Fuzzy Math

A recent paper in the Journal of the American Medical Association notes that the current Medicare Bundled Payment for Care Improvement Advanced (BPCI-A) ended up costing the government money and is, in fact money-losing. The paper reports that: "The analysis observed an aggregate spending change of ?$75.1 million across the 428 670 episodes in BPCI-A model years 1 and 2. However, CMS disbursed $354.3 million more in bonuses than it received in penalties." In other words, the program generated $75 million in "gross" savings, but CMS doled out a net $354.3 million in shared savings, and therefore incurred a "net loss" because $354 million is more than $75 million.

WTF ???!!!???

To anyone with basic math skills, it's a bit difficult to understand how any organization -- especially a government agency that has to account for every penny -- can distribute substantially more in shared savings than it actually saves. But that's where the fuzzy math kicks in. And to be clear, the fuzzy math isn't the true accounting math that led to the $354 million in shared savings, it's the math that led to the conclusion in the paper. And that fuzzy math is the same that has been used by the folks that have been evaluating the Medicare Bundled Payment program since its inception.

Before I get into the 2=1 math, it's important to understand how the program works:

1. Hospitals, health systems and physician groups volunteer to participate and they select a group of bundles in a specific clinical category -- for example orthopedics;
2. For each participating provider, Medicare sets a target price across the selected bundles. The target price is calculated as follows: Baseline costs - 3% * (1+/- trend rate). It's a little more complicated than that, but not by much. (a) The baseline is derived from several years of historical performance, so the better your performance in past years, the harder it is to achieve savings. That's why most providers select bundles at which they have historically sucked. (b) CMS takes its upfront vig by notching down the baseline by 3%. (c) CMS applies a trend rate that can be either positive or negative. The trend rate reflects an expected increase or decrease in the costs of the bundles. These trend rates have largely been negative. So, to recap, and using a simple numeric example, if the baseline is $100, CMS takes off $3, then reduces the balance by a trend -- let's use $2 -- and that generates a target price of $95;
3. At the end of a performance period, CMS compares the amount of money that was spent on each contracted episode to the target price. If the actual amount is lower, then there are shared savings that are owed to the provider, and if the amount is higher, there are shared losses that have to be paid by the provider. But wait, there's more.
4. If the consultants that CMS uses to calculate the up front trend feel that they got it wrong, they can come on the back end and retro-adjust the target price. That is what has happened every year in this program, and the retro-adjustment has typically been negative. Leveraging the example above, while the up front (known) target price was $95 (baking in $5 of guaranteed savings to Medicare) there is a retro-adjustment of another $2 that puts the revised target price at $93 (baking in $7 of guaranteed savings to Medicare), and it's to that revised target price that the actual costs are compared.
5. From a pure accounting standpoint, CMS performs these calculations and determines whether or not the actual costs were higher or lower than the revised target prices. Therefore, in order for CMS to distribute $354 million in savings it has to generate at least that much in calculated savings: revised target price - actual costs. It cannot simply send out money for nothing. So the gross savings, which could be determined as Baseline Costs - Actual Costs During the Performance Period, are much higher because Medicare knocks down the baseline to collect it's initial vig plus what it expects providers to save at a min. The $354 million in distributed savings is the portion of savings that providers were able to generate beyond the 7% collected by CMS.

Back to the fuzzy math. How in the world did these researchers come up with $75 million in savings when, from an accounting standpoint, it was substantially higher (close to 5 times higher)? The answer requires something close to a PhD, but I'm going to break it down for you into high school math. Not because you can't get PhD math, but because you'll never finish this blog if I go down that road.

The researchers perform a completely artificial and academic calculation that goes something like this: Savings from Participants - Savings from Matched non-Participants = Actual Savings to Medicare. In other words, they look at what happened to the costs of similarly calculated bundles for providers that didn't participate, to the savings from the ones that participated, hypothesizing that if the program hadn't existed, Medicare would have gotten savings from the Participants that are at least equivalent to the non-Participants. And so those savings don't count, and it's only the incremental savings that were generated by the Participants that should count. That's how you get to $75 million in incremental savings between the two groups -- not the actual savings of the Participant group.

But then they commit what I think is an absurdity, which is to deduct the portion of savings distributed by Medicare, which was derived from normal accounting practices, and conclude the program lost money (because $354 is greater than $75). But it didn't. From a pure accounting perspective, it saved a lot. And the contribution of the paper should stop at the observation that the Participants saved more than the non-Participants, which is the goal of these programs.

This has been long enough and I applaud those who stuck through the math -- fuzzy and not fuzzy. What irks me to no end in these evaluations is the inference that there was an accounting loss when there wasn't. What the researchers are attributing as a loss is a purely theoretical and academic mathematical equation that has nothing to do with the practical accounting of the success or loss of a program. In addition, it completely glosses over the fact that the reason why Participants participate, and select bundles at which their historical performance sucked, is because they believe they can improve that performance and generate savings. Absent the program, they would just keep sucking at the bundles and Medicare, and all of us, would pay the price. So we're all better off for the program having existed.