Predictive analytics have transformed the way the telecom industry communicates with existing customers and distributes its resources to maximize customer retention. Can healthcare companies leverage the same analytics techniques to combat patient non-adherence?
According to research published in Risk Management and Healthcare Policy, nearly half of the adults residing in the United States live with at least one chronic disease. As the healthcare industry — and the broader scientific community — has attempted to reduce the prevalence of chronic conditions among the country’s population, the percentage of adults in the U.S. taking at least one prescription drug has increased from 38 percent between 1988 and 1994 to 49 percent between 2007 and 2010.
That being said, as the study points out, medication non-adherence rates continue to be troublingly high, with somewhere between 25 percent and 50 percent of patients failing — for any number of reasons — to follow their physician’s’ guidance. This, the study explains, has had a significant impact on the U.S. healthcare system. Among older adults, for example, as many as one in ten hospitalizations can be directly attributed to medication non-adherence. Overall, non-adherence accounts for between $100 and $300 billion of unnecessary annual healthcare spending in the U.S., amounting to between 3 percent and 10 percent of the country’s total healthcare costs.
As the healthcare industry attempts to rein in its costs and improve patient outcomes, it’s clear that combating medication non-adherence must be a central focus. And while healthcare providers (HCPs) have yet to come to a consensus regarding the best way to incite the widespread behavioral change necessary to overcome this expensive impediment to our country’s health, advances in predictive analytics now offer an opportunity to not only identify trends in non-adherence, but anticipate patient non-adherence before it occurs.
Drawing Inspiration from an Unlikely Source: Telecommunications
Telecommunications consumers have a constantly evolving set of choices for video, high-speed Internet, and telephone services. They can switch providers easily and are quick to notice the differences in the quality of customer service. “Retaining current customers, and finding ways to capture a larger share of their media and communications spending, is a top priority for telecommunications companies,” says John Lucker, Global Advanced Analytics and Marketing Leader at Deloitte.
In an effort to combat constantly rising churn rates, many companies have begun to use predictive analytics algorithms to better identify churn drivers, profile at-risk subscribers, and better understand how to communicate with both existing and prospective customer segments.
In a case study released by Forbes Insights, Parimala Narasimha, Director of Advanced Analytics at Cox Communications, emphasizes the importance of predictive analytics in strengthening customer loyalty and in turn, lowering churn rates: “if they are already a customer, you want to know: Why are they leaving us? Is it because their promotional offer ended? Or is there a better offer in the market? Did they have trouble with our product? We need to understand both sides.”
By using predictive analytics to make sense of — both organizationally and analytically — the wealth of raw data at their fingertips, telecom companies like Cox are able to hone in on what really makes their customers tick. Data related to a customer’s recent purchase history, frequency of purchases, and total money spent is certainly helpful in its own right, but predictive analytics enables companies to push beyond these surface metrics and investigate how variables like emails opened, duration of website visits, mobile device usage, and frequency of call center interactions bear upon a customer’s likelihood to purchase or continue using a service.
These deeper insights help companies like Cox understand which customers are likely to add to their existing bundle, simply maintain their current subscription, or defect to another provider — and why.
Applying Predictive Analytics to Patient Non-Adherence
A similar set of analytical tools that companies like Cox use to personalize retention-based communications can be applied in the healthcare industry to predict patients at the highest risk of medication non-adherence.
The key is to start with a mathematical simplification of the potential patient behaviors at hand. In the case of modeling patient persistency – a useful simplification is in representing long-term adherence as a cascading set of discrete probabilistic behaviors. We have found success in modeling these behaviors within the context of a Markov Chain representation:
Once the mathematical framework has been developed to quantify persistency-oriented behaviors in a structured manner, any number of supervised segmentation algorithms (such as neural networks or support vector machines) can be trained to make predictions (personalized lapse risk assessments) with respect to transitional persistency probabilities.
In order to fuel the algorithmic training cycle, there are a variety of determinant features of medication non-adherence that can be leveraged via various sources of APLD. Demographic characteristics like age, and gender often play major roles in predicting non-adherence, in addition to other highly important factors such as payer type, income, geography, and out of pocket costs (often, the most important factor). In our experience, additional information about the prescribing physician and dispensing pharmacy can also glean significant information about predicted patient adherence behaviors (such as the specialty of a prescribing HCP, or the parent company of the dispensing pharmacy).
While the ability to assess which patients are most likely to lapse therapy is not a solution in and of itself — it lays important groundwork in developing robust, individualized communication strategies whose end goal is to combat medication non-adherence. By taking a predictive analytics-first approach to communications planning, healthcare companies can design tailored engagement programs fueled by predicted behavioral data to laser-focus resources, support services, and messaging for those patients at the highest risk.