Presentation (CTI)

OBJECTIVES: To map where RWE and data science methods are used across the evidence-generation lifecycle in ISPOR 2025 AI-focused research and to define priorities for decision-grade adoption.
METHODS: We performed a taxonomy-based scoping review of 230 abstracts presented across the three ISPOR scientific conferences held in 2025, retrieved from the ISPOR abstracts repository using a structured keyword strategy (AI/ML, NLP, LLM/GenAI, agentic systems, automation). Each abstract was charted into a standardized framework capturing evidence-generation phase/sub-phase and data science branch/sub-branch. Charting used LLM-assisted extraction with rule-based label harmonization and completeness checks. We summarized method distributions by evidence phase.
RESULTS: Most abstracts addressed Post-Launch Evidence (RWE/Medical Affairs) (88/230; 38.3%), Market Access & HTA (Value Demonstration) (73/230; 31.7%), or Cross-Cutting Methods/Infrastructure (54/230; 23.5%). GenAI/LLMs were most prevalent (61/230; 26.5%), followed by Predictive Modeling (47/230; 20.4%), Decision Intelligence/Automation (39/230; 17.0%), Descriptive/Statistical Analytics (37/230; 16.1%), and NLP (30/230; 13.0%); Agentic AI (10/230; 4.3%) and Causal Inference (6/230; 2.6%) were least frequent. GenAI/LLM work clustered in evidence synthesis and dossier/report workflows, whereas predictive models clustered in post-launch safety, adherence/persistence, and utilization. Reporting of external validation, calibration/uncertainty, and traceability/governance needed for decision-grade deployment was often limited.
CONCLUSIONS: ISPOR 2025 shows rapid uptake of GenAI-enabled evidence workflows but relatively limited causal inference. Actionable priorities are: (1) minimum reporting standards for AI-enabled RWE (data provenance, confounding strategy, performance, external validation, calibration/uncertainty, fairness); (2) traceability-by-design for GenAI (source-linked evidence, versioning, human-in-the-loop review); and (3) shared benchmarks and head-to-head evaluations to support HTA- and regulatory-ready deployment.