An Approach to Reconciling Competing Ethical Principles in Aggregating Heterogeneous Health Preferences
Total Page:16
File Type:pdf, Size:1020Kb
ORIGINAL ARTICLE An Approach to Reconciling Competing Ethical Principles in Aggregating Heterogeneous Health Preferences Barry Dewitt, MSc, Alexander Davis, PhD, Baruch Fischhoff, PhD, Janel Hanmer, MD, PhD Background. Health-related quality of life (HRQL) scores that rely on HRQL scores, such as cost-effectiveness analy- are used extensively to quantify the effectiveness of medical ses. We propose an analytical-deliberative framework for interventions. Societal preference-based HRQL scores aim to choosing one (or a set of) aggregation procedure(s) in a produce societal valuations of health by aggregating valua- socially credible way, which we believe to be analytically tions from individuals in the general population, where sound and empirically tractable, but leave open the institu- each aggregation procedure embodies different ethical prin- tional mechanism needed to implement it. Conclusions. ciples, as explained in social choice theory. Methods. Using Socially acceptable decisions about aggregating heteroge- the Health Utilities Index as an exemplar, we evaluate soci- neous preferences require eliciting stakeholders’ preferences etal preference-based HRQL measures in the social choice among the set of analytically sound procedures, represent- theory framework. Results. We find that current preference ing different ethical principles. We describe a framework for aggregation procedures are typically justified in terms of eliciting such preferences for the creation of HRQL scores, social choice theory. However, by convention, they use only informed by social choice theory and behavioral decision one of many possible aggregation procedures (the mean). research. Key words: health state preferences; health- Central to the choice of aggregation procedure is how to related quality of life; health utility; equity; cost-effectiveness treat preference heterogeneity, which can affect analyses analysis. (Med Decis Making XXXX;XX:xx–xx) eneric health-related quality of life (HRQL) mea- be asked to assess, in numeric terms, the relative Gsures place individuals’ health status on a com- quality of life with blindness and 20/20 vision. mon scale, allowing researchers to compare the effects Although analysts need such aggregation proce- of clinical trials across individuals and summarize dures so that they can incorporate HRQL estimates the results of population health studies.1 Utility-based into their models, aggregating individuals’ estimates measures are a subset of HRQL measures, and attach to societal ones poses a fundamental problem in scores to states of health. These scores can be used for social choice theory.4–8 It has long been known that, outcomes, e.g., quality-adjusted life years, providing under certain general assumptions, there is no unique the estimates needed by regulatory analysts.2,3 solution to this aggregation problem if there is any To produce societal preference-based HRQL heterogeneity in individuals’ preferences. Rather, scores, the common practice is to aggregate the pre- many solutions are possible. Selecting a method to ferences of a sample of individuals designated as collapse a distribution of preferences into a single representing the target population. That might number implies an ethical judgment about what dis- mean eliciting the preferences of patients with a tributional information matters. disease, experts in a disease, or individuals held to Foundational work in utility-based HRQL mea- represent society as a whole. For example, a sample surement9 recognized the preference aggregation of individuals with naturally varying vision might problem. Concurrent research in social choice the- ory revealed the implications of alternative aggrega- tion procedures.10 Here, we integrate the two fields, taking advantage of advancements in both during Ó The Author(s) 2017 the ensuing years. We use the Health Utilities Index Reprints and permission: (HUI) as an exemplar,11–13 examining it in terms of http://www.sagepub.com/journalsPermissions.nav 5,8,10 DOI: 10.1177/0272989X17696999 key concepts in social choice theory, thereby MEDICAL DECISION MAKING/MON–MON XXXX 1 DEWITT AND OTHERS providing a concrete example that could be fol- begins with a set of axioms that a preference aggrega- lowed with other societal preference-based scores. tion procedure must satisfy to be deemed rational. One We use ‘‘HUI’’ to refer to both the HUI Mark 2 and commonly used set has these three axioms: Mark 3 systems. When it is necessary to distinguish 1) Unrestricted Domain: Any set of individual pre- between them, we do so with the acronyms ‘‘HUI:2’’ ferences is allowed. and ‘‘HUI:3.’’ 2) Independence of Irrelevant Alternatives: If two We first identify the conditions under which HUI groups of utility functions agree on a subset of is normatively justified, and then show the range of health states, then the societal preferences of the acceptable aggregation procedures, each of which two groups agree on that same subset. expresses an ethical stance. Finally, we offer an 3) Weak Pareto Criterion: If all individuals prefer approach for choosing among these options. Related health state x to health state y, then the societal concerns can be found in assessments of other forms preference should as well. of analysis.14–17 Our approach is generalizable to any societal preference-based HRQL measurement system. (These axioms are applied only to preference aggregation procedures defining societal prefer- METHODS ences that produce a complete, reflexive, and transi- tive ordering over the state space.) A fundamental Below, we describe relevant results from social result from social choice theory is Arrow’s Impossibil- choice theory, applicable to specifying any socie- ity Theorem,5 which states that the only guaranteed tal preference-based HRQL measurement system. way to aggregate individual preferences that satisfies Additional background material on societal prefer- these axioms is dictatorship: impose the preferences ence-based HRQL scores and the HUI system is of one individual on the entire group. It is called an available in the online Appendix. (Also available at impossibility theorem because ‘‘non-dictatorship’’ is https://osf.io/2xz2a.) an axiom in Arrow’s framework, making it impossible to satisfy the full set of axioms. Social Choice Theory Arrow’s result has been interpreted as precluding any non-dictatorial aggregation procedure from being Social choice theory characterizes preference aggre- normatively justified. However, as Sen8 initially gation procedures that define societal preferences. It showed, and Roberts10 elaborated, Arrow’s theorem is a special case of a more general result about prefer- Received 26 June 2016 from Department of Engineering and Public ence aggregation. Sen and Roberts identified two Policy, Carnegie Mellon University, Pittsburgh, PA, USA (BD, AD); aspects of individuals’ preferences that determine the Department of Engineering and Public Policy and the Institute for Poli- tics and Strategy, Carnegie Mellon University, Pittsburgh, PA, USA type of aggregation that is possible: informational con- (BF); Division of General Internal Medicine, University of Pittsburgh, tent and interpersonal comparability;together,they Pittsburgh, PA, USA (JH). This research was done in the Department constitute the informational basis of the preferences. of Engineering and Public Policy, Carnegie Mellon University, and the The informational content of an individual utility Division of General Internal Medicine, University of Pittsburgh. The function reflects its measurement scale; for example, work was presented at the 2015 North American meeting of the Soci- ordinal or interval. Ordinal preferences provide only ety for Medical Decision Making. Financial support was provided by the Swedish Foundation for the Humanities and the Social Sciences enough information to rank options. For example, (Riksbankens Jubileumsfond) Program on Science and Proven Experi- assigning 15 to option A, 5 to option B, and 0 to option ence. Janel Hanmer was supported by the National Institutes of C means that A is preferred to B, B is preferred to C, Health through Grant Number KL2 TR000 1856, and Barry Dewitt and A is preferred to C, but nothing more. With cardi- received partial support from the Social Sciences and Humanities nal preferences, utility is on an interval scale, so that Research Council of Canada through a doctoral fellowship. The fund- units of utility have consistent meaning across the ing agreements ensured the authors’ independence in designing the study, writing, and publishing the report. Revision accepted for publi- scale. Thus, in the example, A is preferred to B by cation 26 January 2017. twice as much as B is preferred to C, meaning that there is intrapersonal comparability of utility differ- Supplementary material for this article is available on the Medical Deci- ences. Ordinal and cardinal preferences, on ordinal sion Making Web site at http://journals.sagepub.com/home/mdm. and interval scales, respectively, define what can be Address correspondence to Barry Dewitt, MSc, Department of Engi- said about an individual’s preferences but say nothing neering & Public Policy, Carnegie Mellon University, 5000 Forbes Ave, about interpersonal comparisons among individuals’ Pittsburgh, PA 15213, USA; e-mail: [email protected]. preferences. 2 MEDICAL DECISION MAKING/MON–MON XXXX ETHICS, PREFERENCE AGGREGATION, AND HEALTH UTILITY Following Sen and Roberts, we distinguish two elicited on a scale from 0 to 1, where 0 represents