A further examination of the problem of double-counting in incremental cost-utility analyses

KEYWORDS:

• Productivity cost
• cost-utility-analysis
• QALY
• double-counting
• net benefit framework

The debate regarding the inclusion of productivity costs (PCs) in cost-utility analyses (CUA) continues [1,2]. The concern is that PCs included in the numerator of an incremental cost-utility ratio (ICUR) may cause double-counting if the productivity loss is also incorporated in health state valuation. ICURs can be sensitive to the inclusion of indirect costs in the numerator [3], and the Washington Panel suggests that in the absence of explicit instructions to respondents to disregard income losses, analysts should assume that health state valuations include productivity loss and exclude them as costs [4].

As an example, Wielage et al. took a societal perspective in their analysis of treatment for osteoarthritis – a condition that affects working age adults. Productivity losses were explicitly included as costs while health state utilities were obtained from multiple sources, including a conversion from Western Ontario and McMaster Universities Osteoarthritis Index (WOMAC) scores, published literature, and the Canadian Community Health Survey (CHS) [5]. It is unlikely that a converted utility would incorporate income losses. The CHS however uses the Health Utilities Index to value health states, an instrument that does not explicitly instruct respondents to ignore income effects [6]. In another example, Tsiachristas et al. examined the cost utility of cardiovascular and chronic obstructive pulmonary (COPD) disease management programs in the Netherlands from the perspectives of the health-care system and society [7]. In addition to treatment and travel costs, the societal perspective analysis included PCs due to absence from paid labor, but these costs were limited to the friction period. Health states were valued using the EuroQoL 5 Dimensions (EQ-5D), which explicitly includes productivity losses from paid and unpaid labor in the usual activities dimension [8].

A systematic review found that in the absence of instructions, only a minority of respondents actually consider income loss in health state valuations [1]. The purpose of this editorial is to reexamine the valuation of productivity loss and consider the potential for double-counting in a net health benefit framework.

A key to resolving the numerator versus denominator conundrum is understanding how to value lost productivity. By using a preference-based approach, one may discover that an individual’s utility for labor productivity may not be exclusively related to income. There may be non-monetary components intrinsic to work participation, socialization, and fulfillment. This is distinct from health-related quality of life related to illness or treatment. Valuation is complicated by transfer payments such as sick pay and also depends on the particular health state. The multidimensional nature of productivity suggests that even when illness-related production losses are accounted for as a cost, there may remain aspects that should be captured in health state valuation.

Double-counting can occur when three conditions are met: (i) the illness affects survival, (ii) the life years (LYs) lost coincide with labor time losses, and (iii) more than one valuation is applied to the coincident LYs lost. The human capital method (HCM) [9] and quality-adjusted life year (QALY) estimation both include LYs. In HCM, lost years of productivity occurring within the study time horizon are multiplied by a wage rate to calculate the PC. For patients receiving treatment T, the PC of patient i is

(1)

where W and LE represent patient i’s wage and life expectancy, respectively, and where LY(T) represent patient i’s survival in LYs due to treatment T. The QALYs of patient i are represented by

(2)

where U(T) represents patient i’s health-state utility weight as a function of treatment T. It is clear that PCs and QALYs depend on LY(T) and each applies an independent valuation, W_i and U(T)_i, respectively. For studies of life-saving interventions, it is preferable to use lifetime horizons, and as the time horizon lengthens, the potential contribution of LYs to the overall value of PC and QALYs also grows. Because the analyst is only interested in incremental effects, LY is important only for interventions that are lifesaving or life-prolonging compared to standard care (S).

The total cost (TC) of patient k in either a treatment or control group can be represented as the sum of PCs and the vector of all component direct health-care costs (C1, C2, C3, etc.):

(3)

For simplicity, let us assume an intervention cost of £35,000 per year for T, £0 for S, and that all other direct health-care costs are negligible. Using Tilling’s example [1], a healthy 45-year-old patient, earning an annual wage of £30,000 (including benefits), will have 20 years of LE over a time horizon extending until retirement (age 65). If this patient suffers a disability that results in death after 10 years, then under standard care conditions TC(S) = PC(S) = £30,000 × (20–10) = £300,000. If a treatment T is available that completely averts morbidity and mortality, then LY = LE, PC(T) = 0, and TC(T) = (£35,000 × 20) = £700,000 (discounting has been omitted for illustrative purposes). The incremental TC(∆C) is equal to £400,000. This equates to an incremental cost of £40,000 per LY gained for T compared to S. Double-counting, perhaps better termed double-valuation, is not an issue because the benefits, expressed as LYs, are not valued in the denominator. However, if morbidity is observed in addition to mortality, both could be quantified in a CUA (where one instructs respondents to ignore income effects during health state valuation). A CUA will detect differences between groups in U as well as LY. If T prevents morbidity as well as mortality, then ∆QALYs will be >10. If ∆QALYs = 12, then the ICUR will be £33,333 and closer to the threshold acceptable to decision-makers.

Double-valuation will not be a problem as long as respondents ignore income effects during health state valuation [4]. However, economic evaluation now favors a linear net benefit framework to facilitate reporting and to depict results under a range of willingness-to-pay assumptions [10]. In an incremental net monetary benefit (INB) framework, another valuation occurs: the observed benefit, expressed as QALYs, is valued in monetary terms according to λ, the shadow price of a QALY gain from the payer decision-maker’s perspective. The INB of T can be determined from

(4)

By assigning a monetary value to λ, a CUA is converted to a linear analysis with cost and consequences expressed in monetary units. One can determine whether T will have a net positive benefit when a decision-maker applies a budgetary threshold value, such as £30,000 in the UK [11] or $50,000 per QALY in the US [12]. One can estimate INB(T) probabilistically by determining the proportion of simulations that result in INB >0 when parameter estimates vary along specified ranges and distributions. An advantage is the ability to assign confidence intervals to INB estimates for a range of λ values.

The question is whether applying λ represents another source of double-valuation. In the above example, there was an incremental cost of £40,000 per LY gained and an incremental cost of £33,333 per QALY gained for T compared to S when PCs were included in the numerator and income effects were excluded from health state valuation. If a QALY gain is valued at £30,000 and ∆QALY is equal to 12, then the INB(T) is equal to a net cost of £40,000, one-tenth of the original ∆C of £400,000!

As society is probably willing to pay for lifesaving interventions, then society should be willing to pay (no more than) the same shadow price for a QALY gained from any intervention. What remains unknown is whether ICURs consistently included PCs in their numerators. If not, then the incremental cost per QALY gained for lifesaving interventions may actually be lower than the accepted thresholds.

One might also ask if assigning a value to λ, i.e. willingness-to-pay for a QALY, is akin to quantifying the demand for health. Health can be viewed as capital whereby investment in health, through the consumption of health care, is a means to increase that capital [13]. A key consequence of building up one’s stock of health capital is increased productivity. Thus, the shadow price assigned to a QALY is conceptually related to valuing productivity time. This connection is explicit in efforts to peg the ceiling threshold to per capita gross domestic product (GDP). The World Health Organization has recommended that interventions with incremental costs per disability adjusted life year (DALY) less than three times per capita GDP be considered cost-effective [14]. While these recommendations are directed at developing countries where per capita GDPs are low, the concept has been applied to the US with a proposed ceiling threshold of twice the per capita private income per QALY [15].

When assigning a shadow price to a QALY, such as in a net benefit framework, it would seem prudent to omit PCs from the ICUR numerator when a societal perspective is taken. Double-valuation is only a risk insofar as the time interval that is being twice (or thrice) valued is the same period. In evaluations of chronic diseases that extend into the later years of life, only a fraction of the time horizon may be used for PC valuation in the numerator while the utility assessment may focus mostly on health effects in retirement years. The use of the friction cost approach would reduce PCs for chronic illnesses and may be preferable in this context. Sensitivity analysis should be used to assess the effects of modeling PCs for the segment of the time horizon during which PCs accrue.

If one is interested in knowing the incremental net benefit of an intervention, one might consider a cost–benefit analysis that steers clear of separate valuations of productivity time and utility, thus avoiding multiple valuations. More research into the use of contingent valuation or discrete choice methods to elicit willingness-to-pay for health-care interventions or for health state improvements that incorporate all relevant attributes, including effects on productivity, income, leisure time, and quality of life, is needed to provide theoretically grounded estimates that can be used in decision-making.

Declaration of interests

The author has 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. There are no conflicts of interest. No writing assistance was received. No funding was received.

Acknowledgment

The author would like to thank Dr. Dean Regier for valuable comments.