To comment on this article please send to email@example.com