INTRODUCTION:

State-transition models are based on the assumption of mutually exclusive health states. Transition probability estimates used in economic models, particularly when obtained from different sources, may however not reflect that feature: event risk often increases with age (e.g. cardiovascular (CV) events) and may even add up to more than 1 when independent risk inputs are used, leading to biased and illogical results. Risk adjustment (RA) methods and their impact on cost-effectiveness (CE) are largely neglected in the literature and economic modelling practice.

OBJECTIVES:

To identify RA techniques and evaluate their effect on predicted event rates based on an example economic model.

METHODS:

Based on basic probability principles, three main categories of potential RA were identified:

1. Arbitrary reductions: decreasing risks until the logical constrains are satisfied
2. Sequencing events: evaluating events in an assumed sequence
3. Creating combined health states: adding states which reflect multiple events occurrence

The effects of these RA were evaluated using a semi-Markov model based on Wilson 2012 CV risk equations and non-CV mortality estimates from life tables. The model includes 3 health states, where patients are at risk of non-fatal CV events, fatal CV events and fatal non-CV events. Additionally, the effect of altering cycle length was assessed.

RESULTS:

The differences in predicted CV rates between RA methods and the unadjusted rates were between - 4% to +2% for the base case inputs. The impact of the RA methods increased with longer cycle length.

CONCLUSIONS:

A number of RA can be implemented and the decision on which one to use, if any, will depend on the inputs, model and resource availability in each particular case. Shortening cycle length reduces the impact of RA. Ignoring to implement RA might substantially affect rate predictions, leading to biases in CE results and ultimately erroneous HTA reimbursement decisions.