Six models were used to evaluate the cost effectiveness of HPV vaccination of adolescent girls in two hypothetical scenarios typical of a low-income and middle-income country. Despite differences in model type, structure, assumptions and complexity, the models reached conclusions that were qualitatively similar about the cost effectiveness of vaccination, although they displayed diverse quantitative features particularly in sensitivity analyses.
The data sets given to model developers were hypothetical and not meant to inform decision making for any actual country. Furthermore, a number of simplifications were made to the exercise. For example, the exercise focused on the most important proven benefit of HPV vaccination, that is, the prevention of cervical cancer and its precursor lesions due to HPV type 16 and 18 infection. Protection against other endpoints (including genital warts, non-cervical HPV-linked cancers and crossprotection against non-vaccine type infection) were not considered. In a real-world setting, these endpoints may provide substantial additional benefits to vaccination. Also, model developers were given a month to conduct the exercise, following receipt of the standardised data set. They were asked to restrict themselves to a single model, even though some groups had several models that they could deploy, often in order to crossvalidate results. This simulates a typical real-world situation in low-income and middle-income countries, where there may be constraints on the time and resources available to conduct an economic evaluation. The limited time to conduct the exercise required other trade offs of the investigators. For example, they were asked not to alter their existing natural history parameters to better fit outcome data provided, which may account for the variation in model predictions of HPV prevalence and cervical cancer incidence, as well as the difficulty some models had in accurately reproducing both outcomes in the same model. Also, only one-way sensitivity analysis was performed. In practice, investigators may typically conduct multidirectional, probabilistic and scenario sensitivity analyses to capture uncertainty in HPV natural history and epidemiology.
Only a subset of available HPV models was evaluated. The models selected were some of those perceived to be potentially useful for deployment in low-income and middle-income settings; however, this does not imply that other available models (either from the same or from other model developers) could not also be used. Also, the type of model to be used will depend on the kind of question that needs to be answered. For example, a dynamic model would be needed to investigate the impact of herd immunity, and the incremental benefit of interventions that rely on this effect, such as catch-up campaigns. However, the input and computational power needed for such a model may not be available in all settings.
While unrealistic for directly informing policy, the results of this exercise are useful for identifying features of models and model comparisons that are useful to policymakers wishing to commission cost-effectiveness evaluations. The results of the exercise suggest that even when using a standardised data set, there are important differences between model predictions, although broad qualitative features are common to most model results. This finding highlights the importance of not relying on a single model type or set of modelling assumptions for decision making. If independent models using different structures and assumptions reach similar qualitative results (such as HPV vaccination being cost effective under a particular threshold) there may be greater confidence in the robustness of such conclusions. This is particularly the case for low-income and middle-income countries that are limited by lack of input data, and hence may have to make assumptions where data are not available.
Specific recommendations for model developers and policy makers are listed in the Appendix.