ASSESSMENT OF PROPORTIONAL HAZARDS IN NICE SUBMISSIONS: A REVIEW OF RECENT NICE APPRAISALS IN CANCER

OBJECTIVES: Conducting a network meta-analysis (NMA) of hazard ratios (HRs) for time-to-event outcomes, when the proportional hazards (PH) assumption is violated, could lead to biased treatment effect estimates. Published algorithms exist to reconstruct individual patient data from digitised plots making formal testing of PHs possible. In the UK, NICE has stressed the importance of using an appropriate scale, thus Health Technology Assessment (HTA) groups may expect justification for the choice of scale in reimbursement submissions given the accessibility of formal diagnostic testing techniques. Taking the UK as a case study we aimed to identify whether this was the case in recent NICE appraisals. METHODS: We searched the NICE website for Technology Appraisal (TA) guidance documents in cancer published in the last three years. Searches were run in December 2016. We reviewed the TAs to identify whether manufacturers had assessed the PH assumption and justified the choice of scale when synthesising time-to-event endpoints for a NMA. HTA groups' comments on these assessments were also identified. RESULTS: A total of 45 TA submissions in oncology were identified. 27 included an NMA of survival endpoints, of which only 6 manufacturers assessed PHs using established methods on trials included in the NMA. In an additional 10 submissions the HTA group noted that PHs does not hold in one or more of included trials and should have been considered in determining the analysis methods. Methods used to address non-proportional hazards included; independent modelling of treatment arms, fractional polynomial models, and piecewise modelling. CONCLUSIONS: Choice of scale is important to avoid bias in an NMA. HTA groups have criticised submissions that have not justified or formally tested the PHs assumption for NMAs synthesising HRs. Manufacturers should justify choice of scale used in NMAs for reimbursement submissions and conduct formal assessment of PHs when synthesising HRs for time-to-event outcomes.