Risk adjustment is crucial for identifying high-risk patients and predicting costs, especially for Accountable Care Organizations (ACOs) participating in the Medicare Shared Savings Program (MSSP). Do you know what your ACO population will cost you next month? Best guesses are still just guesses and don’t provide the level of accuracy your organization needs.
Lumiata solves this problem by using machine learning to make predictions at the claim level for your ACO population, such as likely future claims, costs, and more. With our innovative tools and prediction models, you’ll know what costs and revenues to expect so you can plan ahead, allowing you to save time and money while providing improved member care.
Want to see proof? We recently partnered with a large ACO in the South to test our prediction models and compare them to the organization’s actual results. As a preliminary figure, we were at least 95% accurate in estimating the monthly cost of our customers (MSSP) for a population of about 15,000 members during the second half of 2020.
How Our Prediction Models Work
Lumiata can help predict at the claim level for your ACO population. We start by cleaning and enriching your data for use in our machine learning models, which we use to predict procedures, doctor visits, prescribed medications, and other claims. This gives you the ability to better control your costs and plan accordingly, without sacrificing care quality. Take a look below to see how our prediction models helped our customers predict expenses and revenues.
Case Study: Large ACO in the South
While we’re confident in our prediction models, we know it’s important to have real-world data to prove our capabilities. We recently partnered with a large ACO in the South to forecast allowed total spend amounts for three of their populations on a monthly basis. We also projected the benchmark per member per month (PMPM) target for one of their ACO populations. Our predictions were then compared to our customer’s actual expenses and revenues.
Task 1: Expense Projections
The first part was to forecast on a monthly basis the allowed total spend amounts for three of our customer’s populations. Predictions were made based on historic, current, and incoming financial information included in the claims data files. Each month, we provided financial estimates for the preceding four calendar months. Each estimate featured an aggregate per member per month (PMPM) amount along with approximated PMPMs across these service categories: Inpatient, Outpatient, Professional, Pharmacy, and Other.
We made predictions according to four snapshots of data. The “true cost” value was defined as the total allowed amounts for each population, each month, and each claim type as of 4.5 months past the last date in the projection month. For example, July 2020 was considered fully paid by December 15th, so our customer used a December 15th claims snapshot to determine the true paid amount for July claims. To be precise in this example, our customer used claims incurred in July, paid through December 15th to determine the ‘true cost of July for all three populations.
Moreover, Lumiata used eligibility information as of the current data to determine who was eligible in the three populations during the projection months. Medicare Shared Savings Program claims were lagged 2 months for reporting, so this included paid data through mid-Oct. For MSSP, we used mixed paid and allowed amounts from the different service categories to aggregate to a ‘total cost.
Lumiata’s total and claim type monthly cost predictions were compared with our customer’s cost predictions for two of their populations. The ‘error metric’ used was the absolute error between predicted and actuals. We also offered member-level predictions per month, population, and claim type, though this exercise was secondary to the total per month and the MSSP target calculation.
Task 2: Revenue Projections
The second part was to project the benchmark PMPM target for one of our customer’s ACO populations based on available claims data, adjusting for enrollment mix type, cost trends, truncation factors, and risk scoring adjustments.
Our customer participates as a Low-Cost ACO in the MSSP Enhanced ACO model, which then adjusts the overall benchmark PMPM based on a 35% regional experience and a 65% national experience blend. This projection aided in projecting any reconciliation amounts for the program after applying a quality-adjusted sharing rate and considering programmatic risk-sharing parameters such as the selected MLR/MSR risk corridors, risk-sharing rates, and risk caps.
Centers for Medicare and Medicaid Services uses the MSSP benchmark PMPM target calculation to determine a reasonable amount of money to pay the claims for the MSSP population, plus how much reimbursement to give an ACO. Target calculations were provided for each retrospective cost prediction.
Lumiata produced a year-end benchmark PMPM target estimate for MSSP. We used all available data to estimate this value, and the evaluation of this prediction was considered against our customer’s own target estimation. We provided these predictions with each monthly prediction submission. Final expense data was used as the sole criteria to evaluate the accuracy of each set of predictions made for four different submissions of projections.
Our results were at least 95% accurate in estimating the monthly cost of our customer’s MSSP for a population of about 15,000 members during the second half of 2020. Because of these more accurate cost predictions, our customers gained better visibility into where their costs will land relative to CMS’s target set for their population for the year. Our predictions also helped guide their cost-control decision-making, provider outreach, and steerage.
Want To Make Claim-Level Predictions for Your ACO Populations?
Lumiata enables you to predict claims, costs, and more, empowering you to prepare ahead of time, save time and money, and provide members with high-quality care.
Ready to see the predictions you can make with Lumiata? Request a demo today.