November 19, 2019

How Predictive Modeling Can Demystify Spillover and Drive Market Access Advantage

Image of strings of light, seen from below

Spillover, as one Merriam-Webster definition notes, is “an unexpected consequence, repercussion, or byproduct.” Put simply, it’s an “if this-then that” type of situation. In the world of pharma, spillover refers to instances in which healthcare providers (HCPs) alter their prescribing behavior in response to — sometimes inaccurate — perceptions of restricted access, or perhaps in response to patient complaints about co-pay burden or drug pricing.

Predicting and quantifying spillover is a universal challenge across industries, but one that has even greater ramifications in biopharma, given patient health may be at stake. In biopharma, spillover cuts both ways for managed care organizations (MCOs) and prescribing HCPs. Consider the following scenarios:

  • An HCP with a dominant payer in their payer mix and a product with a preferred-coverage policy or formulary position could create a comfort level for that provider to prescribe that product for the most appropriate patients, regardless of MCO affiliation.
  • Conversely, if the dominant MCO in the HCP’s payer mix downgrades a product coverage policy or formulary position, the HCP could become frustrated with the new restrictions and hurdles required with increased patient co-pays and out-of-pocket burden and seek new therapeutic solutions.

Popular and historical solutions to the spillover challenge include analogues, heuristics (rules of thumb), and other qualitative research. Many of these solutions have become standard practice because, previously, there was not a quantifiable, predictive way to understand or measure the spillover effect.

How Machine Learning Can Help
Machine learning and predictive modeling are changing the way all of us buy, connect with others, watch entertainment programming, and read the news. Machine learning is, at its heart, a way to have computers help us find patterns and stories in historical data to predict future behavior.

The key component to spillover modeling is exploratory network science. Exploratory network science uses graph theory—much like Facebook, LinkedIn, Apple, and others—to investigate the relationships between entities such as MCOs and their related providers to highlight the influence that they have on one another.

Recent Intouch data science research into HCP relationships and influence has shown that anywhere between 40% to 60% of an HCP’s decision to prescribe a treatment is based on their peer relationships. For spillover determination, the key actors and data resources include MCOs, PBMs and related HCPs, along with prescribing behavior and medical coverage policy/formulary-level data. The spillover model’s output would be to emphasize the potential impact of one entity’s choices and/or actions on the entities with whom they’re related.

Ultimately, the outcomes derived from using a machine learning system can help our partners make better decisions about who to contract with, appropriate level of discounting, and when to walk away from negotiations, as well as quantify payer-mix impact on HCP prescribing behavior, peer and organizational influence, and the true value of the MCO’s ability to “hurt or help.”

Intouch recommends leveraging the combined power of its market access and data science teams to proactively and predictively address the spillover challenge using machine learning. The spillover concept has been investigated across a variety of industries and educational arenas—political science, economics, and psychology, to name a few—and today we believe a machine learning approach can improve the opportunity to not only identify the relationships between entities that influence spillover but the indicators and probability that spillover will occur in the future, as well as the impact that spillover could have on the downstream entities.

Predicting and quantifying spillover has historically been a daunting endeavor, often driven by analogues and common rules of thumb. The application of machine learning, network science, and the historical data available today make this a ripe opportunity to create a truly modern, flexible solution that can be used across the enterprise, from payer contracting to HCP targeting for pull-through initiatives.

This content was produce by Mike Motto, senior vice president of market access and Sam Johnson, senior director of Intouch’s analytics lab.