September 6, 2016

Predictive Modeling for Better Results

Data can help us forecast how plan members will act, how drug prices may change

Scroll to continue
Predictive modeling for better results PDF cover

Being able to see into the future has to be every planner’s fantasy. As our ability to analyze and extract insights from data expands, this dream is becoming more and more of a reality. Increasingly sophisticated algorithms and models are delivering insights that help us and the plans we serve be more effective in all aspects of managing pharmacy benefits — from plan design selection to cost management and member communications.

"Prediction is very difficult, especially if it’s about the future."
—Niels Bohr

Current market volatility makes prediction both more challenging and more critical than ever. Our analytics teams continuously work on evolving and refining our predictive models to help answer fundamental payor questions. For example, we are investing in advanced analytics to provide more personalized services and efficient operations, such as the development of our Predictive Adherence Index (PAI), which needs only three months of prescription claims to predict a member’s long-term adherence trajectory. These analytical algorithms enable us to determine which members are at risk of skipping doses or stopping their medications, and to design educational programs to help them stay on therapy. These types of analytical tools are also designed to ensure compliance with applicable laws and client contracts.

How Will Implementing This Solution Affect Trend?

Our interactive RxInsights® tool allows pharmacy benefit management (PBM) clients to view how various therapeutic categories are driving drug trend. Clients can drill down to identify specific drugs with high inflation or sudden increases in utilization.

RxInsights now also includes a predictive savings model that incorporates the plan’s claim data, along with modeling about prospective drug launches and inflation forecasts. The tool also incorporates behavioral models that predict utilization changes based on modifications to plan features; for instance, how will

members respond to incentives to increase generic utilization. This allows clients to model the impact of various program implementations — formulary, utilization management and network strategies — on their trend.

With this tool, plans can compare the effect of maintaining the status quo against the projected impact of program changes on annual trend. Due to the interactive nature of the tool, plans and account teams can rapidly assess and discuss any combination of changes they are considering.

Executive Savings Dashboard

Computer mouseThis interactive tool enables selection of any combination of formulary, utilization management, channel and health improvement solutions, and shows their projected impact on the client’s trend.

What Drugs Are Driving Up Costs Right Now?

Sudden and drastic price increases for some pharmaceutical products have been a challenge for payors over the last several years. Heavy promotion of drugs by manufacturers has also led to surges in utilization of specific products. Rapid identification and timely response to such challenges can help plans control their costs. Using what we know about historic patterns of price increases and utilization, our analytics team built models to rapidly detect outliers by comparing recent claims against projections of price increases and changes in utilization. 

The team is also further evolving the model to detect outliers in near real time on a plan-specific level.

Our Dynamic Trend Manager solutions were developed to respond to such high, unexpected price and utilization increases. As described in recent issues of Insights Executive Briefing, solutions to date have addressed high trend in the dermatological category and hyperinflating drugs those with triple-digit inflation or more.

Dynamic Trend Manager

Dynamic Trend Manager

Who is Most Likely to Move to a 90-Day Prescription? To a Generic? To Use Mail Service Pharmacy?

It is well known that some choices using generics, filling prescriptions for 90 days and using mail service pharmacy are typically better for members and payors. They can lower cost while helping to improve adherence. But reaching out to members to communicate these benefits is not consistently successful in effecting behavior change.

Our analytics team built predictive models focused on improving the effectiveness of member communications. Our model helps identify the optimal channel based on a member’s likelihood to change to, and stay on, a 90-day prescription at a CVS Pharmacy® retail location or through mail order. Our analytics team took into account not only the best prospects for a particular intervention but also how and when to contact a member about a proposed behavior change. Other models focus on driving other types of preferred behavior such as enrolling in auto refill to help members stay adherent to their chronic medications.

The modeling suite takes a member-centric approach, utilizing robust data points across a member’s profile, including demographics, prior channel utilization, out-of-pocket costs more than 500 data points in all. The suite enables outreach through the most impactful channels and at the preferred time for the individual member, helping ensure not only better results but higher member satisfaction. 

When is the Best Time to Intervene With This Member for Better Adherence?

Pharmacy Advisor® Counseling for members managing chronic conditions has consistently contributed to improved adherence. The standard program includes first-fill counseling, during which members can ask questions about the therapy, learn about side effects and get information about the importance of staying on therapy. If members are subsequently late to refill, a pharmacist reaches out to them to understand their reasons for non- adherence and offer solutions and refill options.

The CVS Health Research Institute has been looking into adherence since its inception. Our research has generated important insights about therapeutic complexity, patient utilization patterns, and the diverse factors that affect adherence. On the basis of work spearheaded by our research and analytics teams, we developed the PAI, which needs only three months of prescription claims to predict a member’s long-term adherence trajectory.

Beyond simply quantifying the member’s risk of non- adherence, the PAI defines patterns of adherence — both how likely it is that a member will drop off therapy and when the drop off is likely to occur. We recently piloted the use of PAI in our Pharmacy Advisor program. 

We applied PAI to a group of 3,000 members matched to a control group. After a baseline period, the PAI group received proactive counseling calls aligned with their defined adherence trajectory. Pharmacists called at the right time to help prevent that particular member from dropping off therapy. Compared to control groups, PAI-timed counseling groups experienced lower therapy discontinuation rates: 11 percent lower for members with diabetes, 9.5 percent lower for hypertension, and 7.8 percent lower for hypercholesterolemia.2

Predictive Models, Real-World Results

CVS Health is using predictive modeling to intervene more effectively with members, to help payors get desired results from plan design changes, and to deter unexpected, budget-busting trend drivers. These enhanced results depend not only on the expertise of our analytics team, but on our connected model and an infrastructure that allows us to reach out through multiple channels

with consistent messaging and high-quality member information. Whether we’re supporting members through a formulary change or helping them stay on therapy, all these systems and capabilities have been honed over the last decade to work together to help achieve better results for plans and higher satisfaction for members.

Bob Darin, Chief Analytics Officer

Bob Darin
Chief Analytics Officer

1. CVS BWH research collaboration: Franklin JM, Shrank WH, Pakes H, Sanfelix-Gimeno G, Matlin OS, Brennan TA, Choudhry NK. Group-based Trajectory Models: A New Approach to Classifying and Predicting Long-Term Medication Adherence. Medical Care, 2013; 51(9): 789-796. 

2. Hypercholesterolemia results not considered statistically significant. 

*Mail pricing at CVS retail for ERISA governed plans.

The Maintenance Choice program is available to self-funded employer clients that are subject to ERISA. Non-ERISA plans such as insured health plans, plans for city, state or government employees, and church plans need CVS Caremark Legal’s approval prior to offering the Maintenance Choice program. Prices may vary between mail service and CVS Pharmacy due to dispensing factors, such as applicable local or use taxes. 

CVS Health uses and shares data as allowed by applicable law, our agreements and our information firewall. Actual results may vary based on factors such as programs adopted by the plan. Client-specific modeling available upon request.  

© 2016 CVS Health. All rights reserved. 106-38980E   090616