How AI Is unlocking new insights from claims and EHR data in Medical Affairs

Medical Affairs is being transformed by AI, unlocking proactive strategies that redefine evidence generation, market shaping, and patient outcomes. AI, particularly large language models (LLMs), enables the integration of structured claims data with unstructured electronic health record (EHR) narratives, facilitating real-time, proactive decision-making. Claims data provides a macro-level view of healthcare utilization, costs, and treatment pathways, while EHRs capture clinical nuances such as lab results and physician notes. AI is revolutionizing insights by dynamically detecting patterns, anticipating patient needs, and shaping evidence strategies in real time. In this piece, we will briefly examine 5 emerging use cases for AI and LLMs and their potential impact for Medical Affairs teams.

1. Unlocking deeper insights from unstructured data

Traditionally, analyses have relied on structured fields (e.g., diagnosis codes, medication lists, and billing details) to infer trends. However, crucial signals often reside in unstructured text. LLMs can parse these free-text narratives to pick up on subtle clinical indicators and can even identify synonyms or colloquialisms that might be used differently by different clinicians. By converting unstructured data into structured insights, LLMs open the door to richer analyses. This allows teams to better detect patterns of disease burden, track emerging side effects, or forecast when patients are at risk of non-adherence.

A notable example is AI-driven seizure prediction, where an LLM trained on seizure-related clinical notes from Boston Children’s Hospital and the MIMIC-III database outperformed traditional analytics,^[1] identifying patterns that help predict recurrence risk. This demonstrates how AI can mine physician notes for actionable insights, empowering Medical Affairs teams to better understand physician decision-making and patient experiences, ultimately guiding evidence-based engagement.

2. Generating patient cohorts for real-world evidence (RWE) studies faster

Real-world evidence studies are a cornerstone within Medical Affairs, providing insights into how therapies perform outside the controlled environment of clinical trials. Yet identifying the right patient cohorts for these studies can be time-consuming, especially in niche or rare diseases. AI is enabling real-time patient cohort identification, drastically reducing study timelines and making dynamic trial designs a reality. This can accelerate cohort identification, enabling faster design and study execution. It also helps unearth those patients who might not be flagged by claims data alone, which can be especially beneficial in rare disease areas where data scarcity is a frequent and major challenge.^[2],[3]

3. Synthetic control groups using AI

AI-driven synthetic control groups are reshaping clinical trial design by leveraging machine learning models to create simulated control arms from RWE, reducing the need for traditional placebo groups. As an example, Flatiron Health employed AI-driven EHR data to develop synthetic control arms for oncology drug trials, a methodology that received FDA acceptance for regulatory submissions, thereby validating its potential to supplement or even replace conventional controls. ^[4] These AI models, trained on historical patient data, facilitated precise matching of real-world patients to trial participants, ensuring robust comparative analyses while lowering trial costs and addressing ethical concerns associated with placebo use. This highlights how AI models, trained on historical patient data, facilitate precise matching of real-world patients to trial participants, ensuring robust comparative analyses while lowering trial costs and addressing ethical concerns.

4. AI-powered predictive modeling for patient outcomes

Beyond basic descriptive analysis, LLMs can feed into predictive models that investigate disease progression, therapy adherence, or event risk (e.g., hospitalization). These models often incorporate data from common (e.g., claims, EHR) and less common data sources (e.g., social determinants like socioeconomic status, and education level). A key difference between modern AI models and traditional approaches lies in feature engineering. Previously, analysts had to manually determine which variables to include (e.g., BMI, or prescription refills per month). LLMs are better at dynamically identifying patterns across multimodal data, teasing out correlations that might go overlooked by human analysts.

Case Study: Using AI to identify health disparities in patient care

Challenge: Social Determinants of Health (SDoH) play a critical role in patient outcomes, yet they are rarely documented in EHR fields. Identifying disparities in care requires analyzing free-text physician notes, a historically manual and time-intensive process.

AI Application: Putnam deployed an AI model that extracted SDoH mentions from unstructured EHR data, identifying patterns in access to care, treatment adherence, and healthcare disparities. To reduce bias, the model was fine-tuned using both synthetic and gold-standard training data.

Impact on Medical Affairs: This approach enabled Medical Affairs teams to better understand patient populations and address disparities in care; aligning scientific engagement with broader public health initiatives and ensuring underrepresented populations are not overlooked.

Combining multimodal insight data to demonstrate impact One of the most powerful advantages of LLMs is their ability to derive meaningful insights from multiple data sources (e.g., such as real-world evidence, claims data, MSL conversation notes, and congress reports) and distill complex information into concise, human-readable summaries. By layering these streams together, Medical Affairs teams identify which patient populations and physicians are most impacted, pinpoint emerging care gaps, and develop more a robust way to analyze impact on patients. This integrated approach enables them to act swiftly on data-driven recommendations. Rather than spending excessive time interpreting raw data, stakeholders can hold real-time discussions to refine strategies for patient engagement, physician education, or market access.^[5] We are embarking on the next evolution of claims data integration with AI-driven medical insights to identify areas of care gaps. For example, patient journey mapping from claims data layered with medical insights enables us to determine the medical content needed for providers so they can make informed care decisions. In another example, claims data is being used as an early impact indicator of AI-driven medical strategies on provider behaviors and outcomes.

Ensuring AI benefits all patients

Although AI-driven analytics are revolutionizing how healthcare data is analyzed, the benefits are not uniformly distributed. In high-income countries, where data infrastructure is relatively robust and interoperability standards like ICD-10 are well-established, it’s easier to deploy advanced AI models at scale. In contrast, many low- and middle-income countries still grapple with inconsistent data collection, fragmented systems, and minimal digital health infrastructure making it more challenging to implement sophisticated AI solutions. Additionally, many AI models are trained predominantly on English-language datasets, which can lead to gaps when applied to regions where medical records are written in other languages or where terminology usage is different. Cultural factors can also influence how symptoms are described or how patients report their conditions, further complicating AI’s ability to generalize across borders.^[6],[7] To ensure AI benefits all patients equally, companies and stakeholders must invest in multilingual AI models, localized training data, and capacity-building initiatives that empower healthcare providers in diverse regions. By adopting inclusive design principles and addressing potential biases at the outset, Medical Affairs teams can help prevent disparities in the next wave of AI-driven healthcare innovation.

The future of AI in Medical Affairs

The ability to harness real-time, data-driven insights has never been more critical for Medical Affairs. AI, and LLMs, bring the promise of analyzing claims and EHR data in a holistic manner, unearthing patterns and insights that were previously obscured by data fragmentation or unstructured text. This transformation moves healthcare analytics from a retrospective look at “what happened” to a forward-thinking perspective on “what’s next?”

The path forward isn’t without challenges, as we will need to ensure AI solutions are globally equitable, culturally sensitive, and regulatory-compliant will require collaboration across industries and geographies. Therefore, investing in robust data infrastructure, federated learning models, and bias mitigation strategies will be key to realizing AI’s full potential.

For Medical Affairs teams embracing AI-driven analytics is no longer optional. It is a critical capability to thrive in a data-rich, rapidly evolving healthcare environment. By leveraging LLMs and other advanced AI tools to integrate structured claims data with nuanced EHR information, Medical Affairs can deliver deeper insights faster shaping better outcomes not just for specific patient cohorts, but for entire healthcare ecosystems worldwide.

At Putnam, we recognize that standard commercial AI tools can fall short when it comes to addressing the nuanced challenges of Medical Affairs. That’s why we offer bespoke AI solutions designed specifically for Medical Affairs with AI-ready data. Putnam’s Data Gateway offering seamlessly integrates client datasets into a multi-sourced tokenized database enabling them to extract faster and higher quality insights that drive strategic decision-making and refine real-world evidence generation. With deep industry expertise at every step, we help you navigate data complexities so you can focus on delivering impactful outcomes.

References:

1. Beaulieu-Jones, B. K., Villamar, M. F., Scordis, P., Bartmann, A. P., Ali, W., Wissel, B. D., Alsentzer, E., de Jong, J., Patra, A., & Kohane, I. (2023). Predicting seizure recurrence after an initial seizure-like episode from routine clinical notes using large language models: A retrospective cohort study. Lancet Digital Health, 5(12), e882–e894 https://pmc.ncbi.nlm.nih.gov/articles/PMC10695164/
2. Bryant, A. K., Zamora-Resendiz, R., Dai, X., Morrow, D., Lin, Y., Jungles, K. M., Rae, J. M., Tate, A., Pearson, A. N., Jiang, R., Fritsche, L., Lawrence, T. S., Zou, W., Schipper, M., Ramnath, N., Yoo, S., Crivelli, S., & Green, M. D. (2024). Artificial intelligence to unlock real-world evidence in clinical oncology: A primer on recent advances. Cancer Medicine, 13(12), e7253 https://pmc.ncbi.nlm.nih.gov/articles/PMC11187737/
3. He, D., Wang, R., Xu, Z., Wang, J., Song, P., Wang, H., & Su, J. (2024). The use of artificial intelligence in the treatment of rare diseases: A scoping review. Intractable & Rare Diseases Research, 13(1), 12–22 https://pmc.ncbi.nlm.nih.gov/articles/PMC10883845/
4. Tan, K., Bryan, J., Segal, B., Bellomo, L., Nussbaum, N., Tucker, M., Torres, A. Z., Bennette, C., Capra, W., Curtis, M., & Miksad, R. A. (2021). Emulating control arms for cancer clinical trials using external cohorts created from electronic health record-derived real-world data. Clinical Pharmacology & Therapeutics, 111(1), 168–178; Huzman, K. T. (2019, July 12). BioCentury – Broadening role for external control arms in clinical trials. Friends of Cancer Research. https://friendsofcancerresearch.org/news/biocentury-broadening-role-for-external-control-arms-in-clinical-trials/ https://pmc.ncbi.nlm.nih.gov/articles/PMC9292216/
5. Fröling, E., Rajaeean, N., Hinrichsmeyer, K. S., Domrös-Zoungrana, D., Urban, J. N., & Lenz, C. (2024). Artificial intelligence in Medical Affairs: A new paradigm with novel opportunities. Pharmaceutical Medicine, 38(5), 331–342 https://pmc.ncbi.nlm.nih.gov/articles/PMC11473552
6. Bryant et al. (2024) https://pmc.ncbi.nlm.nih.gov/articles/PMC11187737
7. Alami, H., Rivard, L., Lehoux, P., Hoffman, S. J., Cadeddu, S. B. M., Savoldelli, M., Samri, M. A., Ag Ahmed, M. A., Fleet, R., & Fortin, J.-P. (2020). Artificial intelligence in health care: Laying the foundation for responsible, sustainable, and inclusive innovation in low- and middle-income countries. Globalization and Health, 16, Article 52 https://globalizationandhealth.biomedcentral.com/articles/10.1186/s12992-020-00584-1