OBJECTIVES

In 2017, the National Institute for Health and Care Excellence (NICE) appraised naltrexone-bupropion (NB32) + standard management (SM) versus SM alone for the treatment of adult obesity (TA494) based on a patient-level model implemented as a discretely integrated condition event (DICE) simulation.

The Evidence Review Group (ERG) had concerns with how the implementation of the model affected its run time, limiting the ERG’s ability to simulate sufficient patients and probabilistic sensitivity analysis (PSA) iterations. This may have contributed to the appraisal committee rejecting NB32 as an appropriate use of resources. This study aims at examining the impact of modelling the same decision problem using a non-DICE, cohort-level approach.

METHODS

Conceptual modelling and literature reviews were conducted to develop a cohort model structure that in its assumptions otherwise matched the NICE-preferred approach based on available materials at time of replication. A non-DICE, 40-state, probabilistic, cohort-level Markov model was specified and constructed in Excel^®.

RESULTS

The mean PSA incremental cost-effectiveness ratio (ICER) from the cohort-level model (£47,729) was higher than both the committee’s preferred ICER (£23,750), and the £30,000 upper limit of the NICE willingness to pay threshold range. For both models, results were very sensitive to changes in the small estimated incremental health benefit (0.0195 versus 0.0434 quality-adjusted life-years). Intrinsic differences between both modelling approaches contributed to differences in results. Data availability issues limited replication efforts. Model execution in the cohort-level model was near-instant.

CONCLUSIONS

Were a cohort-based, non-DICE modelling approach submitted by the manufacturer in TA494, this research suggests the NICE recommendation would have likely been the same. However, the model execution issues faced by the ERG and affecting committee deliberations would have been avoided. Given the number of health states required to approximate the analysis in a cohort-level model, a non-DICE patient-level approach directly capturing patient heterogeneity may have been more apt.