Can we satisfactorily measure the clinical value of new oncology agents with a single summary measure?

• Published in: Journal of clinical oncology Vol. 35; no. 15_suppl; p. 6606
• Main Authors: Jones, Clare Frances, Monnickendam, Giles, Zhu, Mingshu, McKendrick, Jan
• Format: Journal Article
• Language: English
• Published: 20.05.2017
• ISSN: 0732-183X; 1527-7755
• DOI: 10.1200/JCO.2017.35.15_suppl.6606

Abstract only 6606 Background: Current value frameworks (VFs) assess clinical value primarily through using clinical trial endpoints as survival metrics (e.g., median and hazard ratio (HR)). But, if key assumptions do not hold, the interpretation of these summary statistics can become problematic and fail to adequately capture the expected benefit to a patient. This has been observed with innovative oncology treatments. As a proof of concept analysis, we reviewed how two VFs (ASCO and ESMO) dealt with cases where the assumption of proportional hazards (PH) does not hold. Methods: Oncology agents approved by the FDA since 2011 were reviewed and three agents were identified with survival profiles where the assumption of PH was found not to hold because, on visual inspection, the survival curves displayed non-standard patterns: Divergence followed by convergence – panobinostat OS in RRMM; Curves initially track together then diverge – nivolumab OS in NSCLC; Curves diverge steadily then a plateau emerged in the active treatment curve – pembrolizumab PFS in refractory melanoma. We evaluated these agents to assess which measures of clinical benefit were most valued under each VF and how the issue of non-PH influenced the outcome. Results: Clinical benefit/value scores varied: ASCO: 14-27 (maximum 100), ESMO: grade 1-3. The ASCO VF uses a hierarchical approach (incorporating HR and median survival benefit, always prioritising the former) adding a bonus for survival benefit in the tail of the distribution. The combination of HR, median survival benefit 2 and 3 year survival rates in the ESMO non-curative VF can potentially capture aspects of clinical benefit in some cases of non-PH. Overall, the ASCO VF appears less flexible to accommodate non-PH than the ESMO VF. Conclusions: Despite VFs using summary statistics which cannot be easily interpreted under conditions of non-PH, the case of non-PH is not explicitly catered for. Additionally, both VFs may miss important interpretation where value is differentiated across patients groups with different response profiles which may underlie non-standard survival curves. In these situations, a more flexible approach to assessing clinical value may render VFs more relevant for clinical decision making.