Health utility and economic analysis: theoretical and practical issues

Economic evaluation has become increasingly important in health policy. It influences a number of agencies around the world, not only regarding which benefits should be valued, but also who should value the benefit. The potential impact of economic evaluation of healthcare is enormous. In 2008, total US health expenditures were expected to rise by 6.9%, or two-times the rate of inflation, to a level representing 17% of gross domestic product [1]. When economic evaluation was an emerging discipline with little real-world impact, methodological issues were primarily of intellectual importance. This is no longer the case. With the methodology influencing billions of dollars in expenditures across the world and impacting the lives of hundreds of millions of people, even relatively minor issues are of tremendous importance. Our own work has led us to consider several issues we would like to address here that should be taken into consideration when performing economic analysis in healthcare.

Overview of health utility

Health utility is a number that describes the health state or outcome of a patient, ranging from zero (representing death) to one (representing full health). Since the range of values permits a consistent analysis of the value of health states for cost–effectiveness purposes, health utility values are widely used in conducting economic analyses to determine how to better allocate healthcare resources and assist policy- and decision-making processes; for example, which medication should be included into drug lists, treatment A versus treatment B. Importantly, health utility allows morbidity and mortality improvements to be combined into a single weighted measure or quality-adjusted life-years (QALYs) gained [2]. QALYs represent the number of healthy years of life that are valued equivalently to the actual health outcome. In addition, QALYs allow broad comparisons across widely differing programs. QALYs are generally calculated by multiplying the time a person stays in each health state by the health utility of that health state, and summing up the obtained products for all states [3].

Estimates of health utility can be obtained by either direct or indirect methods. Standard gamble and time trade-off methods are currently preferred by health economists, and have been widely used owing to their sound theoretical basis in economic utility theory [2,4,5]. These methods are performed by asking respondents to value various health-state scenarios by explicitly considering how much they would be willing to sacrifice to avoid being in a particular health state. The utility values derived from these direct methods are based on a single utility model, but the impact of illness on health-related quality of life (HRQoL) is multifaceted. Furthermore, direct methods are not practical for use in clinical studies. Given that both methods are related to probabilities and willingness to trade life times, it is difficult for many individuals to estimate the numbers that correspond to specified health states. Indirect methods elicit health utility using HRQoL instruments such as the Quality of Well-Being Scale, Health Utilities Index, 6-Dimension Short-Form (SF-6D) and 3-Level EQ-5D. Data collection using these instruments is more popular than direct methods because they are simple and reproducible. These instruments employ a multi-attribute utility theory or a multi-attribute health status classification system based on a ‘descriptive system’, which is a set of ‘items’ that describe different levels of some aspect of health (usually referred to as a health ‘dimension’ or ‘domain’, such as mobility or anxiety). Corresponding to the descriptive system are community-based preferences that are compatible with the underlying theory of cost-effective analysis and QALYs. This makes them more practical and theoretically appealing [6].

However, if these indirect methods are not available, predicting health utility using mapping techniques is an alternative approach suggested by NICE [7]. Mapping techniques are statistically estimated exchange rates between two HRQoL instruments, either two generic HRQoL instruments or generic and disease-specific HRQoL instruments. Mapping permits conversion of different ‘yard sticks’ into the same measure. Several econometric methods used for mapping purposes include ordinary least squares, Tobit, multinomial logit regression using response mapping techniques, median regression and censored least absolute deviation. Each of these has been used to determine the relationships between two generic measures of HRQoL [4,8–10] or the relationship between a disease-specific and a generic HRQoL measurement [11–14].

Issues with theoretical concepts of HRQoL & utility measures

Generic HRQoL instruments developed for measuring health status are based on the WHO definition, which describes health as “a state of complete physical, mental, and social well-being and not merely the absence of disease or infirmity” [101]. Following this definition, generic HRQoL instruments measure across four main dimensions of health: physical, psychological, role and social functioning, and general perception of health. Generic measures of HRQoL offer the advantage of providing a common estimate of the health across different populations. The major disadvantage is the requisite limitation of the measures, by their focus on the most common denominators of health; aspects of health shared by everyone. As a result, they are less sensitive to the progression or treatment of a specific disease. Furthermore, traditional HRQoL concepts may not cover other aspects of health that are not derived from purely physical or psychological functionings.

For example, vision functioning is believed not to be a part of traditional conceptual models of health status measures. There is some evidence from recent publications showing that a number of patients suffering from age-related macular degeneration symptoms nonetheless reported full health using standard measures [15,16]. Furthermore, a study from McIntosh and colleagues showed that vision functioning is separated from other traditional dimensions of health [17]. Payakachat and colleagues attempted to estimate the relationship between EQ-5D utility values and the 25-item National Eye Institute Vision Functioning Questionnaire (NEI-VFQ 25) using different mapping techniques. The results showed that only a small portion of the two measures overlapped, suggesting that the EQ-5D measured different health constructs when compared with the NEI-VFQ 25, and does not cover all facets of vision functioning [16]. Yet, who would consider the loss of vision in their own life to be anything less than a significant and substantial impairment of their quality of life?

At times, current approaches to HRQoL measurement using generic instruments seem to minimize the negative impacts of certain disease states. However, there is also evidence for overestimation of the impact of disease. A problem arises because health utilities commonly used in economic analysis are generated from a sample of the general population [4,5,18], yet health state values for patients are significantly higher than for the general population [19]. In 2004, Brazier et al. proposed three causes for this discrepancy [19]:

• • Patients and the general population report different values for the same states

• • Patients use different measurement sticks

• • Patients experience shifts in values that are not anticipated by the general population

For these reasons, using patient-versus-general population values remains an unresolved argument [4,20–22]. These factors can create circumstances wherein health policy formulators must use results from cost–effectiveness analysis with caution because they may not accurately reflect the real benefits that patients received from treatments or interventions.

Practical issues of health utility

Statistical issues also merit careful consideration. Ceiling and floor effects can occur in reported health utilities generated from indirect methods. A ceiling effect refers to a great portion of patients reporting an extremely high score at one end of the health utility scale. A floor effect, by contrast, represents a great portion of patients reporting an extremely low score at the other end of the health utility score. A ceiling effect is one of the most problematic characteristics of both the 3-Level EQ-5D utility index and Health Utilities Index, while a floor effect is a problem found in SF-6D [23–26]. This phenomenon implies that some items or dimensions do not have adequate capacity to discriminate higher or lower levels of health. The ceiling effect shows that the EQ-5D and Health Utilities Index do not have sufficient power to discriminate between health statuses along the continuum of health utility, especially at the high end (health utility index = 1). This problem was recognized by EQ-5D developers, and consequently, a new 5-Level EQ-5D has been developed to overcome this problem [27]. However, at least a few years remain before a new validated algorithm can be launched. A large ceiling effect suggests it could be less responsive to improvements in conditions associated with low morbidity. Furthermore, even below the ceiling, any health improvement would need to be large for a change to be registered by the HRQoL categories [23]. Since it reflects that only a small fraction can be increased in cost–effectiveness analysis, it will be difficult to show the benefit of treatments or interventions in low morbidity conditions.

Second, health utilities generated from different generic HRQoL instruments, or even from the the same instrument but using different populations, are not the same. The mean SF-6D utility value was found to exceed the EQ-5D by 0.045, and they were not highly correlated [26]. Population mean EQ-5D utility values elicited from a UK population were found to be significantly lower than the US population, suggesting that using the health utility set generated from the UK population would lead to bigger gains in QALYs [28,29].

Last, health utilities elicited from any indirect methods are merely point estimates. Therefore, using estimated variances from each algorithm in sensitivity analysis is highly recommended in economic analysis. Special caution is merited when using mapping methods to produce health utility used for cost–effectiveness. It will be prone to error and biases since estimated health utilities from mapping methods exhibit less variance than health utilities estimated directly from the HRQoL instruments (e.g., producing health utility directly from the EQ-5D). This issue can cause false reassurance in the accuracy of cost–effectiveness estimates.

Conclusion

There are many issues that healthcare researchers should consider when applying health utilities in economic evaluations. A small increment at the end of the health utility scale may not be seen when the distribution of health utility is skewed, implying gaps in the scale. Evaluating the cost–effectiveness of some treatments or interventions in relatively low morbidities, such as treatments in patients with low vision, may obscure benefits compared with relatively high morbidities. Health utilities are specific to whichever generic HRQoL instrument is used to generate them, as well as the population from which they originated. Different assessment methods lead people to construct different preferences. It may not be valid to consider one method of preference assessment as a standard for another. Sensitivity analysis using estimated variances provided from each algorithm is also highly recommended when performing cost–effectiveness analyses. It is worth exploring how we can create an interval scaling for health utility in specific diseases or conditions, such as age-related macular degeneration, and where it should be relative to health utility values generated from generic HRQoL instruments. Ultimately, health utilities elicited from indirect methods are valuable tools that can be used to compare health statuses across different disease conditions, as long as their limitations are recognized.

The application of economic evaluation is expected to reduce the amount of bias introduced into healthcare policy formulation by quantifying the human condition. Human pain and suffering are a source of intense emotional and political turmoil, which may obscure the best possible solution available. Unfortunately, in the process of attempting to reduce such bias, it is possible for healthcare researchers to introduce additional bias owing to the inherent limitations of current methodology. We must, therefore, always guard against the prospect of a logically satisfying but intellectually lazy assumption that human behavior will tend to conform to our own treasured models.

Financial & competing interests disclosure

The authors have no relevant affiliations or financial involvement with any organization or entity with a financial interest in or financial conflict with the subject matter or materials discussed in the manuscript. This includes employment, consultancies, honoraria, stock ownership or options, expert testimony, grants or patents received or pending, or royalties.

No writing assistance was utilized in the production of this manuscript.