Suman Giri, Ph.D., and Shitang Patel

home

authors

suman-giri-ph-d-and-shitang-patel

Health plans have been working to optimize their performance with the Centers for Medicare & Medicaid Services (CMS) Star Rating Program for years. Organizations that prioritize Star Rating improvement tend to develop better processes, meet member needs more effectively, and increase member recruitment and retention — all of which can boost financial performance.

However, improving a Star Rating is not a trivial undertaking. The dynamic and complex nature of the evaluation criteria often makes planning tricky. The measures on which a health plan is scored, the weights assigned to each measure, and the cut-points used to calculate the score may change from year to year. Adding to the complexity is the fact that plans have to align resources to improve performance across approximately 40 measures and adjust their focus annually based on membership and utilization characteristics. And those efforts don’t include managing and responding to member experience measures, which require a slightly different approach.

Plans that aspire to improve their Star Rating need to continuously refine existing processes to better understand their members’ clinical and behavioral characteristics and know their providers on a more granular level. With a deeper understanding, the health plan can identify prescriptive insights that are laser-focused on the actions needed to drive outcomes.

This is where advanced technology like machine learning (ML) can have a significant impact. This game-changing technology can underpin a robust analytic framework that more accurately predicts performance, prescribes interventions, and enables action to realize Star Rating improvement.

ML algorithms can be used to predict measure-level scores and cut-points for the coming year, generating realistic predictions up to the contract level. For measure-level score prediction, a regression paradigm is recommended with features that capture the essence of a contract’s historic performance and other details about the plan. For cut-point projection, there are a few possible options. One involves predicting all measures for all contracts for the upcoming year and replicating CMS’ cut-point calculation 

 is an issue with this approach, and appropriate statistical techniques should be used to account for that. Another strategy is to predict cut-points directly. Similar to measure-level prediction, incorporating cut-point characteristics is key to effectively framing the prediction.

After measure-level scores and cut-points are predicted, a Star Rating can be calculated using measure-level weights. Figure 1 shows results for contract-level Star prediction for 446 contracts for rating year 2020 (Figure 1). The analysis used a combination of historic 

 and proprietary plan and measure-level information.

Once health plans have an idea of their projected rating, they must determine what measures to focus on to fuel Star Rating improvement. Certain ML frameworks can be utilized to prescribe the exact measures and associated changes that will increase the contract-level Star Rating. By aligning a plan’s resources to the measures that truly matter in its context, the organization can be more efficient about improvement and avoid wasting critical time and resources.

 formulations and approximations can also be equally powerful. Use of penalty terms in such a framework can force the function to find the fewest possible measures that are going to result in an increase in the overall Star Rating. Such frameworks can also take input from the user in the form of cost and difficulty of gap closure at a measure level. This allows the system to consider contextual information from domain experts while leveraging the power of analytics to guide data-driven decisions. By pursuing this strategy, a health plan can receive a concise set of focus measures, the recommended increase in scores for those measures, and the achievable contract-level Star Rating.

With a plan for improving a Star Rating in place, health plans can further leverage ML to achieve their goals. For example, for clinical measures where there are clearly identified gaps between current and optimal performance, measure-level targeting algorithms can be used to zero in on the members and providers that have the highest likelihood of engagement. When designing these algorithms, it’s important to consider the total number of gaps attributed to the member or provider and the importance of those gaps in the overall Star Rating. Any provider-related report cards and member or provider features that are already part of a plan’s standard reporting should be integrated into the targeting algorithm. Once optimal targets are generated, health plans can create campaigns with multiple touchpoints planned for high-value members.

With CMS leaning towards assigning more importance to member experience measures in the future, it is wise to develop a separate analytic framework to optimize member experience scores. This may involve looking at variables that are directly associated with member experience and tracking changes in certain proxy factors instead. For instance, a regression model of Consumer Assessment of Healthcare Providers and Systems (CAHPS) composite score and underlying social determinants, enrollment, utilization, and demographic factors was able to reliably predict changes in a health plan’s CAHPS composite score. A plan could use such a model to understand factors that truly affect its CAHPS score and start proactively running specific engagement campaigns in areas where these factors are underperforming.

Finally, when health plans launch their campaigns, they can use ML to create feedback loops that calibrate resource allocation. Concepts like AB tests can come in handy to understand strategies that work for certain kinds of outreach. In addition, health plans should explore the value of using provider-facing portals, EMR integrations and mobile push notifications. Any analytics performed should integrate with existing campaign management solutions to streamline and effectively focus interventions. By carefully crafting outreach based on embedded intelligence, health plans can bridge the gap between information and action, taking their Star Rating to the next level.

Suman Giri is Director, Data Science, CitiusTech

Shitang Patel is Assistant Vice President, Payer Market, CitiusTech

Articles

Optimizing your Star Rating program with machine learning technology.

Suman Giri, Ph.D., and Shitang Patel

Articles

Supercharge Your Star Rating Improvement Efforts Using Machine Learning