AI Meets GLP-1| How AI is Helping Pharma Marketers Understand Real Patient Journeys at Scale

Written by: Gaurav Kaushik, Ph.D., SVP & Head of AI, Veradigm, and Cheryl Reifsnyder, PhD

Initially approved for diabetes management, GLP-1 agonists (GLP-1s) are now widely used for weight loss and have also demonstrated potential in several other therapeutic areas.

Demand for GLP-1s has skyrocketed. From 2018 to 2024, 1.8 million unique patients received some type of GLP-1 medication. U.S. spending on prescription medication grew 11.4% in 2024 (up from 4.9% in 2023), a $50 billion increase, and GLP-1 medications accounted for 29% of this growth. During this same year, more than $40 billion in revenue came from just 2 GLP-1 medications, semaglutide and tirzepatide. Not only that: Demand for GLP-1s is predicted to continue growing for at least the next 5 years, reaching $105 billion in 2030, according to one forecast.¹

Yet despite the enthusiasm for GLP-1s, 20% to 50% of patients stop GLP-1 therapy within the first year.

Why? To answer this question, marketing teams require greater insight into how different patient populations relate to GLP-1 therapy. Which patients begin GLP-1 therapy? Which patients persist in treatment? Which discontinue? What circumstances or behaviors drive patient response and adherence?

Answering these questions is key to understanding and connecting with the healthcare providers who serve these patients, and ultimately key to knowing how to differentiate in an increasingly crowded market.¹

In this article, we share a recent case study illustrating how Veradigm machine learning (ML) experts developed science-based models that can generate fresh insights into the recently exploding GLP-1 therapeutic market-the type of pattern identification that can help your team take audience targeting to the next level.

GLP-1s: Compounding propensity model

In 2024, approximately 1/3 of the GLP-1 prescription volume was filled by compounders.

Compounding is the practice of formulating a copy of a commercial, brand-name medication to use in its place. The Food and Drug Administration (FDA) allows companies to market compounded versions of drugs only when there are shortages of the brand-name versions. In short supply since 2022, GLP-1s fit this scenario. Multiple compounding pharmacies have offered compounded GLP-1s, helping meet the growing demand for these medications.

Compounded GLP-1s: Benefits and drawbacks

Compounded GLP-1s improved patient access to GLP-1s, often at significantly lower costs than the branded versions. However, the compounded versions were not subject to the same FDA standards as brand-name medications, raising concerns about the drugs' consistency, safety, and efficacy.¹

Compounded GLP-1s are also associated with poorer outcomes. One study found that users of compounded semaglutide exhibited 8% lower weight loss than patients using brand-name semaglutide. This decreased clinical benefit is likely to result in patients taking compounded GLP-1s discontinuing therapy more quickly than patients using the brand-name drugs.¹

Changing availability of compounded GLP-1s

Now, though, GLP-1s are no longer on the FDA's shortage list. Production and sale of compounded GLP-1s is prohibited as of May 2025.

When you consider the rapidly changing availability of compounded GLP-1s, extracting insights specific to patient and provider segments becomes even more critical. With compounding prohibited, GLP-1 access patterns are changing. Some patients may transition smoothly from compounded to branded GLP-1s; others who previously used compounded GLP-1s (at lower cost), though, are at risk of being priced out of the market.¹

Gaining insight into which prescribers favored compounded GLP-1s could help marketers prevent patient drop-off as compounded products become unavailable. Similarly, a better understanding of which patients took compounded GLP-1s could help marketers better understand how patients' experiences with compounded medications may influence their future treatment decisions, adherence, and perception of branded medications.

With this information, teams could better shape messaging to address potential patient concerns stemming from use of compounded GLP-1, significantly impacting the number of patients who continue GLP-1 therapy.

AI proved the ideal tool to start answering these and other questions, to better understand patient choices and the factors driving them.

The key to reliable AI-generated insights

In this era of exploding healthcare data, AI has emerged as an essential tool for generating insights into patient and provider populations. However, the quality of AI-generated insights greatly depends on the quality of data available. AI algorithms need to be trained on high-quality, unbiased real-world datasets that include diverse data representative of the overall patient population.

Patient datasets vary significantly in their completeness, accuracy, and overall quality-which is why the Veradigm Network EHR dataset acts as an "AI power-up."

The Veradigm Network brings together 3 types of data:

• Healthcare provider data, sourced from electronic health records (EHRs) and provider-facing tools in thousands of practices
• Point-of-care connectivity data, sourced from products such as secure messaging and patient engagement tools, directly within the EHR
• Scalable data from specialty registries and disease-focused datasets

Drawing from these 3 sources, Veradigm Network EHR data includes patients of diverse ages, demographic backgrounds, and geographies-over 152 million unique patients from all 50 states. Using data on this scale enables us to identify demographic and regional patterns reflective of "main street" American healthcare.¹

Figure 1 5 Year Time Period. Q1 2020-Q4 2024. Veradigm Data on File, June 2024.

Veradigm Network patient records are sourced from a comprehensive network of ambulatory EHR systems serving over 240 thousand clinicians from across the U.S. Clinicians represent a diverse range of specialties from primary care physicians to dozens of clinical specialties across an expansive range of therapeutic areas. This breadth of physician specialty is vital when looking into a drug with uses spanning obesity, diabetes, cardiology, and mental health domains (such as GLP-1s).¹

Figure 2 <2% of total clinicians are grouped into other. 5 Year Time Period. Q1 2020-Q4 2024. Veradigm Data on File, June 2024.

Using this diverse dataset, we wanted to use AI to gain a deeper understanding of both the patients who took compounded versus branded GLP-1s and the prescribers who favored compounded GLP-1s to prescribe to their patients.¹

Leveraging AI to gain insight into GLP-1 usage

We wanted to better understand how patients were using GLP-1s and how those behavior patterns impacted their outcomes. We began by using AI to transform Veradigm Network EHR data into analysis-ready datasets to understand patient adherence and sentiment towards specific medications. Instead of relying solely on structured EHR data, we trained AI models to evaluate patient notes and detect implicit signals-such as topics of patient complaints and patients' reported experiences with specific treatment-that cannot be found in structured EHR fields. Deciphering unstructured data (which accounts for roughly 80% of patient information in EHRs) allows for more precise data analysis.¹

Next, we examined segmenting providers and patients by their real-world behaviors. We evaluated the segments to answer questions such as:¹

• How do providers' prescribing behaviors impact patient treatment adherence to treatment?
• How are prescribing behaviors influenced by changes in the availability of compounded GLP-1s?

Using Veradigm Network EHR data, we trained and evaluated several models. The best model predicted, with more than 90% accuracy, which prescribers favored compounded GLP-1s. The segmented data also revealed the specific factors driving this behavior.¹

The result? We found that physicians' geography, patient volumes, and practice sizes all strongly influenced whether they prescribed compounded or branded GLP-1s.¹

Similarly, finding relevant and informative patient segments revealed deeper insights into patient treatment choices and how these choices reflected broader behavioral patterns and the availability of compounded versus branded GLP-1s.¹

The goal was to predict which patients would adhere to GLP-1 treatment and achieve positive outcomes, and to identify those most likely to respond to specific prescriptions. This predictive capability can help inform which medication is likely to lead to better adherence and outcomes.¹

Beyond GLP-1s: Point-of-care precision

This article explores how in-depth analytics can provide greater insight into the treatment journey for patients using GLP-1s. By applying AI to the Veradigm Network EHR dataset's broad and diverse data, we uncovered precise insights into how GLP-1s are prescribed and used across multiple therapeutic areas.

Importantly, these insights and these techniques are disease-agnostic. They can be applied across multiple conditions, indications, and patient journeys. The analysis extends beyond identifying which physicians prescribed which medications to reveal the intent and context that shape prescribing behaviors.

Our work with GLP-1s is only one example of what Veradigm's AI expertise can achieve. By combining structured and unstructured data, we create dynamic, behavior-based provider and patient segments. From these segments, life science companies can derive precision insights that support tailored, timely interventions to strengthen care decisions and patient journeys.

AI is a key driver of the future of healthcare intelligence and operations at Veradigm. To learn how our AI expertise can empower your team to improve patient and provider engagement, contact us today.

References:

1. Kaushik, G. AI Meets GLP-1: How AI is Helping Marketers Understand Real Patient Journeys at Scale. Future Pharma, June 26, 2025. Available on Request.