Valuing the Benefits of Weight Loss Programs - Wiley Online Library

Valuing the Benefits of Weight Loss Programs: An Application of the Discrete Choice Experiment Larissa Roux,*† Christina Ubach,‡ Cam Donaldson,*§ and Mandy Ryan‡

Abstract ROUX, LARISSA, CHRISTINA UBACH, CAM DONALDSON, AND MANDY RYAN. Valuing the benefits of weight loss programs: an application of the discrete choice experiment. Obes Res. 2004;12:1342–1351. Objective: Obesity is a leading health threat. Determination of optimal therapies for long-term weight loss remains a challenge. Evidence suggests that successful weight loss depends on the compliance of weight loss program participants with their weight loss efforts. Despite this, little is known regarding the attributes influencing such compliance. The purpose of this study was to assess, using a discrete choice experiment (DCE), the relative importance of weight loss program attributes to its participants and to express these preferences in terms of their willingness to pay for them. Research Methods: A DCE survey explored the following weight loss program attributes in a sample of 165 overweight adults enrolled in community weight loss programs: cost, travel time required to attend, extent of physician involvement (e.g., none, monthly, every 2 weeks), components (e.g., diet, exercise, behavior change) emphasized, and focus (e.g., group, individual). The rate at which participants were willing to trade among attributes and the willingness to pay for different configurations of combined attributes were estimated using regression modeling.

Received for review August 12, 2003. Accepted in final form June 14, 2004. According to U.S. code, all journals requesting payment of author page charges in order to defray the cost of publication are required to publish a disclaimer. This article must, therefore, be marked ‘‘advertisement’’ in compliance with U.S.C. Section 1734 solely to indicate this fact. *Department of Community Health Sciences, University of Calgary, Calgary, Canada; †Division of Nutrition and Physical Activity, Centers for Disease Control and Prevention, Atlanta, Georgia; ‡Health Economics Research Unit, University of Aberdeen, Aberdeen, United Kingdom; and §Department of Economics and Centre for Health Services Research, University of Newcastle upon Tyne, Newcastle upon Tyne, United Kingdom. Address correspondence to Larissa Roux, Division of Nutrition and Physical Activity, Centers for Disease Control and Prevention, 5612 Elm Street, Vancouver, British Columbia, Canada V6N 1A4. E-mail: [email protected] Copyright © 2004 NAASO

1342

OBESITY RESEARCH Vol. 12 No. 8 August 2004

Results: All attributes investigated appeared to be statistically significant. The most important unit change was “program components emphasized” (e.g., moving from diet only to diet and exercise). Discussion: The majority of participants were willing to pay for weight loss programs that reflected their preferences. The DCE tool was useful in quantifying and understanding individual preferences in obesity management and provided information that could help to maximize the efficiency of existing weight loss programs or the design of new programs. Key words: discrete choice experiment, willingness to pay, process of care, weight loss

Introduction With more than one-half of the North American population clinically overweight or obese (1,2), it is now wellrecognized that obesity is a public health crisis, that does not seem to be abating (3). Its accompanying health consequences (e.g., coronary heart disease and type 2 diabetes) also exact large personal and economic tolls from individuals and from society. Many recent studies have expressed the impact of obesity through the cost burden imposed by its associated downstream health effects (4 –7), but few have taken the perspective of costing the potential solutions aimed at treating obesity. With the exception of several efforts that have evaluated pharmaceutical or surgical therapies (8 –11), determination of the value to be gained from interventions to treat obesity has received little attention. There is strong evidence from clinical trial research to support multidisciplinary weight management as the most effective strategy for weight loss (12,13). In clinical practice, delivery of comprehensive care through health care remains challenging, and most noncommercial weight loss programs continue to be modeled as disease-driven and provider-centered (14). Program effectiveness is often based solely on participant weight change. Little emphasis

Valuing the Benefits of Weight Loss Programs, Roux et al.

has been directed toward the study of potential factors perceived as important by enrolled participants of the actual weight loss program process and toward the design of participant-centered programs (15,16). Because the responsibility for achieving successful weight loss, to a great degree, falls on the shoulders of the individuals attempting weight loss and that their success, in most instances, is related to individuals’ willingness and ability to comply with a given program, understanding which factors beyond weight loss may influence program choice and compliance is imperative and deserves more academic inquiry. From its origins in mathematical psychology in the 1960s (17), the discrete choice experiment (DCE)1 survey instrument has more recently been adapted to and applied in a number of different areas in health economics, including eliciting patient and community preferences in the delivery of health services, establishing consultants’ preferences when setting priorities across competing management strategies, developing outcome measures, evaluating alternatives within randomized controlled trials, and establishing patients’ preferences in the doctor-patient relationship (18). At the methodological level, it has been found that respondents complete DCEs in an internally valid and consistent manner (18 –20). This survey instrument is particularly well-suited to the exploration of intervention-specific impacts of service attributes by mimicking everyday decisionmaking processes (21–23). DCEs are attribute-based measures of benefit. This experiment technique is based on the premises that, first, any good or service can be described by its characteristics (or attributes) and, second, that the extent to which an individual values a good or service depends on the levels of these characteristics. The technique involves presenting choices to individuals that vary with respect to the levels of the attributes. Individuals are generally asked which scenario they prefer from a choice set (e.g., “A” or “B”), and choice responses can then be used to inform the following: to establish the importance of a given intervention attribute, to assess the rate at which an individual is willing to give up one attribute to have improvements in another (i.e., the marginal rate of substitution), and, assuming a price proxy is included as an attribute, to estimate willingness to pay (WTP), a monetary measure of benefit (23). Information derived from choice responses can also be used to generate overall benefit scores for alternative ways of providing health care. This latter application allows results to be presented such that they can be used to rank health services against one another, for instance, in a priority setting context. The DCE has recently gained popularity in medical applications and has been applied in a number of clinical areas

1

Nonstandard abbreviations: DCE, discrete choice experiment; WTP, willingness to pay.

(24 –29). However, with the exception of one WTP study, which, employing a different methodology, examined the economic benefits of surgical treatment in the severely obese (30), service-related benefits of nonsurgical weight loss programs, expressed directly or indirectly as WTP, have yet to be investigated by stated preference techniques, including DCE. Using the DCE approach, this study examined participant preferences and estimated WTP for service-related attributes of community-based weight loss programs. From the perspective of the program participant, the relative importance of identified program characteristics and the value attributed to alternative program configurations defined on the basis of these characteristics were explored. For the purpose of this study, the program characteristics explored using DCE were identified through an analysis of three facilitator-led focus group discussions, with 20 participants of a community-based weight loss program, who met the same inclusion/exclusion criteria as did participants in the DCE study sample. The qualitative analysis was aimed at identifying factors that impacted participant choice and adherence to a particular weight loss program, using a structured interview instrument developed for this purpose. The focus group instrument was reviewed by an external team of experts, consisting of a physician, qualitative research methodologist, health economist, weight loss program coordinator, and focus group facilitator, to ensure content validity. All sessions were audio-taped and transcribed for subsequent detailed descriptive analysis. Information derived from discussions was reduced to themes. To enhance validity of the data, the emerging themes were reported back to the group members, who were then asked to verify these themes and to raise issues that had not already been addressed in the session. Data were then sorted and coded according to these themes several times, in an iterative manner. The facilitator and investigator examined the transcribed reports independently and then discussed and compared their individual analyses until they reached agreement regarding the results and no unresolved issues remained. Major themes that emerged from the analytic process included barriers and facilitators to weight loss program involvement. Under the theme of barriers, program costs and time required to access a program were felt to be universally challenging issues for discussants attempting weight loss. Similarly, unanimous consensus was reached that consideration of program focus, of emphasized components of a given program, and of the extent to which a physician was involved in the weight management process was critical to decision-making regarding choice of and adherence to a particular weight loss program. Given the value placed on these five attributes by focus group discussants, these characteristics were chosen to form the basis of the DCE survey instrument explored in this study. OBESITY RESEARCH Vol. 12 No. 8 August 2004

1343


Table 1. Attributes and levels used in DCE Levels

Level coding

250 500 750 1000 Up to 15 Up to 30 Up to 45 Up to 60 None Monthly basic check-up Counseling every 2 weeks Diet only Diet and exercise Diet, exercise, and self-management General group Group-specific Personal

250 500 750 1000 15 30 45 60 1 2 3 1 2 3 1 2 3

Attribute Program cost (for 3 months, $)

Travel time to program (one-way, minutes)

Amount of doctor involvement

Program components emphasized

Focus of program

Research Methods and Procedures This DCE study employed a cross-sectional, nonrandomized experimental design to elicit preferences regarding service characteristics of community-based weight loss programs from participants of such programs. Stages of a DCE A discussion of the five stages of a DCE follows, within the context of assessing the contributions of different characteristics comprising community weight loss programs. Stage 1: Identification of Relevant Effects (Attributes). As mentioned in the introduction, based on focus group discussions undertaken before this study that explored weight loss program service characteristics deemed important to enrolled participants, five attributes were identified as being most important: program cost, program travel time, extent of physician involvement, components of program emphasized, and level of program tailoring or focus. The inclusion of cost allowed estimation of individuals’ WTP values (a monetary measure of benefit) for changes in levels of individuals provided in the model, as well as overall WTP for programs with given attributes. Attributes were also valued in terms of travel time. Stage 2: Establishing Levels for These Attributes. Once attributes were defined, the next step involved assigning levels to them. Identification of levels posed several challenges; including determining the appropriate range for, and 1344


Unit of measurement

Expected effect

Dollars

Negative

Minutes

Negative

Not applicable

Positive

Not applicable

Positive

Not applicable

Positive

intervals between, levels of each attribute and defining levels for qualitative attributes. For example, it was important to achieve a balance between presenting cost levels beyond the price currently being paid for the service (because the actual cost may not represent an individual’s maximum WTP) and not presenting a cost level so high that the individual refused to make a discrete choice. It was also important to choose a “currency” that was both familiar and relevant to participants; that is, costs were expressed in terms of Canadian dollars, and levels were generated with a priori knowledge of the pricing variability of existing weight loss programs offered in the region at the time the study was conducted. Table 1 summarizes the attributes and levels included in this DCE. Stage 3: Determination of Relevant Scenarios to Present. A fractional factorial design was used to develop scenarios based on attribute level combinations (31). Using this approach, the 432 possible scenarios (two attributes with four levels and three attributes with three levels, or 42 ⫻ 33 ⫽ 432) were reduced, using methods described elsewhere (32), to a more manageable number that still permitted inferences about all possible scenarios. This reduction technique, which ensured absence of both multicollinearity and attribute interactions, gave rise to eight choices. These scenario pairs were presented in the form of a paper-based self-administered questionnaire. Two additional choice sets were added to test internal consistency. Here, one of the


pendent and linear effect on preferences, was used to estimate the following benefit function: ⌬benefit ⫽ ␹1⌬cost ⫽ ␹2⌬time ⫹ ␹3⌬doctor ⫹ ␹4 ⌬component ⫹ ␹5 ⌬focus ⫹ ␧1 ⫹ ␧2

Figure 1: Sample question from questionnaire.

presented scenarios of the discrete choice was as good or better across all attributes than the other; therefore, it was expected that respondents would choose the “better” scenario. An example of a discrete choice question is presented in Figure 1. Stage 4: Establishing Preferences. Preferences were elicited from members of the community enrolled in community weight loss programs. Within a 10-week interval in the spring of 2001, in Calgary, Alberta, 200 people expressed interest to participate in this study, in response to a citywide community recruitment effort (i.e., using flyers, public service announcements, radio, television, and print media). Based on the following criteria, 25 years of age or older, overweight or obese (BMI ⱖ 25 kg/m2), recent enrollment in a community weight loss program, not pregnant or nursing, and absence of clinical comorbidities, 172 individuals were determined to be eligible and were invited over the phone or by e-mail to participate. Of those, 165 completed the DCE, which was administered at the University of Calgary Sport Medicine Centre. Of the seven individuals who declined to participate, three were in search of a weight loss study to be part of, and four individuals felt they could not afford the time to complete the DCE survey. This study received ethical approval from the Health Research Ethics Board at the University of Calgary. After written informed consent was obtained, questionnaires were administered and completed at the University site or, for those who were unable to find transportation, at their work sites or personal residences. Confidentiality of responses was ensured with survey coding, secured storage, and collection of data and aggregated data entry into a computerized database for purposes of analysis. Questionnaires were self-administered in the presence of a study facilitator. Participants were encouraged to ask questions at any time and were informed that they could terminate the session or skip individual questionnaire items. Stage 5: Analysis of Data. Regression techniques (random effects probit modeling) were used to estimate an attribute-based utility (benefit) function, after inconsistent and inappropriate responses were excluded. A linear additive model, which assumes that each attribute has an inde-

where ⌬benefit represents the change in benefit of moving from one weight loss program (A) to another (B), ⌬cost reflects the difference in program cost for 3 months between the two programs, and ⌬time, ⌬doctor, ⌬component, and ⌬focus terms, represent the differences in one-way travel time, extent of doctor involvement, program components emphasized, and level of program focus, respectively. ␹j (j ⫽ 1,2,3,4,5) represents the coefficient for each attribute. These coefficients indicate whether the given attributes are influencing the decision-making process. Given that lower levels for travel time or cost are intuitively preferable, it was expected that these attributes would be associated with a negative coefficient. No a priori assumptions for the qualitative attributes were made. The unobservable error terms are represented by ␧1 and ␧2, where ␧1 is the error term due to differences among observations and ␧2 is the error term due to differences among respondents. The ratio between the coefficients for two attributes reflects the marginal rate of substitution between those attributes. The denominator of the ratio provides the currency for value. For example, ␹j/␹2 (j ⫽ 1,3,4,5) reflects willingness to incur additional travel time to program for improvements in attributes. Similarly, ␹j/␹1 (j ⫽ 2,3,4,5) reflects WTP (in money) for improvements in all other attributes. To investigate how preferences varied across subgroups, two additional multivariate regression models, with attribute coefficients segmented by either BMI level (i.e., obese state, BMI ⱖ30 kg/m2, and nonobese state, BMI ⬍30 kg/m2) or income level (⬍ or ⱖ$75,000) were explored. The Wald statistic, for which the null hypothesis assumes no difference in coefficients across BMI level or income level, was used to identify differences between coefficients of obese vs. nonobese individuals or individuals from households with low vs. high incomes for each attribute. Finally, overall WTP for the introduction of a number of hypothetical weight loss programs, where it was assumed that no program previously existed, and which were defined by various combinations of attribute levels, was also estimated. Attribute level configurations for these explored hypothetical programs included several that most resembled the status quo programs in the community in which the study was conducted, whereas others were modeled based on varying suggestions of focus group discussants as to what configurations might describe an “ideal program.” For a given program Y, overall WTP equates to the sum of individual attribute WTP values and is represented by the following expression: OBESITY RESEARCH Vol. 12 No. 8 August 2004

1345


Table 2. Respondent sociodemographic characteristics* Category

% Respondents (n ⴝ 158)

25 to 34 35 to 44 45 to 54 55 to 64 65⫹ Women Men White African American Asian Other Married Never married Widowed/divorced Some high school High school graduate Some university/college University/college graduate Graduate school ⬍15,000 15,000 to 24,999 25,000 to 49,999 50,000 to 74,999 75,000 to 99,999 ⬎100,000 Work full-time Work part-time Unemployed, looking for work Unemployed, not looking for work Homemaker In school Retired Disabled Other

11.4 24.7 33.5 23.4 7.0 82.9 17.1 92.4 0.6 3.2 3.8 64.6 17.7 17.7 5.7 15.3 39.5 24.9 14.6 2.6 8.4 21.2 23.3 14.2 30.4 52.3 10.2 1.9 1.9 7.7 4.5 5.1 1.3 15.1

Characteristic Age group (years)

Gender Race

Marital status

Completed educational level

Household income ($)

Employment status

* BMI was considered as a continuous variable; its distribution was skewed to the right, with a mean of 32.3 kg/m2 (SD ⫽ 6.10 kg/m2) and a median of 31.1 kg/m2.

Overall WTP Program Y ⫽ ⌺共␹2 /␹1 ⫻ ␹3 /␹1 ⫻ ⌬doctor ⫹ ␹4 /␹1 ⫻ ⌬component ⫹ ␹5 /␹1 ⫻ ⌬focus兲

Results

Overweight and obese respondents (165; median BMI ⫽ 31.1 kg/m2; mean BMI ⫽ 32 kg/m2; SD ⫽ 6.7) completed 1346


the DCE (Table 2). The majority of the subjects were women, which is consistent with the higher prevalence rates among women of both obesity and attempted weight loss (33,34). The vast majority of subjects were white, and more than one-half were married and employed full-time. Four individuals (2.4%) failed the consistency check questions, and three individuals (1.8%) gave incomplete answers, suggesting that they misunderstood instructions on how to


Table 3. Results from the regression model

Attribute* Program cost for 3 months One-way travel time to program Amount of doctor involvement Program components emphasized Program focus

Difference (A ⴚ O)

Benefit marginal WTP ($)‡

Benefit marginal time value (minutes)

Coefficient† (SE)

Marginal WTP ($)

Marginal time value (minutes)

␹1 ⫽ ⫺0.0022 (0.0002) ␹a ⫽ ⫺0.019 (0.002)

N/A

⫺0.12

0

250

250

N/A

⫺30

⫺8.45

N/A

0

15

15

⫺127

N/A

␹3 ⫽ 0.158a (0.067) ␹4 ⫽ 0.613 (0.041) ␹5 ⫽ 0.293 (0.055)

No program (0)

Proposed program (A)

71.95

8.51

0

3

3

216

26

278.55

32.95

0

3

3

836

99

133.18

15.75

0

3

3

399

47

Overall WTP§ Overall travel time**

1324 142

* For accompanying levels, see Table 1. † Significance level, p ⱕ 0.01; with exception of a, p ⫽ 0.019. ‡ Benefit marginal WTP for one-way travel time ⫽ (Marginal WTPtravel ⫻ Difference (A ⫺ O) ⫽ (␹2/␹1) ⫻ (A ⫺ O) ⫽ [(⫺0.019/⫺(⫺0.0022)) ⫻ (15 ⫺ O)] ⫽ 127 § Overall WTP ⫽ ⌺ attribute benefit marginal WTP. ¶ Benefit marginal travel time for program focus ⫽ [Marginal travel timefocus ⫻ Difference (A ⫺ O)] ⫽ (␹5/␹2) ⫻ (A ⫺ O) ⫽ [0.293/⫺(⫺0.019)) ⫻ (3 ⫺ O)] ⫽ 47 ** Overall travel time ⫽ ⌺ attribute benefit marginal travel time.

complete the questionnaire. The survey results of these seven respondents were eliminated from further analysis, leaving 1105 useable individual observations from 158 participants (95.8% of data) for analysis. Results from the DCE analysis are shown in Table 3. All attribute coefficients were statistically significant and had the expected sign, respectively reflecting their importance as contributing attributes of a weight loss program and lending support to the theoretical validity of the model. For attributes associated with a positive coefficient, an upward shift in level was preferable to a downward shift, suggesting that one would trade or “give up” something else to move up a level. Conversely, the negative coefficients for program cost and travel time variables reflected that as the

absolute magnitude of each of these attributes rose, the less preferable the attribute became. The relative importance of the different attributes in the design of weight loss programs was inferred from the magnitude of attribute coefficients. However, coefficients were not directly comparable across attributes because the unit of measurement for each coefficient varied across these attributes. For example, although the coefficient for “amount of doctor involvement” was associated with a larger coefficient than was “one-way travel time to program,” suggesting that level of doctor involvement was relatively more important, if travel time was reduced by 10 minutes (a plausible and potentially meaningful reduction), its corresponding coefficient, ⫺0.19, would be greater in absolute OBESITY RESEARCH Vol. 12 No. 8 August 2004

1347


Table 4. Willingness to pay for weight loss program services: comparison of subgroups segmented by BMI level Attributes*

Coefficient

p Value

Marginal WTP

(␹1) Program cost for 3 months (␹2a) One-way travel time to program at BMI ⱖ 30 kg/m2 (␹2b) One-way travel time to program at BMI ⬍ 30 kg/m2 (␹3a) Amount of doctor involvement at BMI ⱖ 30 kg/m2 (␹4) Program components emphasized (␹5) Program focus

⫺0.0022 ⫺0.0139 ⫺0.0222 0.1763 0.6056 0.2901

⬍0.001 ⬍0.001 ⬍0.001 0.043 ⬍0.001 ⬍0.001

N/A ⫺6.32 ⫺10.09 80.14 275.27 131.86

N ⫽ 158. Log likelihood ratio for model ⫽ ⫺388.38 * Coefficient for amount of doctor involvement at BMI ⬍ 30 kg/m2 was not statistically significant; therefore, this subgroup was not included in model.

Restriction/ null hypothesis

␹1a ␹2a ␹4a ␹5a

magnitude than the coefficient associated with a level change in amount of doctor involvement [(⫺0.019 ⫻ 10) ⬎ 0.158, from Table 3]. With time as the trading currency, participants were willing to increase program-related travel time by 33 minutes for an upward shift of one level in emphasized program components and ⬃16 or 9 minutes, respectively, for similar marginal shifts in program focus level or level of doctor involvement. They were willing to accept a 12-minute increase in travel time for a $100 reduction in 3-month program cost (⫺0.12 ⫻ 100). Similarly, when trading was expressed using dollars as the metric, respondents were willing to pay about $85 for 10 minutes less of travel [(⫺0.019 ⫻ 10)/⫺0.0022] or $278 for a single upward shift in program components emphasized. Table 4 shows the results of the BMI segmented model. Based on the Wald statistic, one-way program travel time was the only attribute to vary significantly across BMI categories. Also, respondents within the higher BMI subgroup were willing to pay more (about $80.00 compared with about $72.00) for a positive shift in the level of doctor involvement in their care when compared with the overall cohort. Attribute coefficients for other qualitative variables were not statistically significant for BMI subgroups and, therefore, were simply evaluated without segmentation. The expected income effect was not observed, with the exception of its statistically significant effect on program 1348


⫺ ⫺ ⫺ ⫺

␹1b ␹2b ␹4b ␹5b

⫽ ⫽ ⫽ ⫽

0 0 0 0

p Value 0.88 0.04 0.28 0.92

focus, for which individuals with lower household incomes valued positive shifts more than did higher household income individuals. (These results are available from the authors on request.) Assuming there was no weight loss program being offered, overall WTP for the introduction of a proposed program characterized by a one-way travel time of no more than 15 minutes, physician counseling every 2 weeks, a multicomponent emphasis of diet, exercise, and self-management, and a personalized focus, was $1324 (i.e., ⫺127 ⫹ 216 ⫹ 836 ⫹ 399) (Table 3). In this example, the change in one-way travel time [⫺127, or $127 (the absolute value] contributed least to overall WTP, with the smallest marginal WTP value, whereas program components emphasized contributed most, with a marginal WTP of $836. Using this approach, the introduction of a variety of simulated programs, each with unique attribute configurations, was explored to estimate the overall WTP for each program (Figure 2). The most preferred of the seven alternatives presented was “Program 1” (example in the previous paragraph) because it was associated with the highest overall WTP estimate ($1324). This simulated program assumed the highest level for each qualitative attribute and the lowest travel time level and was suggested in focus group discussions to represent an “ideal” scenario. Program 2 was generated to most closely resemble a program offered in the community of which study participants were a part. This


Figure 2: Ranking of simulated programs based on overall WTP relative to having no available program.

more “realistic” program (i.e., comprised of the following attribute levels: up to 30 minutes of travel time; no doctor involvement; three-component emphasis of diet, exercise, and self-management; and focused at a general group) was ranked fourth. The least preferred program (ranked seventh) of the alternatives presented was “Program 5” (Figure 2).

Discussion Notwithstanding growing evidence of the negative health and economic consequences of obesity, its prevalence continues to rise, and in its wake, a $40 billion/year weight loss industry thrives (35). The current unavailability of therapeutic strategies to address its chronicity and multifaceted etiology and to widely disseminate comprehensive and individualized care perpetuates this alarming trend. This study adds to a growing body of literature on DCE applications to preference elicitation in health care decisionmaking (25,28). To our knowledge, it is the first to use this quantitatively rigorous and theoretically driven approach to evaluate WTP for the design of community-based weight loss programs and highlights some important findings. In this study, substantial value was attributed by weight loss program participants to individual and combined service-related program characteristics. Assuming the best program currently available in the community in which the study took place (i.e., program 2) as the comparator, study participants were willing to pay an extra $600 out-of-pocket for a 3-month weight loss program that was more accessible, comprehensive, and tailored (Program 1, Figure 2). Although, as with other health-related WTP studies, it might be argued that such exercises involve hypothetical expenditures rather than actual purchasing decisions and, therefore, may overestimate WTP, they clearly highlight relative preferences for program characteristics and for alternative overall program configurations.

The potential for obese participants to experience different preferences from nonobese individuals and for WTP measures to reflect ability rather than willingness to pay prompted the exploration of the chosen segmented models. Two findings were particularly interesting. First, a finding that may seem unexpected was that nonobese individuals attributed greater value to reducing travel time. Second, from the income-segmented model, individuals with lower household incomes valued more focused care than did those from higher income households. In both of these instances, these observations might be attributable to varying opportunity costs of the particular subgroups, such that individuals with lower levels of BMI and income have higher opportunity costs, respectively, with respect to travel time and program focus. Thinner participants might not find their weight to be enough of an issue to bother traveling for long periods to address it, whereas heavier individuals might be more willing to accept longer travel times to have access to a program that would otherwise best meet their needs. Furthermore, individuals from lower income households, who have less disposable income, might feel they are “getting more value for their money” by maximizing their interaction with a health care professional through a more focused program. Such data could inform both clinicians and policy makers. Health care providers involved in obesity management could use this information to help tailor programs on the basis of preferred attributes to better meet the needs of individual patients. Although this would need to be empirically tested, it is speculated that meeting individuals’ needs has the potential to result in improved program compliance, which might, in turn, have the added benefit of being associated with better health outcomes. Although the relative preferences of attributes could assist in the design of a given weight loss program, the ranking exercise could facilitate decisions on how to spend additional money to improve an existing program. For instance, if weight loss programs were to be publicly funded, regional decision makers making use of these data could generate different scenarios (even including program configurations not included in the questionnaire) and compare them, and to maximize efficiency, could attempt to introduce the best one possible (the one with the highest overall WTP) with given resources. This study also valued attributes in terms of travel time. For users of health care in a publicly funded system, in which individuals are not used to paying out-of-pocket for health care services, time might be a more familiar trading metric. From a policy perspective, such information could guide decisions regarding the optimal number of required program facilities within a region and their geographic proximity to one another to best meet participant needs. Limitations of this study stemmed from the size and selection of the sample and assumptions made. First, the OBESITY RESEARCH Vol. 12 No. 8 August 2004

1349


study cohort formed a self-selected convenience sample, which may not have been representative of the general adult population attempting weight loss. Specifically, although the characteristics of respondents in our study sample did, in fact, represent those who were enrolled in the communitybased weight loss program we examined, it is certainly possible that men’s views may have been underrepresented in our study. Addressing how measured preferences in a sample of weight loss participants compare with those of the overweight population at large should be a priority of future research. Second, although the distribution of BMIs across this sample resembled that of the overweight and obese adult population in Canada, a limited sample size precluded finer BMI- or income-dependent segmentation. Last, the notion of linearity assumed by the model ignored the potential for nonlinear relationships among attributes to be present. However, in research efforts external to health care, alternatives to the linear function have seldom resulted in a significantly better model fit (36). DCE was found to be a feasible technique for extracting WTP information and served as a particularly useful instrument in eliciting preferences for services that do not currently exist. Although the focus of this study highlighted the specific applicability of DCE to the provision of weight loss programs, it is hoped that readers will appreciate that this method may be applied to a variety of study queries in the arena of obesity research and will encourage further exploration in this area. For instance, satisfaction scores (utilities) or WTP for drug therapy in obesity could be explored in this way: attributes might include chance of 10% weight loss at one year, cost of prescription, side effect profile, and duration of treatment. Results from this study also suggest that service attributes play a marked role in the decision-making practices of individuals faced with choosing a weight loss program. Such findings may provide insight into the design of future programs and may draw attention to another source of data for consideration in future economic evaluations comparing such programs. Future research to expand on these results should include the study of a cohort large and diverse enough to enable more refined subgroup analyses, the incorporation of risk into the decisions faced by respondents (i.e., inclusion of probability for weight loss as an attribute), and the comparison of community-elicited estimates of WTP with those obtained from program participants. A further natural extension of this work might aim to address whether or not introduction of new weight loss programs into a community is worthy of receiving funding within the constraints of a limited health care budget. Incorporation of WTP estimates, such as those derived from this study, into the broader framework of a cost-benefit analysis that would explore the comparison of a variety of competing health care services would assist in answering this important question. 1350


Acknowledgments L.R. held a clinical fellowship awarded by the Alberta Heritage Foundation for Medical Research. She is currently a postdoctoral fellow in the Division of Nutrition and Physical Activity at the Centers for Disease Control and Prevention. C.D. holds the PPP Foundation Chair in Health Economics. M.R. holds an MRC Senior Scholar award in Aberdeen, Scotland. The views expressed in this paper are those of the authors, not the funders. An abstracted version of this paper was presented as a poster at the 2003 Annual Meeting of the American College of Sports Medicine in San Francisco, CA. The authors are grateful for comments received at that meeting. References 1. Mokdad AH, Bowman BA, Ford ES, Vinicor F, Marks JS, Koplan JP. The continuing epidemics of obesity and diabetes in the United States. JAMA. 2001;286:1195–200. 2. Katzmarzyk PT. The Canadian obesity epidemic, 1985-1998. Can Med Assoc J. 2002;166:1039 – 40. 3. World Health Organization. Obesity: Preventing and Managing the Global Epidemic. Technical Report Series 894. Geneva, Switzerland: World Health Organization; 2000. 4. Birmingham CL, Muller JL, Palepu A, Spinelli JJ, Anis AH. The cost of obesity in Canada. Can Med Assoc J. 1999; 160:483– 8. 5. Thompson D, Wolf AM. The medical-care cost burden of obesity. Obes Rev. 2001;2:189 –97. 6. Colditz GA. Economic costs of obesity and inactivity. Med Sci Sports Exerc. 1999;31(Suppl 11):S663–7. 7. Finkelstein EA, Fiebelkorn IC, Wang G. National medical spending attributable to overweight and obesity: how much, and who’s paying? Health Affairs: Web Exclusive. http:// content.healthaffairs.org/cgi/reprint/hlthaff.w3.219v1.pdf (Accessed June 30, 2003). 8. Craig BM, Tseng DS. Cost-effectiveness of gastric bypass for severe obesity. Am J Med. 2002;113:491– 8. 9. Hauri P, Horber FF, Sendi P. Is bariatric surgery worth its costs? Obes Surg. 1999;9:480 –3. 10. O’Meara S, Riemsma R, Shirran L, Mather L, ter Riet G. A rapid and systematic review of the clinical effectiveness and cost-effectiveness of orlistat in the management of obesity. Health Technol Assess. 2001;5:1– 81. 11. O’Meara S, Riemsma R, Shirran L, Mather L, ter Riet G. The clinical effectiveness and cost-effectiveness of sibutramine in the management of obesity: a technology assessment. Health Technol Assess. 2002;6:1–97. 12. NHLBI Expert Panel on the Identification, Evaluation, and Treatment of Overweight and Obesity in Adults. Clinical guidelines on the identification, evaluation, and treatment of overweight and obesity in adults: the evidence report. Obes Res. 1998;6(Suppl 2):51–209S. 13. NHS Centre for Reviews and Dissemination, University of York. Bulletin on the Effectiveness of Health Service Interventions for Decision Makers: The prevention and treatment of obesity. http://www.york.ac.uk/inst/crd/ehc32.pdf (Accessed June 30, 2003).


14. Serdula MK, Kettel Khan L, Dietz WH. Weight loss counseling revisited. JAMA. 2003;289:1747–50. 15. Mattfeldt-Beman MK, Corrigan SA, Stevens VJ, et al. Participants’ evaluation of a weight-loss program. J Am Diet Assoc. 1999;99:66 –71. 16. Stevens VJ, Rossner J, Greenlick M, Stevens N, Frankel HM, Craddick S. Freedom from fat: a contemporary multicomponent weight loss program for the general population of obese adults. J Am Diet Assoc. 1989;89:1254 – 8. 17. Luce D, Tukey J. Simultaneous conjoint measurement: a new type of fundamental measurement. J Math Psychol. 1964;1:1–27. 18. Ryan M, Gerard K. Using discrete choice experiments to value health care programmes: current practice and future research reflections. Appl Health Econ Health Policy. 2003; 2:55– 64. 19. Viney R, Lanscar E, Louviere J. Discrete choice experiments to measure preferences for health and health care: expert review. Pharmacoeconomics Outcomes Res. 2002;2: 319 –26. 20. Ryan M, Watson V, Amaya-Amaya M. Methodological issues in the monetary valuation of benefits in health care: expert review. Pharmacoeconomics Outcomes Res. 2003;3: 717–27. 21. Ryan M, Farrar S. Using conjoint analysis to elicit preferences for health care. BMJ. 2000;320:1530 –3. 22. Ryan M. Using Consumer Preferences in Health Care Decision Making: The Application of Conjoint Analysis. Office of Health Econimics, London: White Crescent Press Ltd.; 1996. 23. Ryan M. Using conjoint analysis to take account of patient preferences and go beyond health outcomes: an application to in vitro fertilisation. Soc Sci Med. 1999;48:535– 46. 24. Bryan S, Buxton M, Sheldon R, Grant A. Magnetic resonance imaging for the investigation of knee injuries: an investigation of preferences. Health Econ. 1998;7:595– 603. 25. Phillips KA, Maddala T, Johnson FR. Measuring preferences for health care interventions using conjoint analysis: an application to HIV testing. Health Serv Res Dec. 2002;37: 1681–705.

26. Ryan M, McIntosh E, Dean T, Old P. Trade-offs between location and waiting times in the provision of health care: the case of elective surgery on the Isle of Wight. J Public Health Med Jun. 2000;22:202–10. 27. Ryan M, Hughes J. Using conjoint analysis to assess women’s preferences for miscarriage management. Health Econ. 1997;6:261–73. 28. Kleinman L, McIntosh E, Ryan M, et al. Willingness to pay for complete symptom relief of gastroesophageal reflux disease. Arch Intern Med. 2002;162:1361– 6. 29. Van der Pol M, Cairns J. Establishing patient preferences for blood transfusion support: an application of conjoint analysis. J Health Serv Res Policy. 1998;3:70 – 6. 30. Narbro K, Sjo¨ stro¨ m L. Willingness to pay for obesity treatment. Int J Technol Assess Health Care. 2000;16:50 –9. 31. Hahn GJ, Shapiro SS. A Catalog for the Design and Analysis of Orthogonal Symmetric and Asymmetric Fractional Factorial Experiments. Scemelady, NY: Canadian Electricity Research and Development Center; 1966, pp. 66 –165. 32. Zwerina K, Huber J, Kuhfled W. A General Method for Constructing Efficient Choice Designs. Durham, NC: Fuqua School of Business; Duke University; 1996. 33. Green KL, Cameron R, Polivy J, et al. Weight dissatisfaction and weight loss attempts among Canadian adults: Canadian Heart Health Surveys Research Group. Can Med Assoc J. 1997;(Suppl 1):S17–25. 34. Serdula MK, Mokdad AH, Williamson DF, Galuska DA, Mendlein JM, Heath GW. Prevalence of attempting weight loss and strategies for controlling weight. JAMA 1999;282: 1353– 8. 35. Nunn RG, Newton KS, Faucher P. 2.5 years follow-up of weight and body mass index values in the Weight Control for Life! program: a descriptive analysis. Addict Behav. 1992;17: 579 – 85. 36. Emery D, Barron F. Axiomatic and numerical conjoint measurement: an evaluation of diagnostic efficacy. Psychometrika. 1979;44:195–210.


1351