Intro
Modern epidemiological studies generally report confidence or uncertainty intervals around their estimates, often based on the variation observed in sample data. Estimates of the burden of disease and of risk factors, which extrapolate from specific data sources and epidemiological studies to population-level measures, are subject to a broader range of uncertainty because of the combination of multiple data sources and value choices. Hence, the reported uncertainty intervals should ideally include all sources of uncertainty, including those arising from measurement error, systematic biases, and modeling and extrapolation to compensate for incomplete data. In contrast to uncertainty analysis, which attempts to formally quantify the limitations of available data, sensitivity analysis examines how key analytic outputs vary when input quantities are systematically varied. Following Murray and Lopez (1996b), this chapter uses sensitivity analysis to examine the specific effects of social values that have been incorporated in the design of the disability-adjusted life year (DALY).
Taking account of uncertainty in such value parameters as the rate of time preference used to discount future outcomes is not common. Even if there is empirical evidence on population preferences for discount rates and uncertainty in these estimates, investigators have argued that the choice of discount rate for use in analysis is essentially a social value judgment and should not include uncertainty (Morgan and Henrion 1990). Although there is uncertainty about the social value judgment and about its effects on decisions based on the analysis, varying the value deterministically in the analysis and performing a sensitivity analysis to examine the impact on the outcomes of interest is usually preferable to uncertainty analysis. Thus, the 1990 Global Burden of Disease (GBD) study (Murray and Lopez (1996b)) examined the sensitivity of the ranking of causes of the burden of disease globally when discount rates and age weights were varied across a range of possible values.
Health state valuations, which link mortality information with information on nonfatal health outcomes in summary measures of population health, fit somewhat more ambiguously within the framework of uncertainty analysis. If we conceptualize a health state in terms of levels in multiple domains of health, health state valuation involves the weighting of these domains to arrive at an overall assessment of the health level associated with the state. These valuations, unlike values such as time preference, do not have any clear normative basis; that is, while we might rely on philosophical arguments about intergenerational equity in choosing a discount rate, no obvious arguments pertain to the relative importance of mobility versus cognition in overall assessments of health levels. The choice of measurement strategies for eliciting health state valuations does sometimes introduce normative questions, but these pertain to additional considerations, such as concern for fair distribution, which are orthogonal to the assessment of the health state itself.
