Different Methods for Modelling Severe Hypoglycaemic Events: Implications for Effectiveness, Costs and Health Utilities

Abstract

Background

Clinical trials report severe hypoglycaemic events as the number of patients with at least one event out of the total randomised or number of events for a given total exposure. Different network meta-analysis models have been used to analyse these different data types.

Objective

This aim of this article was to establish the impact of using the different models on effectiveness, costs and health utility estimates.

Methods

We analysed a dataset used in a recent network meta-analysis of severe hypoglycaemic events conducted to inform National Institute for Health and Care Excellence recommendations regarding basal insulin choice for patients with type 1 diabetes mellitus. We fitted a model with a binomial likelihood reporting odds ratios (using a logit link) or hazard ratios (complementary log-log link), a model with a Poisson likelihood reporting hazard ratios and a shared-parameter model combining different types of data. We compared the results in terms of relative effects and resulting cost and disutility estimates.

Results

Relative treatment effects are similar regardless of which model or scale is used. Differences were seen in the probability of having an event on the baseline treatment with the logit model giving a baseline probability of 0.07, the complementary log-log 0.17 and the Poisson 0.29. These translate into differences of up to £110 in the yearly cost of a hypoglycaemic event and 0.004 in disutility.

Conclusion

While choice of network meta-analysis model does not have a meaningful impact on relative effects for this outcome, care should be taken to ensure that the baseline probabilities used in an economic model are accurate to avoid misrepresenting costs and effects.

http://ift.tt/2ErzOeL