When does economic model type become a decisive factor in health technology appraisals? Learning from the expanding treatment options for relapsing-remitting multiple sclerosis

Objectives: Specific economic model types often become de facto standard for health technology appraisal over time. Markov and discrete event simulation (DES) models were compared to investigate the impact of innovative modeling on the cost-effectiveness of disease-modifying therapies (DMTs) in relapsing-remitting multiple sclerosis (RRMS). Fingolimod was compared to dimethyl fumarate (DMF; in highly active [HA] RRMS), alemtuzumab (in HA RRMS) and natalizumab (in rapidly evolving severe RRMS). Comparator DMTs were chosen to reflect different dosing regimens. Materials and methods: Markov and DES models used have been published previously. Inputs were aligned in all relevant respects, with differences in the modeling of event-triggered attributes, such as relapse-related retreatment, which is inherently difficult with a memoryless Markov approach. Outcomes were compared, with and without different attributes. Results: All results used list prices. For fingolimod and DMF, incremental cost-effectiveness ratios (ICERs) were comparable (Markov: 4206 pound/quality-adjusted life year [QALY] gained versus DES: 3910 pound/QALY gained). Deviations were observed when long-term adverse events (AEs) were incorporated in the DES (Markov: 25,412 pound saved/QALY lost, versus DES: 34,209 pound saved/QALY lost, fingolimod versus natalizumab; higher ICERs indicate greater cost-effectiveness). For fingolimod versus alemtuzumab, when relapse-triggered retreatment was included in the DES, large cost differences were observed (difference between incremental cost is 35,410 pound and QALY is 0.10). Limitations: UK payer perspective, therefore societal approach was not considered. Resource utilization and utilities for both models were not derived from the subpopulations; as the focus is on model type, input limitations that apply to both models are less relevant. Conclusions: Whilst no model can fully represent a disease, a DES allows an opportunity to include features excluded in a Markov structure. A DES may be more suitable for modeling in RRMS for health technology assessment purposes given the complexity of some DMTs. This analysis highlights the capabilities of different model structures to model event-triggered attributes.