A flexible formula for incorporating distributive concerns into cost-effectiveness analyses: Priority weights.

BACKGROUND: Cost effectiveness analyses (CEAs) are widely used to evaluate the opportunity cost of health care investments. However, few functions that take equity concerns into account are available for such CEA methods, and these concerns are therefore at risk of being disregarded. Among the functions that have been developed, most focus on the distribution of health gains, as opposed to the distribution of lifetime health. This is despite the fact that there are good reasons to give higher priority to individuals and groups with a low quality adjusted life expectancy from birth (QALE). Also, an even distribution of health gains may imply an uneven distribution of lifetime health.
METHODS: We develop a systematic and explicit approach that allows for the inclusion of lifetime health concerns in CEAs, by creating a new priority weight function, PW = α+(t-γ)·C·e-β·(t-γ), where t is the health measure. PW has several desirable properties. First, it is continuous and smooth, ensuring that people with similar health characteristics are treated alike. For example, those who achieve 50 QALE should be treated similarly to those who achieve 49.9 QALE. Second, it is flexible regarding shape and outcome measure (i.e., caters to other measures than QALE), so that a broad range of values may be modelled. Third, the coefficients have distinct roles. This allows for the easy manipulation of the PW's shape. In order to demonstrate how PW may be applied, we use data from a previous study and estimated the coefficients of PW based on two approaches.
CONCLUSIONS: Equity concerns are important when conducting CEAs, which means that suitable PWs should be developed. We do not intend to determine which PW is the most appropriate, but to illustrate how a flexible general PW can be estimated based on empirical data.

Introduction

Prioritisation of limited health care resources involves saying no and yes, hard ethical problems and reasonable disagreements. In health economics, cost-effectiveness analysis (CEA) is used to identify the most efficient allocation of health care resources. Within such a framework, informed decisions based on explicit CEA rankings can be made. However, CEAs have been extensively criticised for not being sensitive to fair distribution of health benefits, and there are other important equity concerns which may be considered, and few standardised methods are available to directly adjust incremental cost-effectiveness ratios (ICERs) [1-3]. CEAs are used to inform policy decisions on the introduction or reimbursement of new technologies or pharmaceuticals in countries such as the UK, Norway, Sweden, the Netherlands, New Zealand and Australia, in some cases including considerations beyond CEA through concepts like severity or burden of disease [4-14].

Typical health metrics in CEAs are quality-adjusted life years (QALYs) gained and disability-adjusted life years (DALYs) averted [15]. These aggregate measures take into consideration not only the life expectancy (LE) of a group, but also the quality of life. Interventions and health programs are typically ranked by ICERs, where those that maximize QALYs gained or DALYs averted are favoured. However, such rank-orders of interventions based on ICERs have been criticized because they disregard concerns for those who are worst off [16]. There are good reasons to give higher priority to individuals and groups with impaired health [16]. As opposed to health maximisation, one could aim for equal distribution of health across all individuals in society. This would require that resources should always be directed at people with the most impaired health, so as to minimize inequality in health outcomes, even if this means that the total health in the population will be reduced. People’s intuition about what is fair may lie somewhere between strict health maximisation and strict equality. A compromise between these perspectives is that health maximization is an important goal, but health gains to people with impaired health are assigned more weight than similar gains to people with better initial health [17]. This “in between” position is called prioritarianism and has been defended by the philosophers like Derek Parfit [18] and Matthew Adler [19]. According to prioritarianism, a cheap and effective intervention directed at the healthiest may be preferred to an expensive and ineffective intervention directed at those who have worse health if left untreated. Still, if cost and effect were equal, the allocation of resources should be directed towards people with more impaired health if left untreated. Finally, when comparing competing health programs, studies have found that people favor priority to worse off, and they favor interventions that benefit those that are worse off over slightly more cost-effective interventions that benefit better off groups [20, 21]. Such equity concerns are too important to leave out from standard CEAs.

One may also argue that non-health concerns, like age, indirect benefits, productivity, and financial risk protection, are also important regarding equity. However, this paper focuses on health related concerns. At the time of a prospective health intervention, concerns for health distribution can be roughly divided into two perspectives: The first perspective is forward looking and is only concerned with the distribution of the health gains, typically measured in QALYs or DALYs. E.g., one may think that one QALY gained for 50 people is better, equal or worse than 50 QALYs gained for one individual. The second perspective is dealing with the distribution of health over a lifetime, often measured as quality adjusted life expectancy from birth (QALE) [22, 23]. E.g., consider two individuals who may benefit from some treatment. One has QALE = 80 without treatment and 81 with, and the other has QALE = 20 without treatment and 21 with. Their gains would be equal, but the second is expected to get much less lifetime health. Hence, if an equal distribution of lifetime health in the population is a priority, one would value the set {80, 21} above {81, 20}. The view that expected lifetime health is what matters draws support from the literature on health priorities and some empirical studies [24-28]. Note that although QALE often increases with age, a high age does not necessarily correspond to a high QALE. E.g., the QALE of an old person who has experienced decades of chronic illness may be far less than that of a healthy child. In other words, current age is not a good measure of QALE, and weights based on age are therefore different than weights based on lifetime health. In this paper we focus on applications of the lifetime health perspective.

Our aim is the development of a systematic and explicit approach to empirically include these equity concerns in the framework of CEAs, by creating a new priority weight function. The function should be smooth and continuous in order to treat people with similar characteristics alike. Further, it should be flexible regarding shape and outcome measure (e.g., cater to other measures than QALE) to be able to encompass a broad range of values. Finally, coefficients should have distinct roles, to ensure that different functions can be easily compared, and allow for easy manipulation of the function’s shape.

Methods

Moving from utility curves to priority weights

A social welfare function (SWF) is a function that represents the sacrifices that society is willing to make to promote a more equitable health distribution. The input should be a health distribution (individual QALYs, DALYs, etc.), and the output should be a real number, so that different distributions can be easily ranked. An SWF incorporates trade-offs between total health gain and health inequality in a population. Well-known SWFs are the Gini index and the Atkinson index [29, 30]. Both include a parameter that indicates the degree of inequality aversion, ranging from 0 to infinite. Such SWFs are data hungry, because complete data about the health or health gains for an entire population are needed. Often the data required do not even exist. A different set of SWFs is the one based on the aggregated QALY model. These utilitarian SWFs have the form where ti is the number of QALYs received by individual i (i = 1,…,n), and the utility function u(t), defined over ti, is positive and monotonously increasing (u(t)≥0, du/dt>0). If u(t) is concave (d2u/dt2<0), W represents societal preferences for equality. Equality indifference yields u(t) = t, and reduces W to be the sum of the QALYs received across all individuals in society. Finally, if u(t) is convex (d2u/dt2>0), society is assumed to prefer inequality. The function u(t) has been the focus of several papers [29, 31–34]. As opposed to the Gini and Atkinson indexes, calculating u(t) does not require complete information about the health or health gains of an entire population (t1,…,tn). We refer to Rodriguez and Pinto for a more detailed discussion on u(t) [34].

However, u(t) does not account for the distribution of lifetime health. We approach this problem by considering priority weight functions (PWs) instead. These are functions that adjust the weight of health gains according to health related equity concerns, such as health achievement in a lifetime perspective. A general form of an SFW based on PWs is where ti is the health level of individual i at the time of intervention, si is the health gain, and

The vertical bars denote that the absolute value is to be considered, and x is a dummy variable used for integration. As an example of how (2) may be calculated, consider a sick individual with a health level without intervention of QALE = 60, and an intervention with a gain of 3 QALYs. The priority weighted gain of the intervention is now .

Although the literature is lacking, one candidate for PW is the age-weight function of the Global burden of disease (GBD),

Here ti is the age of individual i [35]. As mentioned, age weights are beyond the scope of this paper. Still, by letting ti denote the health gain or lifetime health of individual i, (3) becomes a PW with quite an appealing form, although lacking in flexibility. A lack of flexibility would also be a problem if one would use a linear PW, such as

Flexibility is an important property of a PW, as it should be able to reflect a wide range of health related equity concerns. Further, if the coefficients of the PW have distinct roles, they can easily be adjusted to accommodate such concerns. E.g., one may wish to set an upper limit for the maximum weight, or set the maximum weight at a fixed value of t. Also, if the PW is fit to empirical data on people’s preferences, the roles of the coefficients will help to disentangle different concerns of society. One may, e.g., wish to estimate both the t for which the PW reaches its maximum, and what the maximum is. This is not straightforward in (3). While these concerns could be addressed using (4), a linear PW falls short in other aspects. E.g., a hump is useful if one wants to give lower weights to health gains benefitting those with a very low QALE (typically children born with severe conditions). If one wishes to do the opposite, and prioritize health gains to those who are expected to achieve very little lifetime health, PW may peak at t = 0, and then drop at a faster rate when t is close to 0 than when t is higher. A PW should also be twice differentiable with respect to t, so that it is continuous (no jumps) and smooth (no breaking points), ensuring the approximately equal treatment of individuals with similar values of t.

A priority weight function

In this section we present a new flexible PW with coefficients that have distinct roles, which is suitable in a lifetime health framework via (2). The formula is as follows:

Setting α = γ = 0 reduces (5) to (3). Setting C = 0.1658 and β = 0.04 we get the weights shown in Fig 1A. Panels B and C in Fig 1 show how PW1 can be altered by manipulating C and β. As seen, C has the effect of inflating or deflating PW1 (Fig 1B). For β>0, it can be shown that (5) ultimately approaches α with increasing t. We see that for larger β’s this happens more quickly than for smaller β’s (Fig 1C). When β≤0, (5) increases with t, and never approaches α at all.

Fig 1

Properties of the coefficients of (5) when α = γ = 0. Panel A: Priority weights according to t. Panel B: Altering C causes an inflation or deflation. Panel C: Altering β changes the “pointyness” of the hump, causing the peak to become more or less pronounced.

The intercept α of (5) shifts the curve up or down (Fig 2A), whereas γ provides a right-left shift (Fig 2B). As is shown in Fig 2B, it is possible to ensure that the maximum is at t = 0, making (5) strictly decreasing. Note that C is interpreted slightly differently for (5) than for (3). For (5), changing C will affect the relative weights between different values of t unless α = 0. Also, C≤0 is not a problem for (5), because α can be adjusted to make sure that PW1>0 for all relevant t’s. A negative C simply causes (5) to be reflected through the line PW1 = α, or in other words, turned up-side-down. If C = 0, (5) reduces to the line PW1 = α, corresponding to inequality indifference. This is also true if β approaches infinity.

Fig 2

Properties of the coefficients of (5) when C = 0.1658 and β = 0.04.

Panel A: Altering α causes an upward or downward shift. Panel B: Altering γ causes a left or right shift. Note that one would typically chose a PW where the maximum weight was at zero.

As demonstrated in Figs 1 and 2, manipulating the different coefficients of (5) allows us to flexibly change its shape. E.g., if we want to ensure that people with a certain value of t are given a certain weight (anchoring the curve), we may simply change α. Letting anchors the curve to α0 at t = A. Further, if we want to set a maximum weight, so that max(PW1) = wm, we let

If α0, A, γ, wm and β are given, we can treat (6) and (7) as two linear equations with two unknowns, and solve for C and α. Fig 3A shows (5) anchored to 1 at t = 80 for different values of wm.

Fig 3

Anchoring to 1 for t = 80, with β = 0.04.

Panel A: Varying max t (w) when γ = 0. Panel B: Varying γ when w = 2.5. For γ = 0, there is no shift. For γ = −1/β the curve is shifted left so that the maximum is at t = 0. For γ = −2/β the curve is shifted so that the slope is the steepest at t = 0.

As mentioned, manipulating γ will shift (5) right or left. It may be of particular interest to identify the shift needed to ensure that the maximum is moved to t = 0. This means that the slope is flat at t = 0, and gradually becomes steeper, so that gains for most individuals with poor lifetime health are given weights close to wm. Differentiating (5) with respect to t, we find that the maximum is when t = 1/β. Hence, letting γ = −1/β ensures that the maximum is at t = 0 (Fig 3B). Differentiating (5) once more with respect to t, the t for which the slope is the steepest can be identified as 2/β. Therefore, letting γ = −2/β causes (5) to be at its steepest when t = 0. In such a scenario, only gains to people with very poor health are given weights close to wm (Fig 3B).

Note that although varying α without changing the other coefficients just shifts PW1 up or down without changing its shape (Fig 2A), fixing the other coefficients in this manner is not something one would typically do. In order to keep predetermined properties of PW1, such as anchoring, a change in α would prompt a change in one or more of the other coefficients.

So far the mathematical properties of the coefficients in (5) have been presented, but the policy implications and conceptual meaning of these properties may not be obvious. The next paragraphs are dedicated to the discussion of these implications. First, we consider γ. If γ>−1/β, it is possible that PW(t1)< PW(t2) for t1children born with very severe conditions. In Figs 1 and 2 the default value of γ was 0.04, so that the maximum was at t = 1/0.04 = 25 QALYs, which is probably too high in most scenarios. When γ≤−1/β, a gain of, e.g., 1 QALY will always be weighted higher for the individual with the least lifetime health. From now on we shall assume that γ≤−1/β.

Next we consider α. The particular value of α is not necessarily easy to interpret, but because this coefficient allows for the anchoring of the PW at a certain weight for a certain value of t, it is useful when selecting a reference value of t for which PW = 1 (e.g., one may anchor PW at 1 when QALE = 70). Once a reference is chosen, the interpretation of the other coefficients is easier. A large β corresponds to a narrow peak of the PW, ensuring a steep decline of the curve when t is close to 0. Now, gains to individuals with little lifetime health are given much higher weights than gains to individuals with more lifetime health, whereas gains to individuals with intermediate and high levels of lifetime health will be weighted similarly. A small β yields a flat peak, so that the weight drops steadily as the initial health level improves (t increases). Although β also will affect the range of possible weights, this property is more closely connected to the C. The choice of C decides how much value will be assigned to a given t relative to the reference. E.g., one may think that the maximum weight should be no more than three times the weight of the reference. A very small C would ensure that all health gains were treated almost the same.

Application of the priority weight function to experimental data

In order to illustrate the potential use of (5) in a CEA within a lifetime health framework, we fit the function to data from a convenience sample. The procedure involves two conceptually simple steps. First, data must be collected, and then the curve must be fit. However, both these steps may be performed a number of different ways. We shall focus on an approach that is suitable for a dataset presented in Ottersen et al. (2014).

Ottersen et al. used a computer based questionnaire to ask 96 students in Norway how they valued QALYs gained for people at different initial health levels, measured in “initial QALE” (QALE without intervention). The reference case was a group of healthy people with QALE = 70 who were subject to an intervention that would give them 10 extra QALYs (i.e., move from QALE = 70 to QALE = 80). Respondents were asked to assign QALYs to individuals with worse initial health (QALEs in the set {10, 25, 40, 55}) in such a way that they were indifferent between the different scenarios. Now, priority weights could be calculated for each individual respondent based on the ratio between the QALYs assigned to the unhealthy groups and the reference group. E.g., consider a respondent who was indifferent between a situation where the group with initial QALE = 25 gained 5 QALYs (i.e., moving from 25 to 30) and the reference group gained 10 QALYs (i.e., moving from 70 to 80). This would mean that gains in the group with a worse initial health, where initial QALE = 25, are regarded as more important than gains in the healthy group, where initial QALE = 70. Mathematically, this extra weight can be expressed as V(25) = 10/5 = 2 relative to the reference group (QALE = 70). We refer to Ottersen et al. for a more thorough discussion of data and study design.

One may organize the data as follows. Two lists, V and T, are constructed. V contains all the weights assigned by the respondents, and T contains the initial QALEs, so that the weight Vi corresponds to the QALE Ti. For purposes of estimating the coefficients of (5), we treat the observations as independent. Applying a least squares approach, we get that the sum of squared errors (SSE) is

This expression may be minimized using a numerical routine or by solving the set of equations

If one wishes to anchor the curve to a specific value (e.g., the data from Ottersen et al. has V(70) = 1), it is straightforward to use (6) in (8). Setting maximum weights would imply using (7) in (8). Also, any coefficient in (8) may be kept constant.

Minimizing SSE ensures that the curve is fit to the mean V for each T in {10, 25, 40, 55}. Using the R function optim() to minimize SSE, we get

However, minimizing instead the sum of absolute errors, the curve is fit to the median. Now PW becomes

Note that when fitting the curves, α and γ were set to ensure that PW(10) was the maximum (i.e., PW(10) = wm) and PW(70) = 1. Fig 4 shows (9) and (11) plotted with the empirical data. Note that the PWmean is more affected by extreme values than are PWmedian.

Fig 4

Estimating PW from data.

The red line minimizes (8) to estimate a PW based on the means (black +). The coefficients are α = −0.42, γ = −22.2, C = 0.27, and β = 0.031. The pink line minimizes (10) to estimate a PW based on the medians (blue ×). The coefficients are α = 0.79, γ = −8.85, C = 0.17, and β = 0.053. In both scenarios PW(10) = wm (maximum weight at T = 10) and PW is anchored at PW(70) = 1.

Discussion

We have presented a new priority weight function (PW) with key characteristics that are important in consistent priority setting across patient groups and populations. As demonstrated, the PW was suitable for CEA in a lifetime health framework. The PW is flexible and easy to adjust to a broad range of equity considerations by altering a limited number of comprehensible parameters, each with its distinct role. We may set the range of priority weights by modifying C, we may shift the curve upwards or downwards by modifying α, we may shift the curve left or right by modifying γ and we may set the “pointiness” of the curve by modifying β. These characteristics make it easy, e.g., to anchor the curve or to set maximum weights. Considering (9) and (11) (Fig 4), we see that βmedian>βmean, indicating that PWmedian is pointier than PWmean. By this we mean that for values of QALE around 10, PWmedian decreases at a faster rate relative to its maximum than does PWmean. As QALE approaches 70, the relative rate of decrease is greater for PWmean. Further, we have that Cmedian

Drawing the PW by hand would of course be more flexible than using (5). However, we argue that setting some restrictions for the behaviour of the PW is useful. First, it offers consistency when comparing different curves. Second, the space of possible PWs is limited by (5), but the major restrictions caused by (5) seem reasonable (smooth and maximum one hump). Hence, a large number of nonsensical PWs are impossible (discontinuous and multiple humps).

We did not account for discounting of future health. It is not clear how this would affect rankings of interventions. A positive discount rate would put different weights on equally sized gains that were experienced at different times from the start of intervention (e.g., 0.5 QALYs gained immediately is valued higher than 0.1 QALYs gained each year in five years, or 0.5 QALYs five years from now). Hence, interventions with immediate effects would be valued over interventions where the benefit occurs in the future. Also, discounting health gains would give a relatively low priority to those with large gains, who are expected to have less lifetime health. This is counterintuitive in a lifetime health priority perspective.

A PW could be used theoretically to assess the impact of applying various ethical principles to rank health interventions, but ideally, the PW should be inferred from data. Although we do give an example of how this can be done, future work includes developing methodology to do inference about parameter estimates based on empirical data. This would allow for the testing of differences in preferences between groups. E.g., conditioning on the other variables, one could test if men and women preferred different values of C (determining the maximum weight) or γ (determining the t for which the maximum weight was obtained). When an appropriate PW (or a range of PWs) has been selected, it could be applied to evaluate the equity impact of different interventions on reduction of mortality or morbidity. Also, work should be done to include non-health concerns like socioeconomic status and productivity in the model, and to explore the consequences of priority-weighted CEAs across settings.

In this paper we do not intend to determine which PW is the most appropriate, but rather to illustrate what can be achieved by a flexible general PW. However, we would like to stress the fact that a flat PW is also a PW. In other words, there is no such thing as “not using a PW” in CEA. As illustrated, our framework allows for the estimation of PWs based on empirical data.

19 Aug 2019

PONE-D-19-18169

A flexible formula for incorporating distributive concerns into cost-effectiveness analyses: priority weights

PLOS ONE

Dear Mr. Haaland,

Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process.

Please find below the reviewers' comments.

We would appreciate receiving your revised manuscript by Oct 03 2019 11:59PM. When you are ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file.

If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter.

To enhance the reproducibility of your results, we recommend that if applicable you deposit your laboratory protocols in protocols.io, where a protocol can be assigned its own identifier (DOI) such that it can be cited independently in the future. For instructions see: http://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols

Please include the following items when submitting your revised manuscript:

A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). This letter should be uploaded as separate file and labeled 'Response to Reviewers'.

A marked-up copy of your manuscript that highlights changes made to the original version. This file should be uploaded as separate file and labeled 'Revised Manuscript with Track Changes'.

An unmarked version of your revised paper without tracked changes. This file should be uploaded as separate file and labeled 'Manuscript'.

Please note while forming your response, if your article is accepted, you may have the opportunity to make the peer review history publicly available. The record will include editor decision letters (with reviews) and your responses to reviewer comments. If eligible, we will contact you to opt in or out.

We look forward to receiving your revised manuscript.

Kind regards,

Valerio Capraro

Academic Editor

PLOS ONE

Journal Requirements:

1. When submitting your revision, we need you to address these additional requirements.

Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at

http://www.journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main_body.pdf and http://www.journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_title_authors_affiliations.pdf

Additional Editor Comments:

I have now collected two reviews from two experts in the field. Both reviewers are positive towards the paper but suggest minor revisions. Therefore, I am glad to invite you to revise your paper following the reviewers' comments. Looking forward for the revision.

[Note: HTML markup is below. Please do not edit.]

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #1: Yes

Reviewer #2: Yes

**********

2. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: Yes

Reviewer #2: I Don't Know

**********

3. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #1: Yes

Reviewer #2: Yes

**********

4. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #1: Yes

Reviewer #2: Yes

**********

5. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #1: The authors of this paper have proposed a flexible priority weight function that is designed to incorporate concerns about the distribution of lifetime health across individuals into cost-effectiveness analysis. I appreciate the contribution the authors have made. I have a few suggestions and questions below.

Additional clarity in the abstract may help readers better understand the contribution. First, it’s not clear what is meant by ‘similar health characteristics’. Is this measured using preference-weighted measures of health, like a utility weight corresponding to a given health state, at a moment in time, or does it account for accumulated health? Second, in the following sentence, the authors state that the function if flexible regarding shape and ‘outcome measures’. Do those outcome measures reflect health outcomes measured over the remaining lifetime for patients?

Also, I would prefer that the authors do not include the PWmean and PWmedian functions with weights in the abstract as those weights should not be construed as generalizable. They are based on a small convenience sample of students.

Unfortunately, there are no page numbers on the printed version of the paper.

On the first page after the abstract, line 34, the authors state that “… those that are worse off over slightly less cost-effective interventions”. I would have expected this to be ‘more cost-effective interventions’ to demonstrate that people generally prioritize less efficient investments in those that are worse off. Please confirm.

In the paragraph after [Figure 1 here], it is stated ‘in Figure 2B, the maximum can be moved beyond t=0 making (7) strictly decreasing’. It appears that when ϒ is -40, the maximum PW is at t=0 with decreasing PW. But, what is meant by ‘beyond t=0’? less than t? I note that later in the paper, it is stated that letting ϒ= -1/β ensures that the maximum is at t=0, so that seems to be a better way to ensure that the maximum PW is at 0.

In the paragraph after [Figure 2 here] ‘if we want to ensure that people with a certain value of t are given a certain weight (anchoring the curve), we may simply change α’. It appears to me that Figures 3A and 3B represent cases when the PW is anchored at t=80, indicating that one would want to vary Wm or ϒ. It appears that varying α would change the PWs across different values of t. Please try to explain this more clearly.

In the section on ‘Application of the priority weight function to experimental data’, the questionnaire described for the Ottersen et al. study does not seem to represent a discete-choice experiment wherein respondents were asked to choose which QALE and QALY gain scenarios they prefer. Instead, respondents were asked to assign QALYs with varying levels of initial health such that they would be indifferent. This weighting exercise does not represent a DCE. Please explain.

In the Discussion, lines 14-16, the authors point out that βmedian is greater than βmean indicating that PWmedian decreases at a faster rate than PWmean. In my view, according to Figure 4, it appears that PWmean decreases at a faster rate than PWmedian as PWmean starts at a higher value and both lines converge. Please explain. It is difficult to discern that PWmedian is ‘pointier’.

Reviewer #2: The authors address an important question which is well motivated by the background. I have comments about the communication and presentation that are not critiques of the overall approach.

While in general I appreciate the specificity, presenting the equation in the abstract without defining the variables in unclear, it is impossible for readers to interpret.

There should be clearer summary or conclusions in the abstract that are more detailed. The final sentence in the abstract is too general to be useful.

At the end of the second to last paragraph of the introduction, the authors discuss the relevance of age to QALE calculation. I found this to be seemingly important but the relevance was unclear. Perhaps the authors can rewrite this section.

The authors refer to another paper by Rodriquez and Pinto for discussion of u(t), but it would be useful to have more explanation here since it seems to be important for estimates.

It would be useful to answer the question of why we need a mathematical function at all, why not a linear slope?

Do the authors suggest that investigators pick values and then look at the curve, is the overall idea to find a curve that makes sense, or values that make sense, and should these be done a priori to looking at the data?

In figure 2, it seems unusual that the function should be lower at 0 and close to 0, is this intentional and/or desirable? It would be useful to have text explaining this.

**********

6. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: No

Reviewer #2: No

[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files to be viewed.]

While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email us at figures@plos.org. Please note that Supporting Information files do not need this step.

12 Sep 2019

Response to reviewers.

All page numbers and line numbers refer to the manuscript with tracked changes.

Initial comment from authors: The equation numbers were wrong, and we have changed these throughout the manuscript. We have also done some other minor changes to improve readability. These can be found in the manuscript with tracked changes.

Reviewer #1:

Additional clarity in the abstract may help readers better understand the contribution.

Comment: First, it’s not clear what is meant by ‘similar health characteristics’. Is this measured using preference-weighted measures of health, like a utility weight corresponding to a given health state, at a moment in time, or does it account for accumulated health?

Reply: We have added the following to the second paragraph of the abstract (lines 13-15 on page 2):

“For example, those who achieve 50 QALE should be treated similarly to those who achieve 49.9 QALE.”

We have also defined QALE at the end of the first paragraph (lines 7-8 on page 2).

Comment: Second, in the following sentence, the authors state that the function is flexible regarding shape and ‘outcome measures’. Do those outcome measures reflect health outcomes measured over the remaining lifetime for patients?

Reply: We have added the following to the Abstract (line 15 on page 2):

“(i.e., caters to other measures than QALE)”

and the last paragraph of the Introduction (lines 18-19 on page 4):

“(e.g., cater to other measures than QALE)”

Comment: Also, I would prefer that the authors do not include the PWmean and PWmedian functions with weights in the abstract as those weights should not be construed as generalizable. They are based on a small convenience sample of students.

Reply: We agree, and have deleted lines 20-23 on page 2.

Comment: Unfortunately, there are no page numbers on the printed version of the paper.

Reply: We are so sorry about that! Page numbers have been added!

Comment: On the first page after the abstract, line 34, the authors state that “… those that are worse off over slightly less cost-effective interventions”. I would have expected this to be ‘more cost-effective interventions’ to demonstrate that people generally prioritize less efficient investments in those that are worse off. Please confirm.

Reply: Of course, the reviewer is right. We have changed “less” to “more”, as suggested (line 34 on page 3).

Comment: In the paragraph after [Figure 1 here], it is stated ‘in Figure 2B, the maximum can be moved beyond t=0 making (7) strictly decreasing’. It appears that when ϒ is -40, the maximum PW is at t=0 with decreasing PW. But, what is meant by ‘beyond t=0’? less than t? I note that later in the paper, it is stated that letting ϒ= -1/β ensures that the maximum is at t=0, so that seems to be a better way to ensure that the maximum PW is at 0.

Reply: The relevant text now says (lines 11 and 12 on page 6):

“As is shown in Figure 2B, it is possible to ensure that the maximum is at t=0, making (5) strictly decreasing.”

Comment: In the paragraph after [Figure 2 here] ‘if we want to ensure that people with a certain value of t are given a certain weight (anchoring the curve), we may simply change α’. It appears to me that Figures 3A and 3B represent cases when the PW is anchored at t=80, indicating that one would want to vary Wm or ϒ. It appears that varying α would change the PWs across different values of t. Please try to explain this more clearly.

Reply: We thank the reviewer for pointing out that this was not clear, and have added a paragraph on lines 15-18 on page 7:

“Note that although varying α without changing the other coefficients just shifts PW_1 up or down without changing its shape (Figure 2A), fixing the other coefficients in this manner is not something one would typically do. In order to keep predetermined properties of PW_1, such as anchoring, a change in α would prompt a change in one or more of the other coefficients.”

Comment: In the section on ‘Application of the priority weight function to experimental data’, the questionnaire described for the Ottersen et al. study does not seem to represent a discete-choice experiment wherein respondents were asked to choose which QALE and QALY gain scenarios they prefer. Instead, respondents were asked to assign QALYs with varying levels of initial health such that they would be indifferent. This weighting exercise does not represent a DCE. Please explain.

Reply: We thank the reviewer for this comment. We have changed the text in Application of the priority weight function to experimental data (lines 6-7 on page 8) to the following:

“In order to illustrate the potential use of (5) in a CEA within a lifetime health framework, we fit the function to data from a convenience sample.”

We also changed the text in the Discussion (lines 17-21 on page 10) to the following:

“A PW could be used theoretically to assess the impact of applying various ethical principles to rank health interventions, but ideally, the PW should be inferred from data. Although we do give an example of how this can be done, future work includes developing methodology to do inference about parameter estimates based on empirical data.”

Comment: In the Discussion, lines 14-16, the authors point out that βmedian is greater than βmean indicating that PWmedian decreases at a faster rate than PWmean. In my view, according to Figure 4, it appears that PWmean decreases at a faster rate than PWmedian as PWmean starts at a higher value and both lines converge. Please explain. It is difficult to discern that PWmedian is ‘pointier’.

Reply: We have changed the relevant text (lines 21 on page 9 - line 1 on page 10) to:

“Considering (9) and (11) (Figure 4), we see that β_median>β_mean, indicating that PWmedian is pointier than PWmean. By this we mean that for values of QALE around 10, PWmedian decreases at a faster rate relative to its maximum than does PWmean. As QALE approaches 70, the relative rate of decrease is greater for PWmean.”

Further, we have changed lines 34-35 on page 7 to:

“A large β corresponds to a narrow peak of the PW, ensuring a steep decline of the curve when t is close to 0.”

Reviewer #2:

Comment: While in general I appreciate the specificity, presenting the equation in the abstract without defining the variables in unclear, it is impossible for readers to interpret.

Reply: See reply to Reviewer 1

Comment: There should be clearer summary or conclusions in the abstract that are more detailed. The final sentence in the abstract is too general to be useful.

Reply: We have revised the last sentence of the abstract (lines 24-27 on page 2) as follows:

“Equity concerns are important when conducting CEAs, which means that suitable PWs should be developed. We do not intend to determine which PW is the most appropriate, but to illustrate how a flexible general PW can be estimated based on empirical data.”

Comment: At the end of the second to last paragraph of the introduction, the authors discuss the relevance of age to QALE calculation. I found this to be seemingly important but the relevance was unclear. Perhaps the authors can rewrite this section.

Reply: We have tried to clarify the text (lines 9-13 on page 4):

“Note that although QALE often increases with age, a high age does not necessarily correspond to a high QALE. E.g., the QALE of an old person who has experienced decades of chronic illness may be far less than that of a healthy child. In other words, current age is not a good measure of QALE, and weights based on age are therefore different than weights based on lifetime health.”

Comment: The authors refer to another paper by Rodriquez and Pinto for discussion of u(t), but it would be useful to have more explanation here since it seems to be important for estimates.

Reply: We thank the reviewer for this comment. The brief review of u(t) and the social welfare functions is meant to motivate the development of PW, but we do not use them directly. Currently, we write:

“However, u(t) does not account for the distribution of lifetime health. We approach this problem by considering priority weight functions (PWs) instead.”

If this is not clear enough, we would be happy to rewrite the text. However, we prefer not to include more information about u(t), because u(t) cannot be used in a lifetime health perspective, which is the focus of this paper. At one point in the preparation of the manuscript, we did discuss u(t) in more detail, but internal reviewers found it confusing.

Comment: It would be useful to answer the question of why we need a mathematical function at all, why not a linear slope?

Reply: This is an important point, and we have added the following text where we discuss flexibility in Methods (lines 21-23 on page 5):

“A lack of flexibility would also be a problem if one would use a linear PW, such as

PW_linear (t_i)=〖a+b⋅t〗_i. (4)”

We also add the following in the next paragraph (lines 31-36 on page 5):

“While these concerns could be addressed using (4), a linear PW falls short in other aspects. E.g., a hump is useful if one wants to give lower weights to health gains benefitting those with a very low QALE (typically children born with severe conditions). If one wishes to do the opposite, and prioritize health gains to those who are expected to achieve very little lifetime health, PW may peak at t=0, and then drop at a faster rate when t is close to 0 than when t is higher.”

Finally, we have changed the text on lines 21-25 on page 7:

“First, we consider γ. If γ>-1/β, it is possible that PW(t_1)children born with very severe conditions.”

Comment: Do the authors suggest that investigators pick values and then look at the curve, is the overall idea to find a curve that makes sense, or values that make sense, and should these be done a priori to looking at the data?

Reply: We thank the reviewer for this comment. We have done the following changes in the Discussion (lines 1-4 on page 10):

“A PW could be used theoretically to assess the impact of applying various ethical principles to rank health interventions, but ideally, the PW should be inferred from data. Although we do give an example of how this can be done, future work includes developing methodology to do inference about parameter estimates based on empirical data.”

Comment: In figure 2, it seems unusual that the function should be lower at 0 and close to 0, is this intentional and/or desirable? It would be useful to have text explaining this.

Reply: We thank the reviewer for this comment, and have added the following to the figure text of Figure 2A (line 8 on page 13):

“Note that one would typically chose a PW where the maximum weight was at zero.”

Submitted filename: Response.docx

Click here for additional data file.

1 Oct 2019

A flexible formula for incorporating distributive concerns into cost-effectiveness analyses: priority weights

PONE-D-19-18169R1

Dear Dr. Haaland,

We are pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it complies with all outstanding technical requirements.

Within one week, you will receive an e-mail containing information on the amendments required prior to publication. When all required modifications have been addressed, you will receive a formal acceptance letter and your manuscript will proceed to our production department and be scheduled for publication.

Shortly after the formal acceptance letter is sent, an invoice for payment will follow. To ensure an efficient production and billing process, please log into Editorial Manager at https://www.editorialmanager.com/pone/, click the "Update My Information" link at the top of the page, and update your user information. If you have any billing related questions, please contact our Author Billing department directly at authorbilling@plos.org.

If your institution or institutions have a press office, please notify them about your upcoming paper to enable them to help maximize its impact. If they will be preparing press materials for this manuscript, you must inform our press team as soon as possible and no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact onepress@plos.org.

With kind regards,

Valerio Capraro

Academic Editor

PLOS ONE

Additional Editor Comments (optional):

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation.

Reviewer #1: All comments have been addressed

Reviewer #2: All comments have been addressed

**********

2. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #1: Yes

Reviewer #2: Yes

**********

3. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: Yes

Reviewer #2: I Don't Know

**********

4. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #1: Yes

Reviewer #2: Yes

**********

5. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #1: Yes

Reviewer #2: Yes

**********

6. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #1: (No Response)

Reviewer #2: Thank you for addressing all of my previous comments.

**********

7. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: No

Reviewer #2: No

2 Oct 2019

PONE-D-19-18169R1

A flexible formula for incorporating distributive concerns into cost-effectiveness analyses: priority weights

Dear Dr. Haaland:

I am pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department.

If your institution or institutions have a press office, please notify them about your upcoming paper at this point, to enable them to help maximize its impact. If they will be preparing press materials for this manuscript, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact onepress@plos.org.

For any other questions or concerns, please email plosone@plos.org.

Thank you for submitting your work to PLOS ONE.

With kind regards,

PLOS ONE Editorial Office Staff

on behalf of

Dr. Valerio Capraro

Academic Editor

PLOS ONE