I just read the commentary by McMahon et al.

 I very much support their arguments and would like to share a 30-year-old anecdote. 

In the early 1970s, as a graduate student at Carnegie-Mellon University, I was involved in a project that applied cost-effectiveness analysis to medical screening tests in use at that time, one of which was Pap smears. The faculty concluded that cervical cancer screening was not cost effective, based on a mathematical model that used the low overall incidence of cervical cancer, among many other inputs. I and another student independently re-did the analysis, separating the data by ethnicity and age. We found that screening guidelines should vary by both the woman's ethnicity and age. 

Thirty years on, this concept of targeting screening or treatment to those for whom it would most benefit and not providing it to those for whom it would least benefit is still not accepted. Some may say this is rationing–if so, it is rationing based on clinical and epidemiologic information. In a society with limited resources for government programs such as Medicare and Medicaid, and with 48 million people without health insurance coverage, rationing based on clinical and epidemiologic information is preferable to rationing based on economic criteria. 

Articles

LETTERS TO THE EDITORS

Correspondence regarding, "Univariate solutions in a multivariate world: Can we afford to practice as in "the good old days"?

(1) To determine if the Health Plan Employer Dataand Information Set (HEDIS) asthma inclusion criteria consistentlyidentify persistent asthma on a year-to-year basis and (2) to explorewhether variation in the number of years of qualification is associatedwith medication and resource utilization outcomes.

We identified 132 414 patients in a large healthcareprogram who were included in 1 or more HEDIS persistent asthmacohorts between 1999 and 2002 and who had continuous insuranceand pharmacy benefit coverage for the entire 4-year observationperiod. Medication, emergency department, and hospital usein 2002 was identified using electronic claims and pharmacy information.

Overall, 47.9% of the patients were identified as havingpersistent asthma in only 1 of 4 years, 40.8% had at least 2 consecutiveyears, and 28.2% had at least 3 consecutive years. Inbivariate and multivariate analyses, more consecutive years ofHEDIS persistent asthma qualification significantly increased thelikelihood of frequent short-acting &#946;-agonist use, inhaled anti-inflammatorycorticosteroid use, at least 1 emergency departmentvisit, and at least 1 hospitalization. The strongest relationship wasfor 3 or more consecutive years of HEDIS qualification.

A significant portion of the HEDIS persistent asthmacohort does not qualify on a year-to-year basis, suggesting thatthe current 1-year qualification period or the underlying administrativecase definition for persistent asthma may be suboptimal.Further clinical validation studies are needed to determine theoptimal criteria for a more useful HEDIS persistent asthma casedefinition.

Asthma is a chronic inflammatory lung disease.

There is no gold standard test for the diagnosis ofasthma. It is typically based on clinical symptomsand is ideally confirmed by the demonstration ofreversible airway obstruction or bronchial hyperreactivity.

Clinical audits of asthma diagnoses have consistentlydemonstrated significant underdiagnosis andoverdiagnosis.

The ability of administrative data systemsof most health plans to identify the diagnosis ofasthma is still unknown.

The current Health Plan Employer Data andInformation Set (HEDIS) measure for asthma uses administrativedata collected during 1 year to identifypatients with presumed persistent asthma and evaluatescontroller therapy during the next year. It has beenreported that this asthma case identification usingadministrative data in children cannot distinguish persistentasthma from transient wheezing syndromes.

Itis also difficult to determine asthma severity usingadministrative data.

One of the historical limitations of administrativedata from a single treatment center or healthcare networkis migration of patients into and out of the healthcareplan.

Kaiser Permanente Health Care Programcares for a large group of patients year after year andcan provide multiyear data to evaluate asthma prevalenceand persistence. Vollmer and coworkers,

in theKaiser Permanente Northwest Region, recently publishedan article on asthma prevalence using multipleclinical information systems. They found that asthmacase identification based on 1 year of pharmacy datawould capture only 61% of the prevalent asthma casesand that the electronic health record alone would identifyonly 66% of the total asthma cases.

Despite the clinical limitations of administrative datato define persistent asthma, such data are available tomost health plans and are potentially useful from a qualityof care perspective. Identifying poor performance onthe HEDIS measure for asthma can be used to targetquality improvement and outreach efforts.

Because few reliable and valid assessment tools existto diagnose persistent asthma, further efforts are neededto improve administrative methods to define persistentasthma. A limitation of the current HEDIS asthmainclusion criteria is the inclusion of a significant portionof patients with intermittent asthma, based on a 2-yearanalysis.

As a result, it will be important to developmore robust administrative measures so that chronictreatment can be directed to those who will benefit fromit and so that health plans that provide the best care arerewarded.

With this background in mind, the present study hasthe following 2 general objectives: (1) to evaluate theoccurrence of persistent asthma on a year-to-year basisas defined by the current HEDIS measure for asthmaand (2) to explore whether alteration of the administrativedefinition of persistent asthma results in the identificationof a population that uses more asthma-relatedhealthcare, including medication use and resource utilization,which are likely markers of persistent asthma.

We identified 132 414 patients with persistent asthmaaged 5 to 56 years who qualified in at least 1 of 4HEDIS persistent asthma cohorts identified from 1999through 2002. The sample was drawn from patientsreceiving care from Kaiser Permanente Health CareProgram, a large vertically integrated staff-modelhealthcare system. Computerized data from patientswith persistent asthma were drawn from the following4 geographic regions: northern California, southernCalifornia, Colorado, and the Pacific Northwest (northwestOregon and southwest Washington).

Patients were selected based on 3 inclusion criteria.First, patients had to meet 1 or more of the followingcurrent HEDIS criteria for persistent asthma caseidentification during at least 1 of the years observed:(1) at least 4 asthma medication-dispensing events,(2) at least 2 asthma medication-dispensing eventsand 4 asthma-related outpatient visits, (3) at least 1asthma-related hospitalization, or (4) at least 1 asthma-related emergency department (ED) visit. Second,patients were required to maintain continuous insuranceand pharmacy benefit coverage during the entirestudy period (January 1, 1999, through December 31,2002). Third, patients had to stay within the HEDIS-eligibleage range of 5 to 56 years for the entire durationof the study. If patients aged out during the 4-yearobservation period, they were excluded from the studypopulation.

Primary Independent Variable of the Number of

The number of years of HEDIS qualification was theprimary independent variable. This variable was categorizedinto the following 6 mutually exclusive groups: 1year only, 2 nonconsecutive years, 2 consecutive years,3 nonconsecutive years, 3 consecutive years, and 4consecutive years.

Medication and resource utilization data were evaluatedduring 2002. Data on the use of weighted canisterequivalents of inhaled anti-inflammatory corticosteroidsand short-acting &#946;-agonists were collected fromelectronic pharmacy databases. The calculation ofweighted canister equivalents was based on drug potencyand doses per canister (available from the author).Resource utilization included 1 or more asthma-relatedED visits and hospitalizations during 2002. Both measureswere collected from electronic databases.

All analyses were conducted using the StatisticalPackage for the Social Sciences (SPSS version 11.1;SPSS Inc, Chicago, Ill). Descriptive statistics were calculatedoverall and stratified by pediatric and adult populationsfor demographic variables, number of years ofHEDIS qualification, medication use, and resource utilization.We then examined the method of entry intothe HEDIS persistent asthma cohort by the number ofyears of HEDIS qualification. Next, we examined thebivariate association of consecutive years of HEDISqualification with medication and resource utilization,using the c

test. Finally, logistic regression models wereconstructed to examine the independent association ofthe number of consecutive years of HEDIS qualificationwith medication and resource utilization, adjusting forage, sex, and geographic region.

A total of 216 025 patients had membership in atleast 1 HEDIS persistent asthma cohort between 1999and 2002. Of those, 146 531 (67.8%) maintained continuousinsurance and pharmacy benefit coverage forthe entire 4-year observation period (data not shown).There were 132 414 patients (90.4%) with continuousinsurance and pharmacy benefit coverage whoremained within the HEDIS-eligible age range for theentire observation period. The total health plan populationwithin the age cutoffs with continuous insuranceand pharmacy benefit coverage was 2 224 804 patients.The HEDIS-defined patients with persistent asthmathus accounted for 6.0% of the health plan population.

Overall, the study population included slightly morefemales (55.0%), with less female representation in thepediatric population and greater female representationin the adult population (

). About half of thestudy population qualified during only 1 of the 4 yearsobserved, slightly more for the pediatric populationand slightly less for the adult population. Overall, 40.8%had at least 2 consecutive years of HEDIS qualification,and 28.2% qualified during 3 or more consecutive years.A higher proportion of adults maintained 2 or moreyears of HEDIS qualification compared with children.

Within the pediatric population, most patientsentered the HEDIS persistent asthma cohort each yearthrough medication use alone (range, 71.3%-79.9%) orthrough medication use in combination with 1 or moreof the other entry criteria (range, 91.9%-94.3%). Aslightly higher proportion of the adult population qualifiedthrough medication use alone (range, 78.7%-85.5%), while a similar proportion qualified bymedication use in combination with other entry criteria(range, 93.9%-96.6%).

The number of consecutive years of HEDIS qualificationwas strongly associated with medication use inbivariate analysis (

) for the pediatric and adult populations. Inboth populations, those with 2 or more consecutiveyears of HEDIS qualification were more likely to use 1or more canister equivalents of inhaled anti-inflammatorycorticosteroids and to overuse short-acting&#946;-agonists (&#8805; 14 canister equivalents). As might be expected,in the multivariate comparison, those with 4consecutive years of HEDIS qualification were themost likely to use inhaled anti-inflammatory corticosteroidsand to overuse short-acting &#946;-agonists, withthose with only 1 year of HEDIS qualification as the referentgroup (Table 3).

The number of years of HEDIS qualification was significantlyassociated with resource utilization, althoughthis association was not as strong as that found for medicationuse (Tables 2 and 3). For the pediatric and adultpopulations, those with 3 or more consecutive years ofHEDIS qualification were significantly more likely tohave asthma-related ED visits and hospitalizations comparedwith those with 1 year of HEDIS qualification(Table 3). Utilization of asthma-related ED visits andhospitalizations did not differ for those with 2 consecutiveyears of qualification compared with those withonly 1 year of qualification.

This study found that the current 1-year HEDIS qualificationperiod does not consistently capture the persistentasthma cohort on a year-to-year basis. Abouthalf of children and adults qualified for the persistentasthma cohort for only 1 of the 4 years observed.However, our results suggest that the manner in whichpatients entered the HEDIS persistent asthma cohortdoes not vary over time, with most children and adultsentering through the use of asthma medications.

Our findings suggest that at least 2 consecutive yearsof HEDIS qualification comprise a better qualificationperiod to identify persistent asthma. For the pediatricand adult populations,patients with2 or more consecutiveyears ofHEDIS qualificationwere significantlymore likelyto use any inhaledanti-inflammatorycorticosteroids andto overuse short-acting&#946;-agonists,which are likelymarkers of persistentasthma. Theoverall relationshipwith the numberof years ofHEDIS qualificationand the occurrenceof adverseevents was consistentwith medication use, but the critical point was 3 ormore consecutive years of qualification, rather than 2consecutive years. In addition, the multivariate resultssuggest that medication use is a more sensitive indicatorto detect differences in the number of consecutive yearsof HEDIS qualification compared with asthma-related EDvisits or hospitalizations.

These findings have implications for the manner inwhich the quality of asthma care is measured in theUnited States. Results indicate that a change from a 1-year qualification period in the HEDIS measure forasthma to a 2-year qualification period may be warranted.Such a change would likely reduce the numberof false-positive results in the current HEDIS denominator,which in turn would affect performance on theHEDIS numerator. With a more accurate denominatorthat reflects true persistent asthma, it is likely that therate of long-term controller use will increase and thatthere will be less variation in the HEDIS measure forasthma overall. The change to a 2-year qualificationperiod would not require additional resources from ahealth plan perspective, because the current windowexamines medication and resource utilization during a2-year period. Further analysis (data not shown) foundthat the HEDIS persistent asthma cohort would bereduced by about 30% when changing from the 1-yearqualification period to a 2-year qualification period.Therefore, it is likely that most health plans would beable to continue participating in the HEDIS measure forasthma and would not be affected by the reduction inthe persistent asthma cohort size.

Furthermore, if healthcare programs are encouragedto provide inhaled anti-inflammatory corticosteroids topatients with chronic asthma symptoms, there needs tobe a corresponding effort to ensure that these patientsactually have a condition that would be helped by regularuse of this therapy. Further random audits of actuallung function and bronchial hyperreactivity of patientsselected by the various current and future proposedadministrative measures for asthma are essential.

This study had several limitations. First, the qualityand reliability of administrative pharmacy and resourceutilization data can vary across geographic regions andfacilities. Furthermore, the use of asthma qualificationmedications may also vary widely, affecting selection ofasthma patients in the HEDIS persistent asthma cohort.However, adjusting for geographic region in the multivariateanalysis in this study did not alter the results(Table 3). Second, no clinical assessments of asthmawere measured (eg, symptoms and forced expiratoryvolume in 1 second). Therefore, it is not entirely clearwhether the number of years of HEDIS qualification isassociated with clinically defined persistent asthma.Third, results can only be definitely applied to insuredpopulations receiving care in a large staff-model healthcareprogram, rather than to other delivery systems.

Future research is needed to validate the findings ofthis study. Specifically, further studies are needed toexamine the association of the number of years ofHEDIS qualification with clinical assessments of persistentasthma. Such work will further clarify the appropriatenumber of years of the qualification period thatassesses persistent asthma. Finally, further work isneeded to examine the effect of modifications to theduration of the HEDIS qualification period on overallperformance on the HEDIS measure for asthma andwhether findings are consistent across delivery systems.

We would like to acknowledge the following Kaiser Permanente HealthCare Program analysts whose dedication and effort made this study possible:Mary-Jean Clements, BSc, Doug Felton, MA, Corey Flohr, MHS, AiliGong, MS, Michael Nash, BS, and Nelli Tadevosyan, MPH.

From the Care Management Institute, Portland, Ore (DM, TS), and Oakland, Calif (JG,JB); Departments of Allergy, San Diego (EM, MS) and Vacaville (GM), Calif; KaiserPermanente Clinical Research Unit, Denver, Colo (JM); and Kaiser Permanente Centerfor Health Research, Portland (DM); Kaiser Permanente Health Care Program.

This work was supported by Kaiser Permanente Health Care Program.

Address correspondence to: David Mosen, PhD, MPH, Kaiser Permanente Centerfor Health Research, 3800 N Interstate Avenue, Portland, OR 97232. E-mail:david.m.mosen@kpchr.org.

Strategy for Asthma Management and Prevention. 

1. Global Initiative for Asthma, National Heart, Lung and Blood Institute. 

Bethesda, Md: Global Initiativefor Asthma, National Heart, Lung and Blood Institute; 2004.

2. Hunter CJ, Brightling CE, Woltmann G, Wardlaw AJ, Pavord AD. 

A comparisonof the validity of different diagnostic tests in adult asthma. 

4. Montnemery P, Hansson L, Lanke J, et al. 

Accuracy of a first diagnosis of asthmain primary health care. 

METHODS

How Well Do the HEDIS Asthma Inclusion Criteria Identify Persistent Asthma?

To use a simulation model to estimate the costs andbenefits of bariatric surgery among full-time employees.

Multivariate regression analysis of nationally representativesurvey data sets to estimate the costs of obesity and asimulation model of the number of years until breakeven underalternate assumptions about the costs and benefits of bariatricsurgery.

We used a 2-part model to estimate medical costs ofobesity based on the 2000-2001 Medical Expenditure PanelSurvey. We estimated work loss with a negative binomial regressionbased on the 2002 National Health Interview Survey. Usingthese results, we simulated the expected number of years requiredfor a bariatric surgery procedure to become cost saving.

Nine percent of the full-time US workforce, or 29% ofthe obese workforce, is eligible for bariatric surgery. Obese workerseligible for bariatric surgery have 5.1 (

< .01) additional daysof work loss and $2230 (in 2004 dollars) (

< .01) higher annualmedical costs than persons of normal weight.

Although the cost implications of bariatric surgeryamong full-time employees depend on many factors, the simulationsreveal that 5 or more years of follow-up are most likely required forthese operations to become cost saving unless the employee bearsa significant fraction of the total costs of the surgery.

The increased prevalence of obesity and obesity-relateddiseases among Americans is well documented.

The overall rate of obesity in adultsgrew to 30.5% in 1999-2000, from 22.9% in 1988-1994and less than 15% in the 1970s.

The prevalence of morbid,or severe, obesity has increased at a much fasterrate than obesity in general.

A substantial body of literaturehas also shown large financial consequencesfrom obesity. For example, obese adults incur 36%greater annual medical expenditures than normal-weightpersons,

and overweight and obesity account for9.1% of total annual medical expenditures.

Increased prevalence of obesity among the workforcemay have several financial consequences for employers.

As increased rates of obesity contribute to risingmedical costs, this will likely exacerbate health insurancecosts. Recent research has shown that 12% of therise in inflation-adjusted per capita medical spendingbetween 1987 and 2001 was attributable to theincreased prevalence of obesity.

Obese employees havealso been shown to be absent from work more oftenthan their nonobese counterparts.

found that obese men were absent 2.7 more days peryear than normal-weight men, and obese women missed5.1 more days per year than normal-weight women.

With attention focused on the costs and prevalenceof obesity, bariatric surgery—in particular, Roux-en-Ygastric bypass and gastric banding—has recentlybecome a common form of treatment for severe obesity.

Employers and insurers have, in turn, beenforced to make or revisit decisions about coverage forbariatric surgery and other obesity treatments.

Although surgical operations have been shown to beeffective in reducing weight and resolving or reducingcomorbidities,

few investigations have addressedpotential cost savings resulting from the surgery.

applied the conventional cost-effectivenessframework to compute the cost per quality-adjustedlife-year for surgery; results ranged from-$4000 per quality-adjusted life-year (net savings)

to$35 600 per quality-adjusted life-year.

looked at reductions in specificcost components resulting from the surgery, but only 1study

reported the number of years before bariatricsurgery results in cost savings, and this study was limitedto medical costs in the Canadian healthcare system.The study reported that Roux-en-Y gastric bypass andvertical banded gastroplasty were cost saving after 3.5years. However, because the Canadian healthcare systemis different from the US healthcare system, resultsmay not be generalizable across borders.

In this study, we use nationally representative datafor the US full-time employed population to quantify theincrease in annual medical costs and work loss associatedwith obesity among the bariatric surgery-eligibleand bariatric surgery-ineligible obese populations. Wethen use these results in a simulation model to estimatethe potential benefits associated with coverage forbariatric surgery under various assumptions. We areaware of no available randomized controlled trial orquasiexperiment with sufficient data to permit adetailed cost-benefit analysis of bariatric surgery. In theabsence of such data, a simulation model is feasible andappropriate for providing base-case estimates.

We used 2 nationally representative data sets of thecivilian noninstitutionalized population to quantifyannual work loss and medical costs attributable to obesity.We used the 2002 National Health Interview Survey(NHIS) to analyze work loss due to illness or injury andthe 2000-2001 Medical Expenditure Panel Survey(MEPS) for medical costs. We applied common sampleselection criteria to both data sets. We restricted thesample to individuals aged 18 to 64 years who reportedworking full time (&#8805; 35 h/wk) for the entire year. We alsoexcluded pregnant women and individuals with missingbody mass index (BMI) data, calculated as weight inkilograms divided by the square of height in meters.

The NHIS is the principal source of information onthe health of the household population of the UnitedStates. Besides self-reported height, weight, and healthconditions, the NHIS includes information on workdaysmissed due to illness or injury and sociodemographiccharacteristics, including race and ethnicity, sex, age,education, family size, employment status, occupation,hours of work per week, and income. For analysis ofwork loss, we began with the 31 044 adults in the 2002NHIS. After applying the sample restrictions, the finaldata set included 12 019 full-time employed adults(6641 men and 5378 women) with sampling weights togenerate nationally representative estimates. Forty-sixpercent of the weighted regression population (41% ofmen and 53% of women) reported missing at least 1 dayof work due to illness or injury.

The MEPS sample is a subset of the NHIS participants.The MEPS provides additional details on healthconditions and annual medical expenditures, and eachindividual's data can be merged with his or her responsesto the NHIS survey. We used the MEPS and pooleddata from 2000-2001 to increase the sample size.Applying the sample selection criteria already detailedreduced the 41 217 (unweighted) adults (17 558 in 2000and 23 659 in 2001) to the final data set of 20 329 full-timeemployed adults (11 849 men and 8480 women).

Surgery-eligible obesity was defined as a BMI of 40 orgreater, or a BMI of 35 to less than 40 with angina, asthma,osteoarthritis, diabetes mellitus, or hypertension.This approximated the guidelines set by a NationalInstitutes of Health panel on bariatric surgery,

although the guidelines included other comorbiditiesthat we were unable to measure in our data. Surgery-ineligibleobesity was defined as a BMI of 30 to less than35, or a BMI of 35 to less than 40 without the comorbiditiesjust listed. All other BMI values were consideredto represent nonobesity. In the analyses, we includeddummy variables for overweight (BMI, 25 to < 30) andunderweight (BMI, < 18).

Because the dependent variable (workdays missed)is discrete, we used a negative binomial regression witha log link to estimate work loss. We conducted separateanalyses for men and women when reporting results bysex and a pooled analysis (controlling for sex) whenreporting overall estimates. We included binary variablesfor BMI categories (underweight, overweight, surgeryineligible, and surgery eligible) to estimate theeffect of excess body weight on annual missed workdays;normal (BMI, 18 to < 25) weight constituted theomitted reference category. The regressions controlledfor other factors expected to affect the number ofmissed workdays, including race and ethnicity, age,education, family size, marital status, income, hourly orsalaried employee, class of occupation, years at currentjob, smoking status, alcohol use, and any functional limitationsnot self-reported as obesity related (eg, difficultywalking, standing, sitting, stooping, reaching, orgrasping). Our method allowed us to estimate theincrease in work loss associated with obesity among thesurgery-eligible and surgery-ineligible populations.

We used the regression estimates to predict annualmissed workdays for each employee. We then generateda second prediction after setting the 2 binary obesityvariables to zero, thus predicting the number of missedworkdays for a hypothetical nonobese employee with allother characteristics equivalent to those of an obeseemployee.

The difference between these 2 predictionsreflects annual obesity-attributable missed workdays.

We used a standard 2-part model to estimate annualmedical costs attributable to obesity for the surgery-eligibleand surgery-ineligible populations.

We conductedseparate analyses for men and women and apooled analysis (controlling for sex) when reportingoverall estimates. We used the same main set of independentvariables as for estimation of work loss, addinga categorical variable for census region and excludingvariables not directly captured in the MEPS, namely,family size, hourly or salaried employee, class of occupation,years at current job, and functional limitations.Predicted costs were generated as for work loss:we extrapolated one set of predictions using theobserved values and another with the binary obesityvariables set to zero so that the difference reflects estimatedobesity-attributable medical costs. Thesemethods followed those used in a recent study

thatquantified national medical expenditures attributableto obesity. Medical costs were adjusted to 2004 dollarsusing the medical care component of the Bureau ofLabor Statistics Consumer Price Index.

All regressions were performed using Stata 8.2(StataCorp LP, College Station, Tex) with samplingweights that allowed for generating nationally representativeestimates of the full-time employed population.We computed standard errors by bootstrapping usingthe bsample procedure in Stata.

Using the work loss and medical cost estimates, wesimulated the cost implications associated with coveragefor bariatric surgery from a self-insured employer'sperspective (ie, the employer bears all medical andwork loss costs). Using this perspective, a bariatricsurgery procedure would achieve breakeven when thesavings resulting from the procedure become equal tothe costs of the procedure. Algebraically, this occurswhen the following equation holds true: 

indicates the number of years since the procedure;

, the mean annual costs if the procedure is notperformed; 

, the fixed costs associated with the procedure;and 

, the mean annual costs after the procedurehas been performed. From a self-insured employer'sperspective, 

include medical costs and workloss costs due to illness or injury (we assume that 100%of these costs accrue to the firm).

Annual work loss and medical costs in the absence ofbariatric surgery, 

, were drawn from the 75th and 90thpercentiles of the predicted distribution of medicalcosts and work loss for the surgery-eligible populationas estimated using the NHIS and MEPS data. We alsoused the estimated mean values for this population in

However, the 75th and 90th percentilesare likely more representative of expectedcosts for those who elect surgery, as they have beenshown to have poorer health and presumably highercosts than the average of all surgery-eligible employees.

For example, the mean BMI was 46.9 among16 944 patients in 105 extracted studies of bariatricsurgery.

In contrast, the surgery-eligible population inthe MEPS sample used for our estimation had a meanBMI of 41.7; the 75th percentile of BMI was 44.1, andthe 90th percentile of BMI was 48.4.

, depend on the employer's share of themedical costs of the procedure and the value of daysmissed from work for employees who undergo the surgery.The employer's share of the medical costs willvary depending on the cost of the procedure and theemployee copayment. For our base case, we showresults using $20 000 for 

, based on a cost of $25 000for the bariatric surgery procedure

and a 20% employeecopayment. We assumed that 15 workdays wererequired for recovery

and multiplied the number ofmissed days by the mean hourly wage (including benefits)among full-time employees to calculate the value ofworkdays lost. The hourly wage, $26.50, was the nationalmean in 2004 for all US full-time employees

andequals an annual salary of $42 140, before including thevalue (30.8%) of benefits.

For our base case, weassume that bariatric surgery reduces obesity-attributablecosts for the surgery-eligible population by 75%. Arecent meta-analysis

reported that patients lost amean of 61% of excess weight after bariatric surgery; asa result, 77% of diabetes mellitus cases were resolved,62% of hypertension cases were resolved, and 86% ofsleep apnea cases were resolved.

In addition to the base case, we simulated a 100%elimination of obesity-attributable costs for the surgery-eligiblepopulation, as well as a 50% elimination. We alsosimulated results in which the employer pays only 50%of the costs of the surgery (perhaps through a large coinsurancerate), as well as results in which those whoundergo the procedure have about 21/2 times the meanwage rate, or a $100 000 annual salary, before benefits.Higher wage rates will reduce the time to breakevenbecause of the higher costs that accrue to the firm as aresult of work loss. Last, we simulated the results if individualsare able to return to work in 5 workdays, asopposed to 15 workdays. All scenarios use a 3% annualdiscount rate to convert future savings to present value.The simulations were performed using Microsoft Excel2002 (Microsoft Corp, Redmond, Wash).

Table 1 gives predicted annual work loss attributableto obesity among full-time employees. The difference inwork loss between surgery-eligible and surgery-ineligibleemployees was striking. Overall, less severe, surgery-ineligible obesity increased work loss by 0.8 days(

< .01) over normal weight, 1.8 days for women (

= .08). Surgery-eligibleobese employees, on the other hand, missed 5.1 moredays of work on average than employees of normalweight, 4.1 more days among men and 5.5 more daysamong women (

< .01 for all). When both sexes wereevaluated together, the difference between surgery-eligibleand surgery-ineligible employees was statisticallysignificant (

= .02). When the sexes were evaluatedseparately, the difference between surgery-eligible andsurgery-ineligible employees was insignificant for men(

A Cost-benefit Simulation Model of Coverage for Bariatric Surgery Among Full-time Employees

To estimate the effect of price, sociodemographicfactors, and health on the demand for bariatric surgery among eligibleindividuals with private health insurance, in order to enablepolicy makers and insurers to make more informed decisions concerningaccess to care for bariatric surgery.

We conducted an Internet-based contingent valuationsurvey of 1802 obese persons eligible for bariatric surgerybut who had not undergone the procedure.

We used multivariate regression analysis to separatelyestimate the likelihood of having gastric bypass and gastric bandingsurgery at different out-of-pocket costs. We combined the resultswith estimates of the privately insured bariatric surgery-eligiblepopulation from the National Health and Nutrition ExaminationSurvey, 1999-2002, to estimate aggregate demand.

Out-of-pocket cost was negatively and highly significantlyrelated to the self-reported likelihood of having surgery.Persons with higher incomes and younger persons also reported asignificantly higher likelihood of surgery. No effect was found forbody mass index or for most comorbidities. We estimate that about150 000 bariatric operations per year would be demanded by thosewith private health insurance at an out-of-pocket cost of $25 000.At $5000, we estimate a demand of 250 000 bariatric operationsper year.

Price is significantly and negatively related to thedemand for bariatric surgery. At an out-of-pocket cost of $5000,about 2.2% of the bariatric surgery-eligible population wouldstrongly consider surgery.

The number of individuals receiving surgical treatmentfor obesity, also referred to as bariatric surgery,has increased dramatically in recent years.The most common form of bariatric surgery is gastricbypass, which restricts the stomach to a small pouch,thereby limiting food intake and inducing partial malabsorption.

Less common but growing in popularity isgastric banding, which restricts the stomach through theuse of an adjustable band.

In 1999, fewer than 30 000bariatric surgery procedures were performed in theUnited States. By 2003, this number climbed to morethan 100 000. The number of bariatric surgery procedureswas expected to approach 200 000 in 2004; however,the decision by several large private insurancecompanies and many self-insured firms to discontinuecoverage led to a slower than expected increase indemand, so the actual number of surgical procedureslikely fell short of this estimate.

Regardless, given thatroughly 9% of US adults with private insurance are eligiblefor bariatric surgery based on National Institutes ofHealth guidelines and the National Health and NutritionExamination Survey [NHANES], 1999-2002, the potentialexists for a much greater demand in the future.

The prospect of a large increase in demand for thesurgery and the potential effect on costs and premiumswere primary motivators for many insurers to discontinuecoverage. For example, even though CIGNAHealthCare representatives acknowledged the benefitsof bariatric surgery for obese patients, the firm nolonger covers the surgery under its standard insuranceplans, citing the high cost and growing demand for surgery.

Specifically excluding bariatric surgery as a coveredbenefit is one of several options available to insurers.However, this option limits the ability of plan membersto get access to a procedure that has been shown to havesubstantial health benefits.

Moreover, these healthbenefits may translate into financial benefits to theinsurer if improvements in health result in reductions infuture medical expenditures. Another option for insurersis to set copayment rates at levels that meet the dualobjectives of allowing greater access to bariatric surgeryfor those who need it, while minimizing the financialrisk to the insurers and members of the health plans.

Determining the appropriate copayment requires,among other factors, an understanding of the expectednumber of bariatric surgery operations demanded at agiven out-of-pocket cost. In other words, insurers wouldlike to know how "popular" the surgery is likely to be ata given copayment. To address this issue, we conductedan Internet-based survey of 1802 bariatric surgery-eligibleindividuals with private health insurance toanswer the following research questions: (1) What is thepredicted demand for gastric bypass and gastric bandingsurgery at different copayments? (2) What sociodemographicfactors affect the self-reported likelihood ofundergoing gastric bypass and gastric banding? (3)What is the predicted aggregate demand for bariatricsurgery at different price levels among adults with privatehealth insurance? This information will allowinsurers and policy makers to make more informeddecisions concerning the effect of various coverageoptions on the demand for bariatric surgery.

The data for this study were derived from a sample of1802 individuals aged 18 to 64 years with private healthinsurance who participated in an Internet-based survey.The data were fielded in 2 parts, with an initial survey of322 respondents in April 2004 and a second survey of1480 respondents in December 2004. Each survey capturedinformation on sociodemographic characteristicsof respondents and included questions concerning willingnessto pay for gastric bypass and gastric bandingsurgery. Respondents to the surveys were randomlyselected and invited to participate from national panelsmaintained by Harris Interactive, Rochester, NY (thefirst survey), and Greenfield Online, Wilton, Conn (thesecond survey). Invitations to participate in both surveyswere extended to selected participants in eachfirm's existing panels, and a brief screener was used toidentify respondents who met general eligibility guidelinesfor bariatric surgery, including a body mass index(BMI) (calculated as weight in kilograms divided by thesquare of height in meters) of 40.0 or higher, or a BMI of35.0 to 39.9 with selected comorbidities (

). Bothsurvey designs included minimum quotas for sex, age,and racial or ethnic groups to ensure that the voluntaryresponses yielded a sample similar to that of the 1999-2002 NHANES. Anyone who reported having had priorbariatric surgery was excluded. Respondents who completedthe surveys were offered nominal compensationby the survey firms. This study was granted an exemptionby the institutional review board of RTIInternational.

The demand for goods and services as a function ofthe market price is often estimated using actual transactionsdata. With respect to bariatric surgery, however,this "revealed preference" approach is problematic forseveral reasons. First, because of health insurance, theactual price faced by many individuals is likely to bemuch less than the market price. Because informationon out-of-pocket costs for those who have had bariatricsurgery operations is largely unavailable, it would not bepossible to estimate demand using existing data.Second, because of government restrictions (eg, gastricbanding was not approved in the United States until2001), limitations by private payers, and a lack ofskilled providers who perform the surgery, price maynot be the primary mechanism rationing demand.Third, changing technology and an increasing relianceon centers of excellence have improved the outcomesand will likely affect the potential demand for bariatricsurgery in the future. Therefore, we chose to estimatethe demand using contingent valuation (CV) techniques.

Contingent valuation is commonly performed tomeasure willingness to pay for nonmarket goods andhas long been used in applications such as environmentaleconomics. More recently, CV has been applied tohealthcare, and it is well suited to the unique features ofthe market for bariatric surgery already mentioned.

Contingent valuation methods typically rely on surveysof potential consumers. For this analysis, potentialconsumers consisted of the panels of surgery-eligiblerespondents who had not previously undergonebariatric surgery. Respondents were asked to assesstheir likelihood of undergoing gastric bypass and gastricbanding surgery at various out-of-pocket costs. Allrespondents were shown a brief "concept board" (availableas an appendix from the author) that describedeach surgical procedure and typical outcomes and risks.When answering each valuation question, respondentshad an option to refer back to the concept board screenat any point. The survey then asked respondents the followingquestions: "Suppose you had to pay [amount]out of your own pocket for gastric bypass surgery. If thiswere the case, how likely would you be to have this procedurein 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 questionswere asked for gastric banding surgery. For each of theamounts, 11 possible likelihood responses were given,comprising 0% to 90% in 10% increments and 99%. Theorder of questioning about gastric bypass or about gastricbanding was randomized for each participant;amounts were proposed beginning with full coverageand then in descending order from $25 000 to $2500.We converted the response to 1 (yes) if the reportedprobability was 80% or greater and to 0 (no) otherwise.We used 80% as a cutoff because prior studies

foundthat this cutoff best predicts actual behavior. To estimatethe effect of price and sociodemographic factorson the self-reported likelihood of bariatric surgery, weran multivariate regressions with these factors as independentvariables and with the binary likelihood generatedfor each of the survey responses as the dependentvariable. Coefficient estimates of regressions on thecontinuous variables of 0% to 99% likelihood of undergoinggastric bypass and gastric banding were comparablein statistical significance and magnitude to theregression output given in 

(results of theseregressions are available from the author).

We ran separate regressions for gastric bypass andgastric banding. The primary independent variable usedin the regressions included the hypothetical prices; toaccount for nonlinearities in demand (an increasing or adecreasing price effect), we also included the pricesquared and the price cubed. Other variables includedindicators for different BMI categories (35.0-39.9, 45.0-49.9, and &#8805;50.0, with 40.0-44.9 as the omitted referencegroup) and separate indicators for 5 significant comorbidities(coronary heart disease [CHD] or congestiveheart failure [CHF], depression, type 2 diabetes mellitus,osteoarthritis or joint pain, and sleep apnea). We alsoincluded a dummy variable indicating if the respondenthad a college degree or higher and categoricalindicators for household income (<$25 000, $25 000-$49 999, $50 000-$74 999, $75 000-$99 999, and &#8805;$100 000). Demographic indicators were age categories(18-24, 25-34, 45-54, and 55-64 years [with 35-44years as the omitted reference group]), sex, and racial orethnic 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 (1for each price level), we estimated the model using apanel data random-effects regression that accounted forclustering of questions within individuals. We also useda fixed-effects regression, but the price coefficients wereclose to those in the random-effects regression. Becauseour dependent variable was binary, we also used logisticregressions. Panel data logistic regressions with randomand fixed effects generated significance and marginaleffects comparable to those summarized in Table 2.However, when generating predictions, we found thatthe logistic regression results did not fit the data as wellas those from the random-effects regression (results areavailable from the author).

Last, using information on the number of privatelyinsured individuals eligible for bariatric surgery, wegenerated an aggregate demand curve that includedthe predicted demand for both gastric bypass and gastricbanding. Because the procedures are substitutes,we could not simply sum the demand for each procedureat a given price level. For example, some respondentsstated that at an out-of-pocket cost of $2500they would be 99% likely to undergo both proceduresin the next 5 years. To account for this impossibility,we assumed that each participant could get at most 1procedure and estimated the aggregate demand forbariatric surgery by using the minimum ofthe self-reported likelihoods for gastricbypass and for gastric banding. Thisapproach provides a conservative estimateof the aggregate demand for bariatric surgeryoverall. Using the minimum, we thenreestimated the regression and used theresults to predict probabilities of bariatricsurgery at a range of out-of-pocket costsbetween $0 (full coverage) and $25 000.We then multiplied the results by the estimatednumber of privately insured personsaged 18 to 64 years eligible for surgery(10.9 million), calculated from the 1999-2002 NHANES, to estimate the aggregatedemand for bariatric surgery during thenext 5 years. We then divided the predictionsby 5 to generate an estimated annualdemand for bariatric surgery operations.

Table 1 gives summary statistics of thesurvey participants. Slightly less than onethird of the sample were aged 35 to 44years, 28.5% were 45 to 54 years, and 19.7%were 25 to 34 years. Seventy-two percent ofrespondents were non-Hispanic whites,11.3% were non-Hispanic blacks, and 13.9%were Hispanics. Overall, 40.1% of respondentswere male. Compared with the generalpopulation, the sample was highlyeducated. Almost 85% of respondentsreported having some schooling beyond ahigh school degree or general equivalencydiploma. This is likely an artifact of theInternet-based survey and the restriction toindividuals with private health insurance,because persons with private health insuranceare more likely to be employed in jobspaying more and requiring higher educationlevels. Consistent with higher educationlevels, household incomes were alsohigher than in the general population.Twenty-four percent of respondentsreported household incomes between$50 000 and $74 999, and 33.7% reportedhousehold incomes of $75 000 or greater. The meanBMI of participants was 42.2, with almost 45% of thesample being eligible for bariatric surgery by having aBMI of 35.0 to 39.9 and 1 or more comorbidities. Ofthe 5 major comorbidities listed in Table 1, the mostcommon was depression (31.7%), followed byosteoarthritis or joint pain (29.4%) and type 2 diabetesmellitus and sleep apnea (21.3% each).

shows the proportions of men and womenwho reported an 80% or greater likelihood of bariatricsurgery within the next 5 years at each of the 7 price levels.At full coverage, 25.8% and 29.9% of men and womenreported being 80% or more likely to undergo gastricbypass and gastric banding, respectively, in the next 5years. At an out-of-pocket cost of $25 000, these self-reportedlikelihoods dropped to 10.9% and 11.4% for gastricbypass and gastric banding, respectively. Men andwomen indicated a slightly lower likelihood of undergoinggastric bypass compared with gastric banding,although the difference between the 2 procedures wassmall and narrowed as the out-of-pocket cost increased.Men reported a slightly greater likelihood of undergoingeither procedure. Consistent with economic theory, theself-reported likelihood of surgerydecreased with increases in theout-of-pocket cost.

Results of the panel data random-effects regressions for gastricbypass and gastric banding aregiven in Table 2. As expected, theprice variables (price and pricesquared) were highly significantfor both procedures (

= .004, respectively, for gastricbypass and 

= .006,respectively, for gastric banding),suggesting that out-of-pocketcost is indeed a major driver ofinterest in surgery. We also foundsignificant income effects forgastric banding. Individualsfrom households with annualincomes of $75 000 to $99 999and of $100 000 and higherreported greater likelihoods ofgastric banding in the next 5years that were 3.4 (

= .03) percentagepoints higher, respectively, relativeto the omitted reference group($50 000-$74 999). Estimates forgastric bypass were similar butwere not significant at conventionallevels.

The results also reveal thatindividuals aged 45 to 64 yearsreported a much lower likelihoodof undergoing gastric bypass surgery,namely, 4 percentage pointsless likely among those aged 45 to54 years (

= .006) and 5 percentagepoints less likely amongthose aged 55 to 64 years (

=.003), relative to the omitted reference group (35-44years). None of the age terms was significant for gastricbanding. We also found no statistically significant differencesby education level. Consistent with the data shownin Figure 1, men reported a 3-percentage-point greaterlikelihood than women of undergoing gastric bypass(

= .005). Non-Hispanic blacks reported 4-percentage-point (

<.001) lower likelihoods of gastric bypass and gastricbanding surgery, respectively. Self-reported likelihoodsfor other racial or ethnic groups were not statisticallydifferent from those of non-Hispanic whites, theomitted reference group.

Of the 5 included comorbidities,we found a significantlylower self-reportedlikelihood of undergoing gastricbanding (6.6%, 

= .002)for persons with CHD or CHF;the 4-percentage-point lowerlikelihood of undergoing gastricbypass associated withhaving CHD or CHF was notsignificant (

= .10). Thislower self-reported likelihoodmay reflect the notion thatindividuals with CHD or CHFare ineligible for surgery.Surprisingly, none of the medicalconditions that bariatricsurgery has been shown toimprove, such as depression,type 2 diabetes mellitus, osteoarthritis or joint pain, andsleep apnea, significantly affected the self-reported likelihoodof having the surgery.

Body mass index was alsonot a significant predictor.

The estimated aggregate demand curve is shown in

We estimate that, at an out-of-pocket cost of$25 000, about 150 000 surgical operations per yearwould be demanded by individuals with private healthinsurance. This price is approximately equal to the fullcost of gastric bypass surgery.

The figure also showsthat decreasing the out-of-pocket cost from $25 000 to$10 000 results in only a small increase in the totalnumber of procedures demanded. At prices lower than$10 000, demand is more responsive to changes inprice. However, the results suggest that, even when fullycovered by insurance, demand among those with privateinsurance would not exceed about 375 000 proceduresper year, about double the current number.

These analyses generated several important findings.At any given out-of-pocket cost, the number of individualswho reported being likely to undergo surgery forgastric banding surgery was greater than that for gastricbypass surgery. To date, the number of gastric bandingprocedures performed in the United States is much lessthan the number of gastric bypass operations performed.

This apparent contradiction may be explainedby different coverage rates between the 2 procedures.Although gastric banding has been available in Europefor several years, it was only approved in the UnitedStates in 2001, and many insurers have opted not tocover it. Therefore, although the actual price of gastricbanding (approximately $17 000) is less than the$25 000 price tag for gastric bypass surgery, the out-of-pocketcost to date is likely to have been much less forgastric bypass. However, as insurers increasingly optnot to cover either surgery, these results suggest thatthe demand for gastric banding surgery is likely toincrease at the expense of the demand for gastric bypasssurgery.

The higher demand for gastric banding may seemsurprising given that the published literature suggeststhat gastric bypass surgery has much greater effectiveness.

However, gastric bypass surgery is generally notreversible. It is possible that individuals are willing totrade off some of the potential benefits for a fullyreversible procedure.

Although we found a strong and statistically significanteffect of price on the self-reported likelihood ofundergoing bariatric surgery, few other factorsappeared to significantly affect the demand. In fact,neither a higher BMI nor an increase in the number ofobesity-related comorbidities appeared to affect theself-reported likelihood of undergoing the surgery.From an insurer's perspective, this finding may be concerning.Insurers that offer coverage for bariatric surgerymight want to target the surgery to that subset ofthe morbidly obese population that has the most togain (ie, those with BMIs in the higher range or thosewith certain comorbidities) to avoid what they mayperceive as adverse selection. This finding suggests thatmore stringent eligibility criteria may be necessary toachieve this outcome.

As noted in the introduction, close to 200 000bariatric surgery procedures were expected to be per-formed in 2004. The aggregate demand curve shown inFigure 2 would predict this quantity at a mean out-of-pocketcost of roughly $7500. Given that many insuredindividuals likely paid significantly less, while otherspaid the full cost, this estimate does not seem out ofstep with current data.

The aggregate demand curve can be used to predictthe effects of the discontinuation of coverage forbariatric surgery. At a mean cost of $20 000 and noinsurance coverage, the demand for bariatric surgery(gastric bypass or gastric banding) would be roughly160 000 procedures per year. Without insurance coverage,the aggregate demand curve suggests that modestdecreases in the price of surgery, perhaps throughchanges in technology or increases in the number ofphysicians who perform the surgery, are unlikely tohave a significant effect on demand. We estimate thatroughly 150 000 operations would be demanded at anout-of-pocket cost of $25 000, whereas this figureclimbs to only 175 000 when the price falls to $10 000.In other words, the demand for surgery is very inelasticat this end of the curve.

In fact, the aggregate demand curve remains inelasticeven at lower prices. For example, at $2500 we estimatethat roughly 300 000 bariatric operations wouldbe demanded. Doubling the price to $5000, whichequates to a 20% copayment at a $25 000 price of gastricbypass surgery, only reduces the total quantitydemanded by 19%. A further increase to $7500 reducesthe total quantity demanded by about 35%.

Given that 1 of 3 adults in the United States isobese, the demand for bariatric surgery may not be aslarge as some have hypothesized. Based on data fromthe 1999-2002 NHANES, we estimate that roughly 11million individuals with private insurance in theUnited States are eligible for bariatric surgery. At anout-of-pocket cost of $5000, about 250 000 bariatricsurgery operations would be demanded each year, orapproximately 2.2% of the eligible population. Aninsurance company with 100 000 adult members anda 9% prevalence of surgery-eligible obesity would likelyprovide coverage for about 200 bariatric surgeryoperations each year. Moreover, a modest copaymenthas the additional benefit of lowering the cost of coveringthe surgery, while simultaneously increasingaccess to those who need it most.

Our analysis has several limitations. Perhaps the primarylimitation is that our sample may not be representativeof the privately insured population eligible forbariatric surgery. As a check, we compared our samplewith a nationally representative sample of privatelyinsured, surgery-eligible individuals (based on as closeto the same definition of eligibility as possible) whoparticipated in the 1999-2002 NHANES (results areavailable from the author). Sixty percent of our samplewere aged 35 to 54 years vs 57.6% of the NHANES sample.Men made up 40.1% of our sample vs 38.0% of theNHANES sample, and both populations were roughly72% non-Hispanic white. The mean BMI in our samplewas 42.2 vs 40.9 in the NHANES sample, and the prevalencesof CHD or CHF and type 2 diabetes mellituswere also nearly identical. However, we found 2 differencesthat may affect our results. A larger number ofparticipants in our sample had more than a high schooleducation (84.1% vs 56.8%), and our sample had a largerpercentage of participants whose household incomewas at least $50 000 per year (57.8% vs 46.4%). Both ofthese features are likely attributable to the online surveysample after survey. Because our results show thatgreater income is positively associated with thedemand for bariatric surgery, it is possible that accessto a more representative sample would have led tosmaller estimates of aggregate demand.

Although CV methods are well established, they arenot without criticism.

One concern is that it is difficultfor people to fully value a procedure before theyexperience it, even when shown the detailed informationabout each procedure that we provided in the conceptboards. Our sample likely includes individualswith varying degrees of prior information about the 2types of surgery, which may differentially affect theirvaluations. Moreover, although the information containedin the concept boards was accurate to the bestof our knowledge, it is possible that the way the informationwas presented may have affected the responses.Another criticism concerns the 80% cutoff to predictwho does or does not get the surgery at a given pricelevel. Additional analyses (available from the author)using cutoffs of 70% and 90% resulted in parallel shiftsof the aggregate demand curve (higher or lower predictedquantities at each price) but did not change theprimary conclusions concerning the relationshipbetween changes in price levels or in sociodemographiccharacteristics on the demand for the surgery.Moreover, the predicted number of surgical proceduresbased on the 80% cutoff appears to be in the range ofthe number of bariatric procedures that were performedin 2004.

Our results are limited to subjects with privatehealth insurance and do not reflect the uninsured ormembers of public insurance programs. Inclusion ofthese populations would suggest that the aggregatedemand for the surgery would be greater than what isreported herein. Our estimates also reflect current populations.If the size of the surgery-eligible populationcontinues to grow at current rates, the aggregatedemand for the surgery will also increase. Changes intechnology that improve the effectiveness of bariatricsurgery might increase demand, whereas changes intechnology for substitute products, such as the introductionof effective weight loss medications, might substantiallyreduce the demand.

Finally, we acknowledgethat clinical guidelines and medical necessity oftendrive coverage decisions and that demand considerationsmay be of secondary importance. Regardless,these estimates provide previously unavailable informationthat should assist insurers and policy makers inappropriate coverage strategies for bariatric surgery.

From RTI International, Research Triangle Park, NC (EAF, DSB); Ethicon Endo-Surgery,Inc, Cincinnati, Ohio (YA); and Copernicus Marketing Consulting, Waltham, Mass (AHT).

This study was sponsored by Ethicon Endo-Surgery, Inc. The data were collected byCopernicus Marketing Consulting for Ethicon Endo-Surgery, Inc. All analyses and datareporting were conducted by researchers at RTI International, who had final right ofapproval of publication of this article.

Address correspondence to: Eric A. Finkelstein, PhD, RTI International, 3040 CornwallisRoad, PO Box 12194, Research Triangle Park, NC 27709-2194. E-mail: finkelse@rti.org.

1. Lehman Center Weight Loss Surgery Expert Panel. 

Commonwealth ofMassachusetts Betsy Lehman Center for Patient Safety and Medical Error ReductionExpert Panel on Weight Loss Surgery: executive report. 

POLICY

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

To estimate responsiveness of prescription demandwithin 9 therapeutic classes to increased cost-sharing comparedwith constant cost-sharing.

Retrospective prescription claims analysis.

Between 1999 and 2001, 3 benefit plans changedfrom a 2-tier to a 3-tier design (cases); 1 plan kept a 2-tier design(controls). Study subjects needed 24 months of continuous coverageand a prescription filled &#8805;3 months before the benefit changefor a nonsteroidal anti-inflammatory agent (NSAID), a cyclooxygenase(COX-2) inhibitor, a selective serotonin reuptake inhibitor(SSRI), a tricyclic antidepressant (TCA), an angiotensin-convertingenzyme (ACE) inhibitor, a calcium-channel blocker (CCB), anangiotensin-receptor blocker (ARB), a statin, or a triptan. Changesin use were compared with the Wilcoxon signed rank test.Elasticity of demand among cases was calculated.

Generally, medication possession ratios decreased forcases and increased for controls between 1999 and 2000. Switchrates increased for cases and decreased for controls for all classesbut CCBs. Switches to lower copayments for ACE inhibitors, statins,and triptans occurred more often for cases. Discontinuation-ratechanges for cases were 2 to 8 times those for controls. Generic-substitutionrates depended on availability and initial generic utilization.Elasticity of demand for drugs was generally low, -0.16 to-0.10, for asymptomatic conditions (ACE inhibitors, ARBs, CCBs,statins), and moderate, -0.60 to -0.24, for symptomatic conditions(COX-2 inhibitors, NSAIDs, triptans, SSRIs).

Use of retail prescription medications within 9specific therapeutic classes decreased as copayment increased.Demand for pharmaceuticals was relatively inelastic with thesecopayment increases.

Because healthcare expenditures are expected tocontinue to rise, health insurers continue to seekways to slow the growth in spending by managingutilization and shifting costs to the consumers whilemaintaining a high level of customer satisfaction andquality outcomes. Although pharmaceutical spendingaccounts for only 7% of total healthcare expenditures,the rate of spending growth for prescription medicationshas outpaced other areas of medical care. In 2001prescription drug expenditures rose 16%, while those forhospital care and physician services rose 8% and 9%,respectively.

Drug utilization review, prior authorization,generic substitution, and closed formularies aresome of the methods used by providers of pharmacybenefits to stem this tide. Most common among employersare incentive-based or multitiered formularies,where consumer cost-sharing increases for products onhigher tiers compared with lower tiers.

In 2000, 80% of health plans with prescription benefitsoffered 3-tier formularies compared with 36% ofplans 2 years earlier.

In 2004 the average copaymentfor generic drugs (tier 1) was $10; preferred brandeddrugs (tier 2), $21; and nonpreferred branded drugs(tier 3) $33.

Joyce et al estimated adding a third tier fornonpreferred brands to a 2-tier plan at a copayment of$30 decreases overall drug spending by 4% among planparticipants with employer-sponsored drug coverage.

Additional studies such as that by Huskamp et al haveshown changing from a 1-tier or 2-tier plan to a 3-tierplan impacts the purchasing decisions of consumersmore substantially with respect to their overall drugconsumption,

but the impact of decreased utilizationon health outcomes has not been evaluated.

Tworecent studies evaluated the impact of increasing prescriptiondrug copayments on uses of other healthcareservices. Motheral and Fairman found that enrolleeswho were moved from a 2-tier to a 3-tier pharmacy benefithad modestly lower prescription utilization with noincrease in physician office visits, emergency room visits,or hospital stays in the 11 months after the benefitchange.

However, Goldman et al found that doublingthe copayments in a 2-tier plan for antidiabetic, antiasthmatic,and antiulcerant agents resulted in a 17%increase in predicted annual emergency departmentvisits and a 10% increase in predicted annual hospitaldays for persons with the respective conditions.

Several studies from the 1980s and early 1990sshowed that demand for prescription drugs were highlyinelastic, with values between -0.33 and -0.10 for smallabsolute changes in price.

As out-of-pocket costs andprice differentials between products on different tierscontinue to rise, consumers may be more sensitive tochanges in prescription drug costs than previouslyreported. This study uses a pre-post analysis of 3 managedcare populations whose pharmacy benefitschanged in mid-2000 from a 2-tier to a 3-tier design,compared with a managed care population that had nochange in a 2-tier benefit design during the same timeperiod. We evaluated changes in utilization patternssuch as medication possession ratio, discontinuationrates, and switches to lower-tier products, as well as theelasticity of demand for medications in 9 major therapeuticclasses.

This study is a retrospective 2-sample cohort analysisof prescription medication utilization among managedcare enrollees. The study population (cases) wasmade up of 3 geographically dispersed managed careplans, each of which adopted a 3-tier pharmacy benefitsystem in the first half of 2000 (

). Immediatelybefore the benefit switch, plans 1, 2, and 3 provided atraditional 2-tier prescription benefit covering genericand formulary brand products. Plans 1 and 3 mandatedthe new benefit for all members, whereas plan 2 adoptedit only for members of a preferred provider organization.Plan 4, the control population (controls), had thesame 2-tiered benefit structure as plan 2, but retainedthe 2-tier formulary during the evaluation period, 1999-2001. The analyses utilized enrollment data and pharmacyclaims for 1999 through 2001.

For inclusion in this analysis, members were firstrequired to have 24 months of continuous enrollment,12 months immediately before their plan benefit change(prechange period) and 12 months after their plan benefitchange (postchange period). If these members hada prescription filled during this time for medicationswithin any of the therapeutic areas of interest, theywere then assigned to 1 or both of 2 analysis groups.First, for each drug class of interest, members with aprescription filled at least 3 months before the end ofthe prechange period were retained for time-seriesanalyses related to drug utilization. Second, the subsetof cases with at least 2 prescriptions filled for a drugclass of interest within the 3 months before the benefitchange was retained for estimation of elasticity ofdemand.

Members were excluded from all analyses if theyobtained any medications through mail-order pharmacies(5%), as there were substantial differences betweencopayment rate and quantity of supply between mailorder and retail prescriptions. Additionally, 20% ofmembers whose copayment for a prescription wasinconsistent with their drug benefit were excluded fromanalysis. This group included those without a copaymentrecorded on their prescription claim or with acopayment amount recorded that did not match thecopayment tier as reported by the health plan. The latterwere excluded to attain a more accurate estimate ofthe impact of the copayment change on drug utilization.

Several criteria were applied when selecting the drugclasses under study. The population of users of medicationswithin the drug class had to be of sufficient size tomake reasonable inferences about the population atlarge. The drug classes also had to be used to treat a varietyof chronic and acute, and symptomatic and asymptomaticconditions. Nine commonly used therapeuticclasses for 5 conditions wereselected for analysis:angiotensin-convertingenzyme (ACE) inhibitors, calcium-channel blockers (CCBs),and angiotensin-receptorblockers (ARBs) for cardiovascularconditions; cyclooxygenase2 (COX-2) inhibitors andnonselective nonsteroidal antiinflammatoryagents (NSAIDs)for pain; selective serotoninreuptake inhibitors (SSRIs)and tricyclic antidepressants(TCAs) for depression; 3-hydroxy-3-methylglutarylcoenzyme A reductaseinhibitors (statins) for lipid lowering; and serotonin 5-HT

October 2005

Correspondence regarding, "Univariate solutions in a multivariate world: Can we afford to practice as in "the good old days"?

How Well Do the HEDIS Asthma Inclusion Criteria Identify Persistent Asthma?

A Cost-benefit Simulation Model of Coverage for Bariatric Surgery Among Full-time Employees

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

Impact of 3-Tier Pharmacy Benefit Design and Increased Consumer Cost-sharing on Drug Utilization

Improving Services for Sex Partners of Chlamydia-infected Patients in an HMO

Assessing Controversial Direct-To-Consumer Advertising for Hereditary Breast Cancer Testing: Reactions from Women and Their Physicians in a Managed Care Organization