The Role of Price, Sociodemographic Factors, and Health in the Demand for Bariatric Surgery

Contingent valuation is commonly performed to measure willingness to pay for nonmarket goods and has long been used in applications such as environmental economics. More recently, CV has been applied to healthcare, and it is well suited to the unique features of the market for bariatric surgery already mentioned.⁷ Contingent valuation methods typically rely on surveys of potential consumers. For this analysis, potential consumers consisted of the panels of surgery-eligible respondents who had not previously undergone bariatric surgery. Respondents were asked to assess their likelihood of undergoing gastric bypass and gastric banding surgery at various out-of-pocket costs. All respondents were shown a brief "concept board" (available as an appendix from the author) that described each surgical procedure and typical outcomes and risks. When answering each valuation question, respondents had an option to refer back to the concept board screen at any point. The survey then asked respondents the following questions: "Suppose you had to pay [amount] out of your own pocket for gastric bypass surgery. If this were the case, how likely would you be to have this procedure in the next 5 years?" Amounts were "full coverage" (which we treated as $0), $2500, $5000, $10 000, $15 000, $20 000, and $25 000. Identical questions were asked for gastric banding surgery. For each of the amounts, 11 possible likelihood responses were given, comprising 0% to 90% in 10% increments and 99%. The order of questioning about gastric bypass or about gastric banding was randomized for each participant; amounts were proposed beginning with full coverage and then in descending order from $25 000 to $2500. We converted the response to 1 (yes) if the reported probability was 80% or greater and to 0 (no) otherwise. We used 80% as a cutoff because prior studies^8,9 found that this cutoff best predicts actual behavior. To estimate the effect of price and sociodemographic factors on the self-reported likelihood of bariatric surgery, we ran multivariate regressions with these factors as independent variables and with the binary likelihood generated for each of the survey responses as the dependent variable. Coefficient estimates of regressions on the continuous variables of 0% to 99% likelihood of undergoing gastric bypass and gastric banding were comparable in statistical significance and magnitude to the regression output given in Table 2 (results of these regressions are available from the author).

We ran separate regressions for gastric bypass and gastric banding. The primary independent variable used in the regressions included the hypothetical prices; to account for nonlinearities in demand (an increasing or a decreasing price effect), we also included the price squared and the price cubed. Other variables included indicators for different BMI categories (35.0-39.9, 45.0- 49.9, and ≥50.0, with 40.0-44.9 as the omitted reference group) and separate indicators for 5 significant comorbidities (coronary heart disease [CHD] or congestive heart failure [CHF], depression, type 2 diabetes mellitus, osteoarthritis or joint pain, and sleep apnea). We also included a dummy variable indicating if the respondent had a college degree or higher and categorical indicators for household income (<$25 000, $25 000-$49 999, $50 000-$74 999, $75 000-$99 999, and ≥$100 000). Demographic indicators were age categories (18-24, 25-34, 45-54, and 55-64 years [with 35-44 years as the omitted reference group]), sex, and racial or ethnic dummy variables (non-Hispanic blacks, Hispanics, and other races or ethnicities [with non-Hispanic whites as the omitted reference group]). Because each respondent generated 7 observations (1 for each price level), we estimated the model using a panel data random-effects regression that accounted for clustering of questions within individuals. We also used a fixed-effects regression, but the price coefficients were close to those in the random-effects regression. Because our dependent variable was binary, we also used logistic regressions. Panel data logistic regressions with random and fixed effects generated significance and marginal effects comparable to those summarized in Table 2. However, when generating predictions, we found that the logistic regression results did not fit the data as well as those from the random-effects regression (results are available from the author).

Last, using information on the number of privately insured individuals eligible for bariatric surgery, we generated an aggregate demand curve that included the predicted demand for both gastric bypass and gastric banding. Because the procedures are substitutes, we could not simply sum the demand for each procedure at a given price level. For example, some respondents stated that at an out-of-pocket cost of $2500 they would be 99% likely to undergo both procedures in the next 5 years. To account for this impossibility, we assumed that each participant could get at most 1 procedure and estimated the aggregate demand for bariatric surgery by using the minimum of the self-reported likelihoods for gastric bypass and for gastric banding. This approach provides a conservative estimate of the aggregate demand for bariatric surgery overall. Using the minimum, we then reestimated the regression and used the results to predict probabilities of bariatric surgery at a range of out-of-pocket costs between $0 (full coverage) and $25 000. We then multiplied the results by the estimated number of privately insured persons aged 18 to 64 years eligible for surgery (10.9 million), calculated from the 1999-2002 NHANES, to estimate the aggregate demand for bariatric surgery during the next 5 years. We then divided the predictions by 5 to generate an estimated annual demand for bariatric surgery operations.

RESULTS