Artemis Health is rolling out a new risk score model for our customers and partners. Risk scoring is a common way in which healthcare experts, benefits professionals, and data scientists measure the health of a population. It’s a means of standardizing and normalizing healthcare claims data to better understand the health of a population and how they compare to others.
Artemis works hard to provide our clients with the most advanced, reliable, and useful data possible, and one of the ways we do this is with data enrichments. Risk scores are a good example of a “data enrichment.” Essentially, data enrichments are algorithms, data models, or other methodologies that take standard data and make it more meaningful. We use a number of these within the Artemis Platform, such as our unique Actionable Overspending model, Milliman Benchmarks, and others. They provide an extra layer of automatic analysis on top of raw data to better inform our users.
Previously, Artemis was using Johns Hopkins ACG (JHU ACG) System as our existing risk scoring methodology. It is a reliable and often used data enrichment in the benefits industry, but upon evaluation, we’ve decided to switch to Milliman Advanced Risk Adjusters (MARA) to replace the Johns Hopkins methodology. We’ve already begun the transition, and clients will see new MARA risk scores and adjustments with their next scheduled data refresh, which happens approximately each month. By July, all Artemis clients will be using MARA’s risk adjustment model.
Let’s dive a little deeper into how risk adjustment models work. There are two broad categories to consider: a concurrent risk adjuster and a prospective risk adjuster.
Concurrent risk adjusters use a given 12-month period’s claims data to calculate the risk score for an individual. So, for example, you could choose the 2019 calendar year, assess the risk of an individual in the population, and get a score. The “benchmark” for a risk score is 1.0. Anyone with a risk score higher than 1.0 is considered higher than average, while those below are considered lower than average. The “scaled to company” version of the concurrent model will normalize the risk to the company’s population. The “unscaled” version of the concurrent risk score is calculated by comparing the company against Milliman’s reference database (more on scaled vs. unscaled risk later). Concurrent risk adjusters use actual healthcare claims and expenditures to calculate risk.
A prospective risk adjuster uses claims data as well, but instead of focusing on what has happened in a 12-month period, prospective models focus on what is expected to happen in the subsequent 12-month period. If you use a prospective model for the last 12 months, it will give you a risk score for the next 12 months. It’s a valuable predictive tool for helping plan for costs and budgeting for benefits. Both concurrent and prospective models take raw healthcare claims data and transform them into useful risk scores. MARA offers Artemis the ability to use both of these risk adjusters to help our clients get insight into their population’s health, risk, costs, and potential future costs.
Why did we choose MARA? We saw three key advantages to using MARA for our risk adjustment model.
MARA stands out because it allows users to break down the risk score by seven specific service categories. Service categories are where/how the claim took place, suchs as inpatient care, outpatient care, primary care, and more. MARA takes these claims, weighs them into different service categories, and provides context for how the risk score was assessed.
In the screenshot above, you can see how this looks in the Artemis Platform. For this organization, the risk score is more heavily influenced by hospitalizations and less influenced by primary care visits. Artemis clients can drill down to see not only how these service categories weighted their organization’s overall risk score, but to the individual member level as well.
One other advantage MARA offers is the ability to “scale” or normalize risk to compare it on a few different levels. While it’s great to see a 30,000-foot view of how their population is doing, it doesn’t mean much to users if they can’t compare it to other similar or even very different populations. MARA offers both scaled and unscaled risk scores:
Many self-insured employers access their benefits data through a consultant or broker, and this offers one additional level of scaling. Brokers and consultants can look at scaled risk scores for an organization compared to their entire book of business. This helps them get a more holistic view of how each client’s population is doing compared to other clients they serve.
MARA Risk Strata may be familiar to you, though you might have heard them called “resource utilization bands.” Essentially, this functionality buckets the members within a population based on their risk score to help you see how many people are high risk, medium risk, low risk, or somewhere inbetween.
This isn’t new to Artemis, as Johns Hopkins risk adjustment model also provided these bands. However, MARA offers an additional resource to help you make sense of these risk strata. They provide a reference population that acts as a benchmark for how many members they’d expect to see in each bucket. This gives Artemis clients the ability to see if their population is less risky, as risky, or riskier than expected.
Here’s an example of how these advantages come together in real-life. One Artemis client was already using MARA risk scores for their Accountable Care Organization (ACO). With Artemis making this change, they are currently calibrating, comparing, and normalizing the data between the two data sets. Once the data normalization is complete, they’ll be able to get an even better look at how their ACO is performing using the Artemis Platform.
Artemis Health will be using the latest version of the MARA software (v4) and will be implementing the MARA CxXPLN risk adjustment model
If you have additional questions about our transition to Milliman Advanced Risk Adjusters, get in touch with your Analytic Advisor.