OBJECTIVES

:

Economic evaluations incorporating survival data often obtain relative treatment effect estimates from network meta-analyses (NMAs) of hazard ratios (HRs) from randomised controlled trials. However, if the proportional hazards (PH) assumption underlying the HRs is violated, this could result in biased estimates that may affect the outcome of the economic evaluation. This study assessed the impact of obtaining overall survival (OS) estimates using an alternative NMA method, fractional polynomial (FP) models, on the cost-effectiveness of fulvestrant versus exemestane for previously-treated postmenopausal, oestrogen receptor-positive (ER+) advanced breast cancer.

METHODS : A cost-utility, partitioned survival model was built from the UK payer perspective. Two versions (OS estimates generated via either an FP NMA or an HR NMA) were constructed and the results compared. Sensitivity analyses were performed to investigate the comparative model uncertainty.

RESULTS

:

Both NMA approaches resulted in a high incremental cost-effectiveness ratio (ICER) for fulvestrant versus exemestane. The FP ICER was higher than the HR ICER (£146,915/quality-adjusted life year [QALY] versus £126,396/QALY, respectively) as the FP model estimated a smaller relative treatment effect (0.10 versus 0.17 QALYs gained, respectively). Both models were highly sensitive to the HR for progression-free survival and the OS HR was a key driver in the HR model. Probabilistic sensitivity analysis demonstrated a wider distribution of probabilistic ICERs and a higher probability of cost-effectiveness at low willingness to pay thresholds in the FP model, as more extreme estimates of incremental costs and QALYs were generated.

CONCLUSIONS

:

The statistical approach for NMAs of survival data may greatly affect the ICER and associated uncertainty when used in economic evaluations. NMAs using FP models may provide improved survival estimates compared with HR approaches, particularly when the PH assumption is violated, which may influence reimbursement decisions. Alternative approaches should be explored to determine the most appropriate method for the available data.