2023 ACA Risk Adjustment Transfer Report - Part 2: Putting it in Context
In our last post, we looked at why risk adjustment exists to begin with, and why looking at the raw transfer alone doesn’t tell you much. In today’s post, we’ll look at different ways you can make sense of the data and what it tells you about the companies that are receiving and paying into risk adjustment.
Normalize it: Per Member, Per Month (PMPM)
First, we should understand the concept of “per member, per month” or PMPM. PMPM is a key measure for any health insurance company, and it’s used in all sorts of metrics - you’ll see claims, premium, utilization, administrative costs, etc. all expressed as a “per member, per month” measure across the industry. Lots of industries try to normalize measures this way: airlines looks at revenue by passenger seat-mile, hotels by bed-night, etc. The math for PMPM is relatively simple - whatever metric you’re looking at (usually a financial one) divided by the number of members for a given month, (or number of member months for a given period). So, if you had $500,000 in premium for a month, and you had a thousand members in that month, your premium is $500 PMPM.
This brings us to a second, related concept: member months. A member enrolled for 12 months of a year is different than a member enrolled for only one month. Thus, a member enrolled for January to December has 12 member months for that year, and if they were enrolled only from July - December, they only have six member months. This seems obvious enough, but it matters because sometimes publicly available information just gives you an enrollment count, which can be misleading if the carrier’s membership isn’t consistent from month to month (which it usually isn’t). Often, you can get close enough by just assuming the enrollment was consistent throughout the year (multiplying membership at a point in time by 12), but this can miss some nuance given seasonality of enrollment and disenrollment patterns. If you can get your hands on member months data, that’s even better. How do you get hands on such data, you may ask? Great question!
PUF: The Magic Data
One of the nice things about working in a regulated industry is that there is a lot of data out there, even from privately held companies. CMS publishes various Public Use Files, or “PUFs” which provide information about things like enrollment, rate filings, plan designs, etc. about health plans regulated by the ACA. These can look a bit different depending on whether the plan is sold on HealthCare.gov or on a state Exchange, but generally the kinds of information you can get is consistent even if the format isn’t always the same. PUFs that I use regularly include:
1) The Plan Data PUFs: this has detailed information about the rate and plan filings that carriers submit to CMS. This has things like what rates are being charged, detailed on the plan designs, what areas a carrier sells its plans in, and even fun administrative data like denial percentages. These usually come out in late October for the next year - i.e. the 2025 plan data will come out in October, 2024. This page is for all the FFM carriers, but there’s another page for State-Based Marketplace (SBM) states here. The SBM one has a less consistent release schedule.
2) The Rate Review PUFs: this has detailed information from the carrier’s rate filing. This is also called the “URRT PUF” - for the Unified Rate Review Template. The URRT is an Excel template that every carrier has to fill out and submit to CMS and state regulators. This is different from the plan data PUFs, although some of the data overlaps. I think of the plan data PUFs as administrative data templates. They have the information you would need to display a plan on HealthCare.gov, like what’s the copay for a specialist visit, and what is the rate for a 42 year old in Broward County Florida. The Rate Review PUF, on the other hand, shows you data around how a carrier came up with that rate (or at the very least, what data they shared with regulators to justify the rate they’re charging). It has both what they project their costs to be in the next year, as well as what their actual costs were in the prior year. It also includes information on off exchange plans, both individual and small group. The PUF is a massive flat file aggregating all the Excel templates that every carrier submitted together, so I find it helpful to look at one of the URRTs from a carrier when exploring this PUF. Here’s an example of one for the 2025 rate cycle for Ambetter’s Georgia plan, obtained via an Open Records Request from the Georgia Office of the Commissioner of Insurance. Georgia releases the PY 2025 templates earlier than a lot of states, so for most states, the PY2025 templates aren’t available. The first tab has carrier-wide information about claims experience, broken down by benefit category; the second tab has plan-design level information but at a less granular level, and the third tab has the geographic factors for each rating area. Spend a little bit of time in this file and you’ll see how much information there is.
This PUF is really useful for looking at things like overall performance for a health plan - but the data is a bit lagged. The rate review PUF is typically released in October for the following year, based on data filed with CMS and state regulators in the summer. So 2025 Rate Review PUFs will be released in October. The most useful information, in my opinion, comes from the carrier’s “experience period” data they present. The experience period data will be 2 years before the “projection period” - the period for which the carrier is filing its rates. So when the 2025 Rate Review PUF is released, the experience period data will be for the 2023 plan year. The data is lagged because of the timing cycle for rate filings - initial rate filings are typically due in May or June, so carriers wouldn’t have much claims data for 2024 yet to use in their filings, while the 2023 plan year should be mostly complete, including claims run out. To come up with their 2025 rates, carriers take their 2023 data and project it forward into 2025. The rate review PUF gives us a treasure trove of data about the experience period: claims expense (both how much the plan paid and how much the member paid), premium, member months, state reinsurance, utilization (how many hospital visits, doctors visits, prescription fills, etc. the plan had) and risk adjustment transfers. Put it all together, and you can see how well a plan did and infer what the drivers of that might be. For most of these measures, this is down to the individual plan design level - so you can see not only what a carrier’s overall experience is, but also what their bronze, platinum, etc. experience looks like. This is my go-to PUF for statewide market analysis, but it isn’t perfect. First, as mentioned, it’s lagged by a year. Second, it’s also limited to carriers still planning to sell their products in the next year - so it sometimes misses some key data points. For example, the 2024 rate review PUF doesn’t have any data for Friday or Bright since both of those carriers exited the market by 2024, but these two carriers had a meaningful amount of claims experience for 2022. Luckily, there’s another PUF that can help fill that gap… the MLR PUF!
3) The Medical Loss Ratio PUF: this PUF has information carriers submit to CMS to show how they’re complying with the ACA’s Medical Loss Ratio (MLR) rule, which requires them to spend 80% of their premium dollars on medical claims (85% in the large group market). Like the Rate Review PUF, the data is a bit lagged: it’s usually released in the fall for the prior year (e.g. 2023’s MLR PUF will be released sometime this fall), but last year it was quite late - not until May of 2024 was the 2022 MLR PUF released. It’s useful because it includes all of a carrier’s medical business, including transitional or grandfathered plans which may not be reflected in the rate review PUFs. It also will include a few other nuggets like how much in prescription drug rebates a carrier received, how much they spend on admin, and how much they spend on certain taxes. However, the inclusion of non-ACA regulated plans like transitional and grandfathered plans can also muddy the waters depending on how you are using the data. It’s also all at the statewide level, and doesn’t have any plan design level information like the rate review PUF does. If I want to drill down to the county or plan level, there are two PUFs that provide some useful information.
3) This brings us to the Open Enrollment PUF. This is the PUF that typically provides the most recent information: it’s published a few months after OE ends and tells you about the current year’s enrollment. It doesn’t have any carrier-specific data, but it does have some county-specific data that’s sliced in lots of useful ways. It has total enrollment by county and ZIP - sometimes redacted for very small counties or ZIP codes… I’m looking at you, Talliaferro (inexplicably pronounced “Tolliver”) County, Georgia. This enrollment information is sliced by metal level, demographic (age, race/ethnicity, gender, etc.) and poverty level. You also get information about average premium at the county/ZIP level and average APTC. It’s useful for looking at market trends, and if you work at a carrier and have access to your own data, you can infer your market share.
4) Lastly, there’s the Issuer Level Enrollment Data PUF. This PUF provides enrollment data by issuer at the county level, and by plan design at the state level. It used to be lagged by a few years, but it’s been getting more up to date as of late: in June of this year, CMS published this PUF for plan year 2023 for both SBMs and the FFM. This is a great data source for looking at market share, and it tells you a lot about enrollment trends and which plan designs are attracting members. One thing that stands out to me: Even though there’s been a massive proliferation in the number of plan designs out there, enrollment still ends up concentrated in just a handful of plan designs. In every FFM state but Illinois, more than 10% of enrollment was concentrated in the top two plan designs. Even in Texas, which has 943 unique plan designs available across the state, the highest enrollment plan (an Ambetter silver plan) has 7% of the entire state’s exchange population enrolled in it. And, in almost half the state (15 out of 33), the top carrier had more than a 50% market share, and in all but two states (Florida and Ohio), the top two carriers combined had more than a 50% market share. One limitation is that it doesn’t include off exchange data, but usually this isn’t a major factor (except in Arkansas, where their Medicaid expansion uses off exchange QHPs).
The Nuts and Bolts: How I used the data
As you can see, there’s tons of info out there. This doesn’t include other sources like SERFF, NAIC filings, and state-specific PUFs produced by different State Exchanges. But it does give you an idea of the type and depth of data you can find. So how do we use this to make sense of the risk adjustment data? We can start by figuring out what the PMPM transfer is at the carrier level. This isn’t entirely straightforward, as there’s not a single PUF that consistently answers the question of how many total ACA-regulated (on and off exchange) member months a carrier had a given plan year. So CMS, if you’re reading this: consider sharing issuer level member months in the transfer report. Issuers will probably cry foul and say its proprietary, but most of that data is already out there or will be out there in one of the other PUFs.
Until they do that, we have to make some educated guesses, so in my analysis I went about it a few different ways. First, I decided to just focus on the individual market. Second, I established a hierarchy of data sources. The best is the Rate Review PUF, because it has actual member months at the carrier and subsidiary level. This only works for PY 2022 since the 2025 URRT data isn’t out yet. But if a carrier was still selling their product in 2024, I used this as a my data source. For those not selling in 2024, I then defaulted back to the Issuer Level Enrollment PUF, multiplying the average plan enrollment column by 12. I then normalized it to the number of member months For the 2023 data, it was a bit trickier. First, I used the issuer level enrollment PUF since it has 2023 member months, but this excludes off exchange plans, so as a backup, I used the 2024 Rate Review PUF. For plans that filed in 2024, they have to report their current enrollment in the URRT PUF, so I took this value and multiplied it by 12.
For both years, I then compared the resulting member months to the statewide total member months reported in the risk adjustment report (CMS doesn’t give you issuer level member months, but they do give you statewide member months). If it was within 5% or so, I just redistributed the difference pro rata to each remaining carrier. For larger discrepancies, I dug a bit deeper and made manual adjustments as needed - for example, in Arkansas, I went to the state Medicaid agency where they had reporting of enrollment in the off exchange Medicaid expansion QHPs and augmented the data from there.
As you can see, this is not an exact science, but it got me pretty close. Now, I’m ready to analyze the data with some real context! So in the coming days, I’ll look at some of the findings - in my next post, I’ll look at how risk adjustment tends to look different for different types of carriers. My plan is to divide them into Blues (other than Anthem/Elevance), publicly-traded carriers, provider-sponsored plans, and “other,” but would love any feedback on whether this is the right way to slice them. Should I look at the carriers whose primary business is Medicaid as their own category, irrespective of their ownership category. In other words, would it be more meaningful to look at Centene, Molina, AmeriHealth Caritas, and CareSource, etc. as one category, even though Centene and Molina are publicly traded, AmeriHealth is a subsidiary of a Blues plan, and CareSource would fall into the “other” category? What categories do you think are meaningful?