EVALUATING THE IMPACT OF LARGE LANGUAGE MODELS ON THE DEVELOPMENT OF COST-EFFECTIVENESS MODELS FOR EARLY HEALTH TECHNOLOGY ASSESSMENT

Author(s)

Raymond H. Henderson, MSc, PhD, Jay Bilimoria, PhD, Holly Gould, MSc, Heritage Kristilere, MPH, MD, Tahera Patel, MSc, Hannah Rice, BSc, Michal Witkowski, MSc, Alison Martin, MSc, MD.
Crystallise Ltd, Colchester, United Kingdom.
OBJECTIVES: Cost-effectiveness models (CEMs) are routinely developed to inform early health technology assessment, pricing, and reimbursement strategy. Recent advances in large language models (LLMs) offer opportunities to automate or accelerate several modelling activities. This study evaluated the impact of LLM-assisted workflows on the time required to develop a de novo CEM and accompanying go/no-go pricing tool.
METHODS: A de novo oncology CEM and associated go/no-go model were developed using a combination of conventional health economic methods and LLM-assisted workflows. Tasks included evidence scoping, survival curve reconstruction and extrapolation, cost and utility identification, worksheet development, sensitivity analyses, and report drafting. LLMs used included ChatGPT and Claude, supplemented by R, VBA, Draw.io, WebPlotDigitizer, and online evidence sources. Time spent on each activity was prospectively recorded and compared with a typical CEM development timeline of approximately four weeks.
RESULTS: Total active development time was 68.5 hours. The most time-intensive activities were survival extrapolation (13 hours), worksheet linking (10 hours), go/no-go model development (8.5 hours), and evidence synthesis for costs and resource use (7 hours). LLMs contributed across most stages, including code generation, formula development, evidence summarisation, documentation, and quality checking. Compared with a conventional development estimate of approximately 160 hours, the LLM-assisted workflow reduced analyst time by 91.5 hours, representing a 57% reduction. The largest efficiencies were observed in programming-intensive tasks, model construction, and drafting activities.
CONCLUSIONS: LLMs may substantially reduce the time required to develop early-stage health economic models while supporting standard modelling workflows. Future research should evaluate the reproducibility, accuracy, and quality assurance requirements of LLM-assisted model development and establish best-practice guidance for their use in health economics.

Conference/Value in Health Info

2026-11, ISPOR Europe 2026, Vienna, Austria

Value in Health, Volume 29, Issue 12S

Code

EE92

Topic

Economic Evaluation, Health Technology Assessment, Study Approaches

Disease

Oncology, Personalized & Precision Medicine

Your browser is out-of-date

ISPOR recommends that you update your browser for more security, speed and the best experience on ispor.org. Update my browser now

×