• Open Access

Uncertainty and variability in influenza cost-effectiveness models

Authors


Correspondence to: Dr Anthony Newall, School of Public Health and Community Medicine, University of New South Wales, Sydney, NSW 2052; e-mail: a.newall@unsw.edu.au

Correspondence to: Dr Vittal Mogasale, Centre for Burden of Disease and Cost-effectiveness, The University of Queensland, School of Population Health, Herston, Qld 4006; e-mail: vmogasale@gmail.com

Mogasale and Barendregt performed a cost-effectiveness analysis of influenza vaccination for Australians aged 50–64 years in Australia.1 A major part of their article was devoted to contrasting their findings with those we had previously published.2 They concluded that several of the central parameters cannot be estimated with a sufficient degree of certainty to reliably study the cost-effectiveness of influenza vaccination in this group. While we agree that there are several areas of uncertainty that warrant further exploration, we do not feel that the authors have adequately identified what the key parameters are and what should be the focus of future research. We refer the authors to our recent review of cost-effectiveness analyses in older adults where we discuss many of the important issues.3

Mogasale and Barendregt focus their analysis on the incidence of influenza-like illness (ILI). However, this parameter is influential in their model not because of the costs and consequences they assume for ILI cases, but rather due to their modelling approach whereby any variation in ILI leads to a proportional change in all other disease outcomes. The use of ILI as the primary basis for the sensitivity analysis and the sequential approach used, may have obstructed the authors from identifying the disease parameters that have a larger direct impact on results. In their analysis, disease estimates other than ILI cases, such as influenza hospitalisations (which accounted for substantially more of the cost savings) and deaths (which accounted for the ‘Disability Adjusted Life Years’) are likely to have had a greater direct impact on the cost-effectiveness ratio.

Estimating the number of influenza hospitalisations and deaths (which may be prevented by vaccination) is complex as only a fraction of the influenza burden is recorded as such. However, over many decades various statistical methods have been developed to estimate hospitalisations and deaths attributable to influenza.4–7 These methods are now routinely applied by researchers around the globe including the US Centers for Disease Control and Prevention7 and US National Institute of Health.5,6 Mogasale and Barendregt state that they apply probabilities of severe complications from “insurer data and hospital records”.1 While the exact method of calculating these probabilities from the references provided8,9 was not transparent to us, it is clear that they do not refer to the available Australian data. In our cost-effectiveness model,2 we applied statistical estimates of the average rate of hospitalisations and deaths attributable to influenza in Australia over multiple influenza seasons.10

It is also important to recognise the role of inter-year variability as it relates to estimating influenza parameters, a factor not well-described by the authors. The influenza disease burden (cases, hospitalisations and deaths) varies from year to year with the circulating strains. Similarly, vaccine efficacy may vary from year to year dependent on the match to the circulating strains. This form of inter-year variability is unavoidable and may partially explain differences in parameter estimates from alternative sources. It is not reasonable to use this type of variability, as opposed to other forms of uncertainty, to draw conclusions that influenza vaccination programs should not be funded.

We agree with the authors that further research should be funded to help inform uncertain key parameters in influenza models. However, the parameters on which to initially focus future research efforts should be those that have the greatest direct impact on cost-effectiveness ratios. This should include parameters associated with severe influenza disease and vaccine efficacy as described in our recent review;3 however, it may also include the burden of less severe disease, where productivity losses and other direct effects from milder disease are given weight by decision makers.

Acknowledgements

ATN holds an NHMRC Australian-based Public Health Training Fellowship (630724).

Appendix

Response from the authors:

Jan Barendregt 3 and Vittal Mogasale 33 Centre for Burden of Disease and Cost-effectiveness, The University of Queensland

We used the Australian ILI incidence data in our base case scenario, despite the unlikely low incidence when compared internationally. This was confirmed by the unlikely low hospitalisations and deaths this incidence produced. The strong suspicion of biased ILI incidence data is an interesting finding by itself.

When we hiked-up the ILI incidence to a more internationally comparable level, we ended up with hospitalisations and deaths very comparable to Australian data estimated in Newall et al.1 Nevertheless, there remained a large difference in terms of ICER with Newall et al., which turned out to be due to vaccine uptake, efficacy, and costs data (Table 4 in our paper).2

We think therefore that, unlike Newall et al. are claiming, we have clearly identified the variables that are in need of better measurement because of their large influence on the ICER.

We agree with the comment that there will be inter-year variability in influenza disease burden and vaccine efficacy. But we never claimed that inter-year variability would be a reason not to fund an influenza vaccination program.

Ancillary