PRM221 - NETWORK META-ANALYSIS IN THE PRESENCE OF NON-PROPORTIONALITY: A REVIEW OF NICE SUBMISSIONS

Abstract

Survival network meta-analyses (NMAs) are a key component of many health technology assessment submissions in oncology when head-to-head evidence is not available for all comparators. Traditional NMA methods of survival outcomes using hazard ratios (HRs) assume proportional effects of treatment over time. More recent fractional polynomial (FP) techniques do not rely on this proportional hazards (PH) assumption. Currently, no formal Decision Support Unit guidance exists on performing survival NMA. The objectives of this work were to examine PH and NMA methods in National Institute for Health and Care Excellence (NICE) submissions and their associated acceptability, and to explore the comparability of results from FP and HR NMAs in the presence of non-PH. We comprehensively identified and extracted information relating to NICE appraisal consultations for oncology products since June 2016. Information extracted included testing of PH, NMA methods, and Evidence Review Group (ERG) and committee comments. A targeted literature review of studies comparing FP and HR NMAs also was conducted. Since June 2016, 60% (47/78) of oncology submissions included survival NMAs or indirect comparisons. Of those, 47% utilised only HR-based NMAs, one (2%) used only an FP-based NMA, 13% used FP and HR, 38% used other approaches. The remaining submissions included limited evidence on comparable populations or availability of head-to-head evidence for all comparators. Tests of PH were reported in 81%. Of those reporting non-PH, approximately half only performed HR based approaches. Where there was evidence of non-PH, and the initial submission reported only HR NMAs, the ERG requested FP analysis. When non-PH was identified, FP and HR analysis sometimes provided conflicting results. It is important for submissions considering time-to-event endpoints to assess and report whether non-proportionality exists using visual and formal tests. Where there is evidence of non-proportionality, both HR and FP techniques are recommended.