Additional contributors to this post include Vijendra S.K, Saksham Garg, and Ishmeet Singh of the Evive Data Science team.
Evive Plan Choice is a new-generation decision support product for employees who find themselves overwhelmed by their annual enrollment options—and employers who want to help them choose the right option for their individual needs.
Without the right guidance, employees often default to the same health plan they had the previous year or select the option with the lowest monthly premiums. Unfortunately, this means they may not factor in current or potential medical conditions; related costs to those conditions; and variation in access due to differing provider participation in networks, underlying the options into their decision-making. In these scenarios, people are often subjected to higher out-of-pocket costs or disruption of care because they didn’t choose the optimal plan for their needs.
Based on large volumes of historical medical claims data and predictive modeling techniques, Evive Plan Choice anticipates future medical claims and provides relevant enrollment suggestions for employees. While Plan Choice estimates the ongoing medical care and incurred cost, users can choose whether to have the product factor in their own past medical claims—and if so, the recommendation engine uses those claims and associated providers to estimate the medical expense for the upcoming year. If the user chooses not to use their past medical claims (or if they’re a new hire who has no claims available to the product), they have the option to use the no-claims flow, where they note their ongoing/upcoming medical care, allowing the recommendation engine to pass that information to our predictive model and compute the estimated expense. On average, about 25% of users go through no-claims flow across our book of business.
While more than 60% of users have been selecting plans based on the recommendations, the Evive Data Science team does not believe in resting on our laurels.
We are continually looking at ways to improve our predictive algorithms and provide Evive’s end users with personalized experiences. With this in mind, we are developing a new and improved Plan Choice platform.
The most significant improvement to the Plan Choice no-claims flow is to use linear regression to identify patterns in medical claims data. This differs from our current version, where the no-claims flow uses mean (average) and median (central) values as predictors of future medical costs.
With the vast amounts of data we analyze, we also needed a sophisticated algorithm to leverage linear regression to analyze medical costs. The trained machine-learning algorithm finds the best-fit trend line and helps us achieve greater prediction accuracy—much faster. A huge benefit of using linear regression is the ability to stack year-on-year medical claims data for trends instead of analyzing all data together as one large category. This facilitates greater accuracy of the Plan Choice algorithm’s predictions.
Plan Choice’s predictions in the no-claims flow have been based on medical claims data, along with estimated pharmacy cost. Previously, the pharmaceutical cost was estimated as a percentage of the estimated medical cost. However, in real scenarios, the pharmaceutical cost varies with respect to different combinations of medical states and demographic factors. To solve this, this upgrade includes related pharmaceutical data in the predictive pipeline.
This is important, particularly for certain chronic conditions where diagnosis and treatment may not constitute the majority of out-of-pocket costs. The cost of prescriptions and refills adds a significant amount to the overall cost. For example, when it comes to conditions like rheumatoid arthritis or multiple sclerosis, pharmaceutical costs can far exceed medical costs.
Factoring this information into the overall analysis greatly increases the no-claims flow algorithm’s precision in predicting total costs.
Frequency of treatment
Certain chronic-disease conditions occur more frequently than others, like diabetes, high blood pressure, and more. These conditions constitute the vast majority of patients being diagnosed and treated. As a result, the related costs for these conditions significantly impact the medical claims data that the Plan Choice algorithm analyzes for its prediction models.
The severity of the condition correlates to greater utilization of office visits, with each visit associated with a cost or list of costs based on the tests or procedures conducted, such as MRIs, CT or ultrasound scans, X-rays, blood tests, and more. So in the no-claims flow, rather than ask patients to characterize their health status, we ask the individual to identify their number of office visits per year to adjust the allowed amount for a particular disease diagnosis or treatment. We categorized the number of visits based on the frequency of occurrence as follows:
- Low: 0-2 visits per year
- Medium: 3-4 visits per year
- High: 5-10 visits per year
By providing more considerable significance to these more frequently occurring disease conditions, Plan Choice is more effective at recommending plans with lower out-of-pocket costs for these common medical conditions versus those suffering from rare conditions. As Plan Choice continues to evolve, we hope to achieve higher accuracy for medical conditions that are infrequent in occurrence as well.
A better experience
These are just a few of many important enhancements being made to Evive Plan Choice. The upgrade will result in even more accurate recommendations to optimize healthcare spend based on procedural and diagnostic costs, particularly for new hires who use the prediction model rather than claims history as the basis for the recommendation.
The field of predictive analytics is advancing at a rapid speed—and in the benefits industry, Evive is at the frontier of leveraging it to help make people’s lives better.