Skip to main content

Applying Predictive Analytics in Adherence Programs

In recent years, predictive analytics and machine learning have become more widely used in healthcare, especially when it comes to predicting disease and its prevalence. These breakthrough technologies have also started to be applied to predict patient medication adherence — but in terms of widespread and consistent use, are still in their infancy.

As an example, a group of recent online seminar attendees were asked how much they had seen their organization leverage data to personalize and tailor patient support and adherence interventions. Only 12% said advanced analytics models were being employed.

Moving away from a broad brush approach

Without predictive analytics capabilities in place, a broad-brush approach is the norm in patient adherence programs — where similar actions are applied to all patients without distinction. Some more advanced programs use business rules-driven logic to create if/then decisions to deliver next best actions. But this type of “flat” logic does not allow for tailored engagement and interactions based on an individual’s unique situation, historical behaviors, preferred communication channels and other relevant characteristics. In flat logic, orchestrated engagements are less personalized, less efficient and, not surprisingly, less effective.

It’s time for the pharmaceutical services industry to start widely applying predictive analytics — and introduce prescriptive analytics — to realize the significant benefits of improved adherence and better patient outcomes.

On the whole, predictive analytics can help pharmaceutical organizations improve customer experience and outcomes, cut costs, streamline operations and protect revenue. In healthcare, if we can predict the risk of a patient dropping off medication, we can design interventions before they stop taking their medication to extend their length on therapy.

By continually analyzing engagement behaviors and marrying patient and social determinants of health (SDOH) data, persuasion models can be introduced to prescriptively drive and reinforce desired behaviors.

With strategic predictive analytics, hub service center agents and digital engagement channels can hone in on patients with a high risk of dropping off therapy. Efforts can be intensified and focused for those high-risk patients and resources can be shifted from low-risk patients.

The strategic use of predictive analytics enables a more robust response to customer needs in real time and ensures actions taken more effectively support patients, providers and pharmaceutical company objectives.

Predictive analytics for adherence consists of four key components:

1. Data – Predictive analytics makes use of all data, including structured data — such as activities, statuses, profiles, transactions, and timelines of patients, as well as unstructured data — such as text-based comments and audio-based conversations with an agent/nurse. Through Natural Language Processing (NLP), a machine learning technique for text and audio data, interactive conversations can be grouped and classified based on characters and sentiment. Metric trends can also be evaluated to boost predictive accuracy. 

2. Predictive Model – To predict ongoing medication therapy adherence, the most common predictive analytics models are Logistic Regression, Random Forest, or Neural Network. These are powerful tools used in predictive analytics and machine-learning.

Figure 1 below shows a graph of a Random Forest application with an AUC (Area Under the Curve, a measure of predictive accuracy), well above 0.8. Numbers above 0.8 are generally regarded as excellent. Figure 2 shows another preferred performance measure: efficiency rate — the ratio between the actual drop-off rate for the highest-risk patients and the average drop-off rate for patients with varying levels of risk across ten deciles. To effectively predict the medication drop-off rate, our experience has shown that an efficiency rate of more than 400% is needed.

Bottom line — a combination of the right analytics model and a high efficient rate ensures superior accuracy predicting which patients are likely to prematurely drop off medication therapy.

3. Proactive Actions Designing pilots and interventions for high-risk patients is a key step in determining the optimal approach to extend the length of time on therapy. With a succesful pilot tested on a small group of patients, the most optimal proactive interventions can be applied to the entire patient population. 

  • For the design: Ensure the pilot will test different interventions with enough patients in randomly selected test and control groups to detect a significant impact. At this stage, the design should make as few intervention changes as possible to the current program to keep the cost of the test minimal or nil.
  • To verify positive impact: Compare the average length on therapy (LOT) in a 3–12-month window between a randomly selected test group that has the designed interventions implemented, and a randomly selected control group without designed interventions implemented. Note that statistical significance may not be achieved if only a limited number of patients are allowed in the test group, though a directionally positive impact could be sufficient. 

4. Extending Length On Therapy (LOT) – If the selected test pilot intervention has shown acceptable positive impact on the test group, the final step is to expand that intervention to all high-risk patients — with the goal of extending the overall length on therapy by 5-10%.

Realizing the benefits

Especially in this current environment, there’s a tremendous need to cultivate and strengthen patient and HCP connections to improve prescription adherence and drive advocacy and education. As market dynamics continue to evolve, pharmaceutical companies are further increasing their use of adherence programs to provide behavioral health support, education and training to patients to improve outcomes and therapy adherence.

Predictive analytics is a critically important advancement that is quickly becoming a game-changer across the healthcare landscape. Given the availability of data (especially unstructured data) and the development of even more sophisticated machine learning techniques, predictive analytics is more important than ever.

The positive ROI with predictive analytics technology is incentivizing its adoption. The gains — not only in terms of adherence, but in other important areas such as hub, PAP and inside sales programs, where predictive analytics enables unprecedented levels of personalization, targeting and efficiency — are too big to ignore.

Learn more about Conduent’s solutions for the pharmaceutical and life sciences organizations. Get in touch with us at to discover how Predictive Analytics can improve the experience and outcomes for your patients.

About the Author

Hongguang is the Director of Analytics for Conduent. He has more than 20 years of comprehensive advanced statistical and Machine Learning analytics experience including program performance monitoring and impact analysis, sales force design and optimization, hub analytics, adherence analysis, Marketing Mix Modeling, and predictive analysis. Hongguang holds a Ph.D. in Statistics from North Carolina State University.

Profile Photo of Hongguang Sun