# GHDP-120-18: Avoiding Unintended Consequences in ACO Payment Model

Use this thread to discuss ANYTHING and EVERYTHING related to this syllabus reading.
Some possible questions include:

• How can this reading be tested?
• I don’t understand a specific topic/formula - Can we discuss this?
• This reading gives me nightmares. Can we talk through it a bit?

Good luck!

If an ACO is eligible for 60% shared savings in a two-sided model, are they to pay back 60% or 40% if they have a loss? I feel like I’ve seen this both ways, and it confuses me.

It depends on their quality score:

ACO gain sharing percent = quality score * 60%
ACO loss sharing percent = min ( 1 - quality score * 60% , 60% )

The min() function just caps the sharing percent so that an ACO with poor quality score doesn’t end up sharing more than 60%, and the “1 - quality score * 60%” part is the complement of the gain sharing percent formula.

So the ACO’s loss sharing percent can really only range from 40% (at 100% quality score) to 60% (at or below 66.7% quality score). As their quality score decreases, loss sharing increases.

I just realized your question may have been referring to problems that don’t address quality.

I don’t think I have seen any problems like this that I can point to, so please share if you’re looking at a specific example. However I think in a problem where no quality information is given, you could reasonably assume one of the following:

• 60% sharing percent for both gains/losses

• Assumes there are no quality incentives in place
• 60% sharing for gains & 40% sharing (i.e. 1 - 60%) for losses

• Assumes quality incentives are in place & the ACO has 100% quality score

Does anyone else have a different understanding?

Let’s say we know quality score is 100%. So, savings percent is obviously 60%. Your formula suggests when a loss beyond the threshold is reached, the ACO must repay 40%. Let’s say \$100M is the benchmark, but the ACO has \$110M in Medicare spending. So, the loss is \$10M. So, the ACO must repay \$4M to CMS in shared losses. That’s what I think you’re saying, and it’s how I thought it worked. However, see the solution to Advanced Spring 2018 #5. It would have the ACO paying back 60%. So, now I’m not so sure.

Yeah I see what you’re saying, and I feel like this is an oversimplification of the payment methodology due to the problem mainly being centered around the incentives of different baseline calculations.

I don’t think the loss share percent was a huge focus because it would yield the same results in terms of the preferred baseline calculation method. I.e. if you substitute 30k instead of the 20k in the solution, you should come to the same conclusion for parts (b) and ©.

But I agree that the “marginal revenue after sharing” should rather be 30,000 instead of the 20,000 in the solution, and I would hope they would have accepted the different answers for part (a) = \$57,000 for (i), and \$84,000 for (ii).

1 Like

Okay… so I was correct to answer the way I had started to (before peaking at the solution). That’s what I thought, but the solution threw me off. Thanks for confirming my understanding.

1 Like

It’s presented both ways in different readings unfortunately. You’re supposed to figure out which reading is being tested. If you’re not sure, your best bet is the 40%. (1-60%)

In this study note they are over looking several minor issues to try and keep focused what the paper was about, unintended consequences, basically non equal weights to average the benchmark. For example they over looked checking the MSR too.

Bottom line is that the authors probably were just saying that you could have a situation with 60% shared losses or in another situation you could have 60% shared gains. But we know from the formula shared losses% = 1- shared gains%, so it can not be both, 60%, at the same time.

Unfortunately the SOA copied the mistake into there solution for Spring Adv 2108 #5. Just know that this paper was trying to show one thing and it was unintended consequences. I would not suggest following there shared savings % method.

3 Likes