INFORMING OVERALL SURVIVAL EXTRAPOLATION IN HEALTH TECHNOLOGY ASSESSMENT USING STRUCTURED EXPERT ELICITATION: A NICE TSD 26 CASE STUDY
Author(s)
Kate Ren, PhD1, Jessica E. Forsyth, PhD2, Monika Achra, MSc3, Javier Baena Espinar, MD, PhD4, Maximilian Hochmair, MD5, Astra M. Liepa, PharmD3, Alan Lenox-Smith, MBBS, FRCP, FFPM, GFMD3, Tarun Puri, MD3, Rahul Rajesh, PharmD3, Yvonne Summers, BSc, MB ChB, MSc, PhD, FRCP6, Anthonie J. van der Wekken, MD PhD7, Min-Hua Jen, PhD8;
1University of Sheffield; ConnectHEOR Limited, Professor of Statistical Health Technology Assessment, Sheffield, United Kingdom, 2University of Sheffield, Sheffield, United Kingdom, 3Eli Lilly and Company, Indianapolis, IN, USA, 4Hospital 12 de Octubre, Madrid, Spain, 5Karl Landsteiner Institute for Lung Research and Pulmonary Oncology, Vienna, Austria, 6The Christie Hospital NHS Foundation Trust, Manchester, United Kingdom, 7University Medical Centre Groningen, Groningen, Netherlands, 8Eli Lilly, Uxbridge, United Kingdom
1University of Sheffield; ConnectHEOR Limited, Professor of Statistical Health Technology Assessment, Sheffield, United Kingdom, 2University of Sheffield, Sheffield, United Kingdom, 3Eli Lilly and Company, Indianapolis, IN, USA, 4Hospital 12 de Octubre, Madrid, Spain, 5Karl Landsteiner Institute for Lung Research and Pulmonary Oncology, Vienna, Austria, 6The Christie Hospital NHS Foundation Trust, Manchester, United Kingdom, 7University Medical Centre Groningen, Groningen, Netherlands, 8Eli Lilly, Uxbridge, United Kingdom
OBJECTIVES: Extrapolation of overall survival (OS) in health technology assessment (HTA) is a major source of decision uncertainty, as alternative statistical models with similar fits can generate markedly different lifetime projections. Although clinical expert input is often used to judge plausibility, this is rarely undertaken using formal structured expert elicitation (SEE). This case study applies National Institute for Health and Care Excellence (NICE) Decision Support Unit Technical Support Document 26 (TSD 26) guidance to elicit expert judgements of long-term survival probabilities and assesses feasibility within an HTA submission.
METHODS: The REVEL trial of ramucirumab plus docetaxel versus docetaxel alone in Stage IV non-small cell lung cancer was used. SEE was conducted using the Sheffield Elicitation Framework (SHELF). Five clinical experts from four countries participated in three virtual workshops. Experts provided probabilistic judgements of OS at four years post-randomisation for both treatment arms. Facilitated discussion explored assumptions about long-term hazard behaviour, with structured scenario testing to assess internal consistency.
RESULTS: Experts were able to provide probabilistic estimates of four-year OS for both arms, supported by qualitative rationale. These judgements were aggregated via mathematical aggregation due to their similarity. Discussion of hazard trajectories and scenario testing demonstrated internal consistency and contextualised survival expectations beyond observed follow-up. Practical challenges arose from conducting SEE across multiple workshops, particularly variable attendance and the need for facilitators to convey perspectives from absent experts.
CONCLUSIONS: This work demonstrates that implementing SEE for long-term survival outcomes in line with NICE TSD 26 is feasible within HTA. Explicit consideration of hazard behaviour is critical for contextualising expert judgements and should be prioritised in SEE design. Formal elicitation of expert survival expectations can reduce uncertainty around survival model choice, supporting more robust HTA decisions and potentially improving timely patient access to effective therapies.
METHODS: The REVEL trial of ramucirumab plus docetaxel versus docetaxel alone in Stage IV non-small cell lung cancer was used. SEE was conducted using the Sheffield Elicitation Framework (SHELF). Five clinical experts from four countries participated in three virtual workshops. Experts provided probabilistic judgements of OS at four years post-randomisation for both treatment arms. Facilitated discussion explored assumptions about long-term hazard behaviour, with structured scenario testing to assess internal consistency.
RESULTS: Experts were able to provide probabilistic estimates of four-year OS for both arms, supported by qualitative rationale. These judgements were aggregated via mathematical aggregation due to their similarity. Discussion of hazard trajectories and scenario testing demonstrated internal consistency and contextualised survival expectations beyond observed follow-up. Practical challenges arose from conducting SEE across multiple workshops, particularly variable attendance and the need for facilitators to convey perspectives from absent experts.
CONCLUSIONS: This work demonstrates that implementing SEE for long-term survival outcomes in line with NICE TSD 26 is feasible within HTA. Explicit consideration of hazard behaviour is critical for contextualising expert judgements and should be prioritised in SEE design. Formal elicitation of expert survival expectations can reduce uncertainty around survival model choice, supporting more robust HTA decisions and potentially improving timely patient access to effective therapies.
Conference/Value in Health Info
2026-05, ISPOR 2026, Philadelphia, PA, USA
Value in Health, Volume 29, Issue S6
Code
MSR136
Topic
Methodological & Statistical Research
Disease
No Additional Disease & Conditions/Specialized Treatment Areas