Skip to main content
Search
Search
Advanced Search
Skip main navigation
The Evidence Base Post

ISPOR 2025 daily round-ups: Day 2

  • Joanne Walker, Katie McCool & Laura Dormer

At ISPOR 2025 (May 13–16, Montreal, Canada), hosted by ISPOR—The Professional Society for Health Economics and Outcomes Research, Day 1 sparked many insightful conversations and lively debate. From the opening plenary on drug price controls and innovation, to sessions on joint clinical assessments, cross-border collaboration, real-world life-cycle evaluation in precision medicine, and the evolving intersection between health economics and outcomes research (HEOR) and venture capital, delegates engaged with some of the most pressing issues shaping the future of health economics and outcomes research.

As the conference continues into Day 2, we’re highlighting some of the key discussions and emerging themes shaping today’s agenda.

Welcome Remarks and Presidential Addresses

The second day of ISPOR 2025 began with a welcome from Dr Eberechukwu Onukwugha (University of Maryland), President of ISPOR—The Professional Society for Health Economics and Outcomes Research. Onukwugha opened the day’s sessions by reflecting on the vital role of HEOR in public discourse and policy, emphasizing that, “HEOR is too important to be left to only us.” She highlighted achievements during her term, including the establishment of a working group for the ISPOR Institute and the introduction of a new conference session to spark deeper engagement with core methodological topics. Acknowledging today’s global challenges, she affirmed the ongoing need for strong, evidence-based healthcare decision-making and expressed deep gratitude for the opportunity to serve the ISPOR community.

Incoming President Dr Uwe Siebert (UMIT TIROL; Harvard Chan School) shared his vision for the Society, centered on five key pillars: scientific rigor, inclusiveness and sustainability, education, catalyzing connections, and globalization. He emphasized collaboration across disciplines and geographies, and closed with a warm welcome to ISPOR’s new President-Elect, Dr Beth Devine (University of Washington in Seattle).

Plenary 2: Evolution of Evidence—Innovating for the Future of HTA

As health systems face growing complexity, workforce pressures, and demands for patient-centered care, the second plenary at ISPOR 2025 explored how health technology assessment (HTA) must evolve to remain impactful. Moderated by Nicole Mittmann (Canada’s Drug Agency), the session brought together a distinguished panel representing researcher, agency and patient advocacy perspectives, including Sudha Kutty (Canada's Drug Agency), Pascale Lehoux (INESSS, Canada), Donald Husereau (University of Ottawa, Canada) and Connie Cote (Health Charities Coalition of Canada, Canada).

Panelists highlighted efforts to embed patient voices more meaningfully, accelerate access to promising therapies, and align HTA with sustainability and equity goals. They called for a shift from supply-driven to needs-driven innovation and emphasized the importance of collaboration, transparency and inclusivity. With strategic plans and pilot initiatives already underway in Canada, the session offered a look both at where we are now, and shared aspirations for what future HTA systems might look like.

Read our in-depth summary of Plenary 2 here >>>

Spotlight Session: Prompt Engineering: Harnessing Generative AI for HEOR

This workshop-style session explored the practical use of generative AI (GenAI) and large language models (LLMs) in HEOR.

Rachael Fleurence (National Institutes of Health) opened the session with an overview of GenAI capabilities, including how LLMs retain and apply context to generate coherent outputs. She outlined four emerging use cases in HEOR: systematic literature reviews (SLRs), health economic modeling, analysis of unstructured real-world data (RWD), and dossier development. While GenAI is particularly well-suited to text-based tasks such as screening abstracts and extracting data, Fleurence cautioned that limitations around accuracy, comprehensiveness, reproducibility, and privacy must be carefully managed. She emphasized that GenAI should be used with human oversight and noted that performance continues to improve alongside growing user expertise.

Turgay Ayer (Value Analytics Labs) provided a practical overview of prompt engineering methods. He introduced several core techniques, zero-shot, few-shot, chain-of-thought, tree-of-thought, persona, and pattern prompting, and explained how these can be strategically combined to guide LLMs more effectively. To illustrate these approaches, he used a simple example: asking an AI model to explain the benefits of exercise. In a zero-shot prompt, the model is simply instructed to list benefits; in few-shot prompting, it is given examples before being asked to generate its own; and in chain-of-thought prompting, the task is broken into logical steps such as first explaining why exercise matters, then listing physical, mental, and social benefits. He also demonstrated how tree-of-thought prompting could explore different dimensions of a topic simultaneously, and how persona prompting, such as asking the AI to respond as a nutritionist, can influence the tone and scope of the answer. Ayer emphasized best practices, including the importance of clarity, providing relevant context, defining constraints, and structuring prompts to reflect the desired output.

Jag Chhatwal (Harvard Medical School / Massachusetts General Hospital) presented several applied use cases, including the use of GenAI to support SLR workflows and to assist in health economic modeling. Using GenAI tools, he demonstrated how prompts could assist in building literature search strategies, screening abstracts, and extracting structured data from published articles. He also shared an example of using GenAI to support the early stages of Markov model development. While the outputs were promising, Chhatwal noted that results varied depending on input complexity and required manual verification. Like Fleurence and Ayer, he underscored the need for guardrails and human validation, particularly in modeling, where current general-purpose tools may lack domain-specific reliability. He also emphasized the value of HEOR-specific AI tools, such as ValueGen.AI, which incorporate domain knowledge and validation mechanisms.

Concurrent Sessions

Day 2 of ISPOR 2025 featured many concurrent sessions across the spectrum of HEOR research. Below are highlights of the session attended by The Evidence Base.

Issue Panel: AI Agents and Guardrails in HEOR: The Ultimate Solution to GenAI Shortcomings or Just Another Overhyped Tool?

Moderated by Foluso O Agboola (Institute for Clinical and Economic Review), this issue panel explored both the transformative potential and the necessary safeguards surrounding the use of artificial intelligence (AI) agents in HEOR. Agboola opened with a brief overview of AI in HEOR, referencing ISPOR’s recent publication and highlighting the accelerating pace of research due to GenAI.

Speaking next, Sven L Klijn (Bristol Myers Squibb) introduced the concept of agentic AI, differentiating it from traditional AI interactions. Unlike task-specific applications, agentic AI systems operate autonomously with goal-setting, adaptive behavior and environmental awareness. Klijn presented use cases including dossier development, “living” HTAs, and simulation of review committee discussions – while stressing the need for rigorous implementation, transparency and lifecycle management.

Tim Disher (Sandpiper Analytics) offered a cautionary view, raising concerns around validation complexity, workforce deskilling and ethical risks. He noted that AI agents could raise expectations for submissions but risk creating formulaic, less insightful outputs. Disher called for smarter validation, benchmarks and prioritization of human expertise in areas where nuance matters.

Ghayath Janoudi (Loon) concluded with a framework for AI guardrails – spanning scientific, technical and agentic considerations. He urged stakeholders to adopt a risk-based mindset and emphasized reproducibility as the ultimate safeguard in HEOR.

The session offered a balanced and pragmatic perspective on advancing AI use in HEOR – encouraging innovation while reinforcing accountability. 

Issue Panel: IRA Under Trump: What Is Next?

This discussion explored how the Inflation Reduction Act (IRA) could shift under a Trump administration and unified Republican Congress, with potential revisions to Medicare drug price negotiation, Part D redesign, orphan drug and qualified single-source drug (QSSD) definitions, and the possible use of international reference pricing. Speakers also examined the broader implications for innovation, value assessment, and patient access.

Lisa Joldersma (Avalere Health) emphasized that the administration is intensifying its approach: “They are full in on negotiation… taking the most aggressive interpretation you can possibly take.” She noted that even products, “under a different New Drug Application (NDA)… provided [they share] methods from the same manufacturer,” may now be grouped under QSSD definitions. She warned that renegotiation could be triggered by new indications and added, “This issue is not done… We could come back. We are just starting to see data.” On cost savings, she remarked, “We know it’s never the patient, and it should be.”

Ulrich Neumann (Johnson & Johnson) presented evidence of emerging trade-offs. “In 2025 there were 464 [Medicare Prescription Drug Plans]… a dramatic 35% decrease in patient choices,” he said, linking plan exits and formulary tightening to unintended consequences. He warned of, “disruptions in the R&D market that hadn’t been there before” and emphasized that, “post-approval research is not an afterthought… that’s the core of innovation.”

Moderator Michael Ciarametaro (Avalere Health) highlighted system-wide pressures on Medicare plan design and pricing: “They can raise premiums, they can try to extract additional rebates, or they can shift their benefit design.” He cautioned, “If your standalone Part D market goes away, the viability of the fee-for-service market goes away,” and described recent policy actions as, “a carrot and stick approach.”

Exhibit Hall Theater: Is IRA Becoming America’s HTA?

In this session exploring the evolving role of the IRA in US drug pricing, Priti Jhingran (Genesis Research Group) opened by questioning whether Medicare’s Drug Price Negotiation Program (DPNP) is beginning to function like a health technology assessment (HTA) body. She noted that while the US pharmaceutical pricing environment remains turbulent, especially following President Trump’s recent executive order, the IRA introduces HTA-like negotiation mechanisms that warrant close examination.

Jhingran outlined how IRA evidence dossiers mirror global HTA submissions, covering clinical benefit, comparative effectiveness, cost offsets, and real-world evidence (RWE). However, key differences remain: the use of QALYs is banned, budget impact is not considered, and CMS does not determine coverage. Still, she emphasized that RWE plays an even more central role under IRA, given the availability of post-launch data.

She also discussed the ambiguous but growing influence of the Institute for Clinical and Economic Review (ICER), whose reports, excluding QALY-based findings, are increasingly referenced during IRA negotiations, despite ICER having no formal role in the process.

Tijana Ignjatovic (Genesis Research Group) addressed the implications of IRA on product lifecycle strategy. She emphasized that IRA evaluations occur years after launch and disproportionately affect small molecules, prompting legislative efforts like the reintroduced EPIC Act to extend their negotiation timeline.

Ignjatovic warned of unintended consequences from single pricing across all indications, potentially deterring indication expansions. She shared lessons from the first negotiation cycle, highlighting the importance of comparative data, robust RWE, and subgroup analyses. Conversely, justifications based on R&D spend, novelty claims, or anecdotal patient input were seen as less impactful. Ignjatovic concluded by stressing the need for proactive planning to navigate IRA’s evolving demands, mirroring the conclusions from previous discussions on the IRA from Day 1 of ISPOR 2025.

Workshop: Identifying Gaps and Establishing a Development Plan for Consensus Real-World Data Standards

Experts examined the need for consensus RWD standards in HEOR, outlining key categories such as documentation, organizational practices, and metadata. They highlighted challenges in applying standards to diverse sources including claims, electronic health records (EHRs), wearables, and AI-driven tools. The discussion focused on improving data quality, ensuring interoperability, and establishing a consistent framework to support credible, reusable evidence in healthcare research and decision-making.

Mary Beth Ritchey (CERobs Consulting) emphasized the importance of structured, interoperable data standards, explaining, “When we have data standards, we have clear guidelines that allow us to have uniformity in the data, to allow us to be consistent and to allow us to share that data back and forth.” She raised concerns about current limitations, noting, “There’s places that exist for the data, but the standards don’t extend to how the data are going to be collected,” particularly in areas like device exposure and payer information.

Outgoing ISPOR President, Eberechukwu Onukwugha (University of Maryland), highlighted the need for structural standards to support RWD use in HEOR, stating, “We don’t yet have a building… to support HEOR, where we're taking study design, layering it onto data, and then trying to draw evidence from that.” She emphasized the importance of cross-system compatibility and added,

“Ultimately, we need a user centred design perspective when it comes to building databases that are fit for our purpose as HEOR scientists.”

Mitch Higashi (ISPOR) emphasized the role of metadata standards in supporting the responsible use of generative AI in healthcare research. He explained that metadata helps ensure, “improved data quality leading to more accurate and trustworthy AI generated insights,” and noted that five of ten evaluation domains in ISPOR’s Elevate AI framework “are highly dependent on data quality.” These include factuality verification, reproducibility, fairness and bias, calibration, and privacy. “We are now at a tipping point,” he said, pointing to the growing use of generative AI in literature reviews and decision-making, and calling for standards that support, “efficient data management, accelerating the processing of large volumes of healthcare data at scale.”

Forum: Clinical Trial Design, Beyond the Regulatory

This Forum session, moderated by Kati Copley-Merriman (RTI Health Solutions, USA), addressed how clinical trial design can (and must) go beyond regulatory requirements to support value assessment, reimbursement and broader healthcare decision-making. Copley-Merriman emphasized the importance of aligning trial design with payer needs and laid out the session’s focus on patient-centered trials, economic modeling and external control arms (ECAs).

Lynda Doward (RTI Health Solutions, UK) opened by urging attendees to design trials, “with the end in mind,” particularly in relation to patient-centered evidence. She encouraged the audience to view patient-generated data through the broader lens of patient experience data (PED), which can enhance reimbursement dossiers, inform economic models, and support patient–provider decision-making. Doward highlighted challenges such as underrepresentation, misaligned endpoints, and internal resistance to collecting robust PED, illustrated with a real-world case study from Canada’s Drug Agency.

Emma Hawe (LCP Health Analytics, UK) discussed the importance of designing trials that can feed into health economic models. Using the PICOs framework, she walked through strategies to enable indirect treatment comparisons and collect preference-based utility data, including the timing and selection of PRO instruments aligned with economic model requirements.

Shannon Cope (PRECISIONheor, Canada) concluded by examining ECA design from an HTA perspective, using the LocoMMotion trial as a case study. She outlined key trial elements – such as target trial emulation, treatment assignment periods and causal contrasts – and highlighted tools like HARPER and suitability checklists to guide robust design.

The session emphasized that early, thoughtful trial planning is key to ensuring post-regulatory success across HTA and access landscapes.

Workshop: Personalizing Race and Ethnicity Data to Improve Real-World Evidence Relevance and Reliability

Lysel Brignoli (Oracle Life Sciences) opened the session by highlighting the importance—and challenges—of collecting race and ethnicity data in healthcare. While the two terms are often conflated, Brignoli clarified that race is a social construct tied to perceived physical traits, whereas ethnicity refers to shared cultural traditions. Despite their limitations, such data remain critical proxies for identifying gaps in care and understanding unmet health needs. However, RWD sources, such as EHRs and claims, often suffer from missing, inconsistent, or misclassified race and ethnicity data due to non-standardized categories and unclear provenance.

Martina Furegato (Oracle Life Sciences) reviewed the types of administrative secondary data sources available, including commercial claims, Medicare, Medicaid, and EHRs. She emphasized that EHRs tend to be more reliable, though they are not immune to undercounting or data entry errors. Oracle-conducted validations have shown moderate concordance with national surveys, but challenges persist in standardizing collection methods and addressing human factors in data capture.

Rachele Hendricks-Sturrup (Duke-Margolis Institute for Health Policy) explored the value of integrating patient-generated health data with EHRs to enhance trustworthiness and regulatory relevance. She referenced recent Duke-Margolis white papers on learning health systems and subgroup representation, emphasizing that racial misclassification and lack of patient trust remain key barriers. Her research using NIH’s All of Us dataset found pharmacogenomic variants disproportionately affecting individuals of African ancestry, highlighting the potential for more tailored care if systems embrace more nuanced, patient-led reporting.

Carla Rodriguez-Watson (Reagan-Udall Foundation) introduced the Real-world Accelerator to Improve the Standard of collection and curation of race and Ethnicity (RAISE Action Framework), a structured approach developed to improve the collection, curation, and exchange of race and ethnicity data. The framework identifies four priority areas: (1) standardize data collection, (2) train the workforce, (3) incentivize data capture, and (4) collect data locally then aggregate. These are supported by actionable strategies promoting cultural humility, data transparency, and system interoperability.

Issue Panel: How Do We Generate RWE in Rare Diseases or Targeted Subgroups? Use of Multimodal Data and Methodologies

This session explored methodological strategies for generating RWE in rare diseases and targeted subgroups, focusing on multimodal data sources and innovative approaches. Moderator Phani Veeranki (Optum Life Sciences) introduced key challenges including diagnostic delays, data fragmentation, and small, heterogeneous patient populations. “How do we use different data assets and multimodal data and different methodological approaches to generate this robust body of evidence?” he asked.

Arpita Nag (AstraZeneca) outlined a range of RWD sources used to study rare diseases, including patient registries, EHRs, claims databases, chargemaster data, chart reviews, and digital studies. She emphasized the value of patient support programs for early evidence generation, noting, “Before your medicine gets into the market, where you see a good critical mass in your claims, this is your way to go.”

Amy Duhig (Takeda) highlighted the importance of combining quantitative and qualitative methodologies, such as grounded theory and mixed-methods approaches, to better understand patient journeys in rare diseases. “Just because somebody has a disease doesn’t mean that they experience the disease the same way,” she noted, emphasizing the value of patient-centered insights alongside traditional data analyses.

Finally, Lucinda Orsini (COMPASS Pathways) explored the use of tokenization to enable privacy-preserving linkages across clinical trial, registry, and RWD sources. “We as researchers are really interested in these longer time horizons, and this is one way to try to understand and link together a really small window of time into a much broader ecosystem of data,” she explained, highlighting its value in extending insights beyond trial settings.

Other ISPOR 2025 sessions

For those not able to attend sessions in person, ISPOR is once again providing on-demand access through its popular Digital Conference Pass. Other sessions from Day 2 include:

Register for free today to become a member of The Evidence Base and receive the latest news straight to your inbox.