BRIDGING THE GAP: ASSESSING THE PREDICTIVE ACCURACY OF EXPERT ELICITATION FOR LONG-TERM SURVIVAL OUTCOMES IN HAEMATO-ONCOLOGY
Author(s)
Joshua Gahan, BSc1, Katie Law, BSc2, Harry Freeman, MSc1, Fern Woodhouse, MChem, MSc3.
1Costello Medical, London, United Kingdom, 2Costello Medical, Bristol, United Kingdom, 3Costello Medical, Cambridge, United Kingdom.
1Costello Medical, London, United Kingdom, 2Costello Medical, Bristol, United Kingdom, 3Costello Medical, Cambridge, United Kingdom.
OBJECTIVES: Validation of long-term survival outcomes by clinical experts represents a critical step for manufacturers of oncology products undergoing NICE technology appraisal (TA), especially following the publication of Technical Support Document 26 (TSD26) in 2025. The accuracy of the different expert elicitation approaches has previously been investigated in solid cancers. We reviewed TAs in haemato-oncology to assess the predictive accuracy of clinical expert estimates for long-term survival outcomes.
METHODS: NICE TAs in haemato-oncology with final guidance published between January 2021 and December 2024 were retrieved. Where available, methods and modes of expert elicitation were recorded, and clinical expert estimates were extracted for long-term survival outcomes. Long-term trial data that have been subsequently published were extracted from the literature and compared with these estimates.
RESULTS: Forty TAs were retrieved, of which 35 included expert elicitation to inform survival curve selection. Where described (n=32), opinion was elicited through validation of survival curves only (n=11), predictive estimates only (n=4), or a combination of the two (n=17); a structured expert elicitation approach was followed in four TAs. Eight TAs contained unredacted clinical expert estimates for which long-term trial data were available from the literature, covering multiple myeloma (n=3), chronic lymphocytic leukaemia (n=2), diffuse large B-cell lymphoma (n=2) and chronic myeloid leukaemia (n=1). An absolute difference of <5% between the clinical expert estimates and the long-term trial data was observed for 66 out of 89 timepoints (74%). For estimates misaligned by ≥5%, under- and over-estimation was observed for 13 and 10 timepoints, respectively.
CONCLUSIONS: Clinical expert estimates aligned well with the long-term trial data for the majority of available timepoints. There was no discernible difference in accuracy between expert elicitation approaches, although generalisability is limited due to the small sample size. These results reinforce the value of expert elicitation for reducing uncertainty in long-term survival outcomes in haemato-oncology.
METHODS: NICE TAs in haemato-oncology with final guidance published between January 2021 and December 2024 were retrieved. Where available, methods and modes of expert elicitation were recorded, and clinical expert estimates were extracted for long-term survival outcomes. Long-term trial data that have been subsequently published were extracted from the literature and compared with these estimates.
RESULTS: Forty TAs were retrieved, of which 35 included expert elicitation to inform survival curve selection. Where described (n=32), opinion was elicited through validation of survival curves only (n=11), predictive estimates only (n=4), or a combination of the two (n=17); a structured expert elicitation approach was followed in four TAs. Eight TAs contained unredacted clinical expert estimates for which long-term trial data were available from the literature, covering multiple myeloma (n=3), chronic lymphocytic leukaemia (n=2), diffuse large B-cell lymphoma (n=2) and chronic myeloid leukaemia (n=1). An absolute difference of <5% between the clinical expert estimates and the long-term trial data was observed for 66 out of 89 timepoints (74%). For estimates misaligned by ≥5%, under- and over-estimation was observed for 13 and 10 timepoints, respectively.
CONCLUSIONS: Clinical expert estimates aligned well with the long-term trial data for the majority of available timepoints. There was no discernible difference in accuracy between expert elicitation approaches, although generalisability is limited due to the small sample size. These results reinforce the value of expert elicitation for reducing uncertainty in long-term survival outcomes in haemato-oncology.
Conference/Value in Health Info
2026-11, ISPOR Europe 2026, Vienna, Austria
Value in Health, Volume 29, Issue 12S
Code
HTA24
Topic
Health Technology Assessment
Topic Subcategory
Systems & Structure
Disease
Oncology