Transparency or proper study valuation procedures missed?

To the Editor:

We wish to thank the Editor for giving us the opportunity to think about and resolve a few potential issues with our paper. Garattini and colleagues have questioned the meaningfulness of the evidence used to inform some of the crucial parameter used in our model. This is because of a misalignment in the reference list, as a result of which, Table 1 in the paper points to the wrong references. We have fixed this and present the corrected version of Table 1 below.

TABLE 1

Distribution of Variables Used in the Model

Incidentally, we notice that the online appendix to the paper1 actually has all the correct references and describes in detail all the aspects of the modeling presented in the paper. We find it slightly bizarre that Garattini and colleagues have taken such a critical stance on our work, but have failed to cross-check the most technical aspects with all the available material.

Garattini and colleagues also raise a few criticisms to our general methodology. Firstly, they question the relevance of data from the region of Basilicata on the parameter representing vaccination coverage. We would like to point out that, although not covering a very large area, Basilicata was the only Italian region to implement a multicohort vaccination program, including 4 cohorts of girls aged 12, 15, 18, and 25 years. The empirical evidence derived by the Basilicata vaccination register has been published in a full paper (the GIOVE study2). In addition, this information is not used at face value, but the uncertainty underlying the estimation is fully acknowledged and propagated through the entire Bayesian model. The prior distribution of parameters that play a relevant role in the cost-effectiveness of vaccination, such as coverage rates in 4 cohorts, were drawn directly from the real-world data (information uniquely registered in the Basilicata Region) rather than from assumptions. The rates of coverage are particularly important when levels ≤50% are achieved in a single cohort of girls; and in this situation, a vaccination including a cohort of both boys and girls can improve the cost-effectiveness as a result of the increased clinical benefits determined by herd immunity. Actually, we must be wondering whether the most economic and clinically effective decision is provided by the immunization of both sexes or by an increase in the coverage rate in a single cohort of females. Probably, the latter might be more complicated and less effective than expected. Increasing the coverage rate may require complex interventions, a long period of time, and a significant incremental cost that could determine a diseconomy of scale. Truly, a scarce result when compared with the huge investment that is needed to increase the baseline rate value by 1 percentage point. Our study reported some indirect and preliminary indication; however, a specific Bayesian dynamic model addressing the cost-effectiveness of a vaccination program that includes a cohort of boys and girls has already been designed and results will be assessed and published shortly.

As for patients’ health-state preferences, we agree with Garattini and colleagues that they represent a highly sensitive variable for the economic evaluations. In this case, we developed an algorithm for the fully computerized administration of a Time Trade-Off questionnaire; this was validated and published in 2011.3 In that publication, the standardized elicitation of utilities was focused on cervical intraepithelial neoplasia (grades 2 and 3), anogenital warts, and cervical cancer exclusively.3 Thus, to include a broader range of human papillomavirus (HPV)-induced pathologies (which were indeed considered in the model developed in the BEST study) and a larger sample size, we used data from an ongoing study that involved >450 patients. Preliminary results from this large study have been communicated or presented in several congresses (including HTAi4) and the overall evaluation will be published as soon as it is completed. We believe that it is noteworthy that the elicitation of each utility used to inform our model relied on a solid and well-acknowledged procedure.3,5,6 Similarly to the point we have made earlier, by using a fully Bayesian model, we incorporated the uncertainty in the estimated values of utilities.

Another issue is about the vaccine price. We modeled this parameter using a probability distribution eliciting the information about the mean unit price of €69.13 and encoding the assumption that 95% of the most plausible values were included in the interval between €60.16 and €79.58. This was based on Regional tenders that occurred in 2008 and 2009 in Italy. Although in a commentary published in early 2012,7 neither an accurate mean price nor a SD were specifically reported for HPV vaccines, a mean price per quadrivalent vial seems to be very close to the range of values we used to inform our model. Although an effective public health intervention is not exclusively a matter of price,8 any value below the lower limit of the range adopted in our study would have had a favorable effect on the cost-effectiveness of the vaccination strategy that we evaluated using a Bayesian framework.

Finally, Garattini and colleagues wonder about the reliability of the results of our model. We are seeking to produce a structured research program, building on the findings of the GIOVE study, which was related to the effectiveness of a multicohort quadrivalent-based vaccination program. Consequently, the BEST study was specifically designed to assess the cost-effectiveness of this predefined vaccination strategy. Although a potential direct comparison evaluating the most cost-effective option between the 2 available vaccines might be interesting, this is an objective that was not consistent with the aim of the BEST study.

Although some biological characteristics of HPV are uncertain, the value of information derived from current clinical trials is improved and the accuracy is increased by the incorporation of prior information in a Bayesian modeling. Further, prior distribution of parameters significantly influencing the impact of vaccination (ie, coverage rates and risk factors having an effect on the dynamic transmission of HPV infection) were directly drawn from the health programs already implemented in Italy and not from assumptions. Although financing and sustaining immunization programs are health governance challenges that public health authorities have to deal with, an assessment of a multicohort or both sexes vaccination strategy with a Bayesian model can inform decision-makers with more reliable data about both the cost-effectiveness of interventions as well as its budgetary implications. In conclusion, Bayesian analytic models have a wide range of uses and can be deemed as important and powerful tool for economic evaluations in health care.9

Especially when associated with the expected value of information, Bayesian models can provide with an accurate valuation of any future implementation of a quadrivalent-based HPV vaccination program.