Page 1

Incorporating Indirect Costs into a Cost-Benefit Analysis of

Laparoscopic Adjustable Gastric Banding

Eric A. Finkelstein, PhD1,*, Benjamin T. Allaire, MS2, Marco daCosta DiBonaventura, PhD3, Somali M. Burgess, PhD4

1Duke-NUS Graduate Medical School, Singapore;2RTI International, Research Triangle Park, NC, USA;3Health Sciences Practice, Kantar Health, New York,

NY, USA;4Global Health Outcomes Strategy and Research, Allergan, Inc., Irvine, CA, USA

A B S T R A C T

Objectives: The objective of this study was to estimate the time to

breakevenand5-yearnetcostsoflaparoscopicadjustablegastricband-

ing (LAGB) taking both direct and indirect costs and cost savings into

account. Methods: Estimates of direct cost savings from LAGB were

available from the literature. Although longitudinal data on indirect

cost savings were not available, these estimates were generated by

quantifyingtherelationshipbetweenmedicalexpendituresandabsen-

teeism and between medical expenditures and presenteeism (reduced

on-the-job productivity) and combining these elasticity estimates with

estimates of the direct cost savings to generate total savings. These

savings were then combined with the direct and indirect costs of the

procedure to quantify net savings. Results: By including indirect costs,

the time to breakeven was reduced by half a year, from 16 to 14 quar-

ters. After 5 years, net savings in medical expenditures from a gastric

banding procedure were estimated to be $4970 (?$3090). Including ab-

senteeism increased savings to $6180 (?$3550). Savings were further

increased to $10,960 (?$5864) when both absenteeism and presentee-

ism estimates were included. Conclusions: Thisstudypresentedanovel

approach for including absenteeism and presenteeism estimates in cost-

benefitanalyses.Applicationoftheapproachtogastricbandingamongsur-

gery-eligible obese employees revealed that the inclusion of indirect costs

and cost savings improves the business case for the procedure. This ap-

proachcaneasilybeextendedtootherpopulationsandtreatments.

Keywords: bariatric surgery, business case, obesity, return on invest-

ment.

Copyright © 2012, International Society for Pharmacoeconomics and

Outcomes Research (ISPOR). Published by Elsevier Inc.

Introduction

Recentevidencerevealsthatthedirect(medical)andindirect(pro-

ductivity loss) burden of severe obesity, defined as having a body

mass index (BMI) greater than 40 kg/m2, is substantial [1]. Bahr et

al. [2] showed that annual obesity-attributable medical expendi-

turesfortheseverelyobesecouldbeashighas$1270formalesand

$2530 for females. Furthermore, they showed that the indirect

costs resulting from severe obesity, which include increased ab-

senteeism and health-related reductions in productivity while at

work (termed presenteeism), comprised an even larger share of

total obesity-attributable costs. They estimated annual indirect

obesity-attributable costs of $6090 for severely obese male em-

ployees and $6690 for severely obese female employees.

Because of the high costs resulting from severe obesity, effective

obesityinterventionshavethepotentialtogeneratesignificantsavings.

To date, the most effective intervention for severe obesity is bariatric

surgery; the two most common types of bariatric surgery are gastric

bypasssurgeryandgastricbanding.Bothprocedureshavebeenshown

tobecost-effectivewhenfocusingondirectmedicalexpenditures[3–8].

Estimating changes in direct medical expenditures after a

medical/surgical intervention is easily accomplished because of

readily available longitudinal medical claims data. Similar data do

not exist for estimating indirect costs. As a result, nearly all cost-

effectiveness and cost-benefit studies focus solely on direct costs.

Given that a bariatric procedure not only generates short-term

work loss but also has the potential to reduce subsequent absen-

teeism and presenteeism, the largest components of obesity-re-

lated costs, and because employers are ultimately responsible for

making coverage decisions for their employees, a lack of informa-

tiononpotentialindirectcostimplicationsresultingfrombariatric

procedures is a significant limitation.

The objective of this study was to estimate the time to

breakeven and 5-year net costs of laparoscopic adjustable gastric

banding (LAGB) taking both direct medical and indirect absentee-

ism and presenteeism costs and cost savings into account. Al-

thoughlongitudinaldataonindirectcostsavingsarenotavailable,

indirect cost savings were generated by estimating the relation-

ship between medical expenditures and absenteeism and be-

tween medical expenditures and presenteeism and combining

these estimates with estimates of the direct cost savings. Al-

though the analysis focuses on LAGB as a treatment for severe

obesity, this approach can easily be applied to gastric bypass or

extended to other populations and treatments.

Methods

Methodological overview

The estimation strategy occurred in four steps. First, estimates of

quarterly percentage reductions in direct medical cost savings

* Address correspondence to: Eric A. Finkelstein, Duke-NUS Graduate Medical School, 8 College Road, Singapore 169857.

E-mail: eric.finkelstein@duke-nus.edu.sg.

1098-3015/$36.00 – see front matter Copyright © 2012, International Society for Pharmacoeconomics and Outcomes Research (ISPOR).

Published by Elsevier Inc.

doi:10.1016/j.jval.2011.12.004

V A L U E I N H E A L T H 1 5 ( 2 0 1 2 ) 2 9 9 – 3 0 4

Available online at www.sciencedirect.com

journal homepage: www.elsevier.com/locate/jval

Page 2

post-LAGB (termed ?i, where i denotes the quarter postprocedure)

were derived from published literature [9]. Second, an elasticity

was calculated that quantifies the percentage change in absentee-

ism for a given percentage change in medical expenditures

(termed ?). Multiplying each ?itimes ? allows for estimating quar-

terly percentage reductions in absenteeism postbanding. Third,

because no data set exists that allows for directly estimating the

percentage change in presenteeism for a given percentage change

in medical expenditures, this estimate was calculated indirectly

by quantifying the percentage change in presenteeism for a given

percentage change in absenteeism, termed ?. This estimate was

then multiplied by ? and then by each ?ito estimate quarterly

percentage savings in presenteeism postprocedure. Fourth, all

savings were monetized and then combined with the direct and

indirect costs of the procedure to quantify net costs. The data and

a more detailed estimation approach are described below.

Data

Themedicalexpenditure/absenteeismelasticity(?)wasestimated

by using the publicly available Medical Expenditure Panel Survey

(MEPS)—a nationally representative survey of the civilian nonin-

stitutionalized population that quantifies an individual’s total an-

nual medical spending by type of service and source of payment.

This includes all expenditures for office-based visits, hospital out-

patient visits, emergency room visits, hospital inpatient stays,

home health care, dental care, vision aids, other medical equip-

mentandservices(e.g.,orthopedicitems,medicalequipment,dis-

posable supplies), andprescription medicines. The survey also in-

cludes the following question in each interview round that allows

for quantifying annual work loss due to illness or injury: “How

many days did [respondent] miss a half day or more of work due to

health problems?”

Other questions capture employment status, self-reported

weight and height, and sociodemographic characteristics of re-

spondents. The MEPS sample was limited to full-time, nonpreg-

nant employees between the ages of 18 and 64 years (N ? 18,143).

For the primary analysis, the sample was further limited to those

eligible for bariatric surgery, which includes those with a BMI of

more than 40 or between 35 and 40 with a significant comorbidity,

including sleep apnea, cardiovascular disease, osteoarthritis, or

diabetes(n?876),andtothoserespondentswithdatainboth2005

and 2006 (n ? 134 individuals representing 268 observations). To

gauge the sensitivity of the elasticity estimate to sample selection,

supplemental analyses were conducted on the larger samples.

MEPS does not include questions on presenteeism. Both absen-

teeism and presenteeism, however, are included in the proprie-

tary National Health and Wellness Survey (NHWS), although it

doesnotcapturemedicalexpenditures.Therefore,the2008NHWS

cross-sectional data set was used to quantify the absenteeism/

presenteeism elasticity (?). NHWS is a self-administered, Internet-

based questionnaire that focuses on the health status and health-

care attitudes, behaviors, and outcomes of adults aged 18 years or

older. It is fielded to 63,000 members of an Internet-based con-

sumer panel and is designed to be representative of the US adult

population. NHWS captures absenteeism and presenteeism by us-

ing the Work Productivity and Activity Impairment (WPAI) index.

The WPAI index is a validated questionnaire, commonly used

across various occupations and disease areas to assess employee

productivity losses related to health [10]. Absenteeism is mea-

sured by using the following question: “During the past seven

days, how many hours did you miss from work because of your

health problems?” Presenteeism is assessed with the following

question, “During the past seven days, how much did your health

problems affect your productivity while you were working?” Par-

ticipants indicate their level of work impairment via a rating scale

ranging from 0 to 10, with 0 indicating that “health problems had

no effect on my work” and 10 indicating that “health problems

completely prevented me from working.” Each response is as-

sumedtorepresentapercentagereductioninproductiveworkdue

to health problems (e.g., a respondent reporting a value of 3 is

assumed to have a 30% reduction in productive work, whereas a

respondent reporting a 10 is assumed to be completely unproduc-

tive while at work).

NHWS also includes questions similar to those in MEPS that

capture self-reported height and weight, employment status, and

other sociodemographic characteristics. Other than the require-

ment of being in two consecutive years of data (data for only 1 year

wereavailablefortheanalysis),thesamesamplerestrictionswere

applied as for the MEPS data. The primary analysis sample in-

cluded2164individualswhowerefull-timeemployeesandeligible

for LAGB; supplemental analyses were run on the larger sample.

Estimation of indirect costs

MEPS provides annual estimates for medical expenditures and ab-

senteeism. To annualize the WPAI index data, each respondent’s

absenteeism estimate was divided by 8 (to convert it from hours to

days)andmultipliedby50,theestimatednumberofworkweeksin

a year. The presenteeism percentage was multiplied by 250 (the

numberofworkdaysperyear)toestimatethenumberofworkdays

per year that the individual was unproductive at work due to

health problems.

Using the annualized values for medical expenditures, absen-

teeism, and presenteeism, regression modules of the following

form were used to estimate the elasticities:

log?ABSi?? ? log?MEDi??? Zi??i

log?PRESi?? ? log?ABSi??? Zi??i

The log-log specification has the advantage that the coeffi-

cients on log(MEDi) and log(ABSi) are estimated elasticities. In the

MEPS model, with annual absenteeism days as the dependent

variable and annual medical expenditures as the key independent

variable, this coefficient is a direct estimate of ?, the percentage

change in absenteeism for a given percentage change in medical

expenditures. In the NHWS model, with annual presenteeism

days as the dependent variable and annual absenteeism days as

thekeyindependentvariable,thiscoefficientprovidesanestimate

of ?; multiplying this estimate times ? provides an estimate of the

percentage change in presenteeism for a given percentage change

in medical expenditures.

Although the log-log model is a convenient method for esti-

mating elasticities, it is problematic when the logged variables

includealargepercentageofzeros,aswasthecaseforthemedical

expenditures, absenteeism, and presenteeism variables. To en-

sure that individuals with zeros for these variables were not

dropped from the models, in the primary specification 0.1 days

were added to all zero absenteeism and presenteeism days and $1

was added for all individuals who had no medical expenditures

during the year. Supplemental analyses—which included 1) anal-

yses on only those with positive values for these variables, 2) gen-

eralized linear models that did not require log transformations,

and 3) larger adjustment factors—tested the robustness of this

approach. We also explored the effects of not restricting the re-

gressions to the surgery-eligible population.

All regressions also included the following control variables:

age, sex (female indicator), and race/ethinicity (non-Hispanic

white, Hispanic white, black [Hispanic and non-Hispanic], other).

For the MEPS regression, individual fixed-effects models were

used to control for unobservable time-invariant characteristics of

individuals. Because only 1 year of data were available for the

NHWS regression, a traditional ordinary least squares model was

used to estimate ?.

The estimates of ? and ? and estimates of ?iwere combined as

noted in the “Methodological Overview” section to generate quar-

300

V A L U E I N H E A L T H 1 5 ( 2 0 1 2 ) 2 9 9 – 3 0 4

Page 3

terly percentage savings in absenteeism and presenteeism. To

convert these percentage savings into dollars, these values were

multiplied by the absenteeism and presenteeism costs attribut-

able to obesity among the surgery-eligible population. These esti-

mates were generated by using the NHWS data following the iden-

tical approach presented in Finkelstein et al. [11], with the only

difference being that the obesity category variable was redefined

to include only those individuals who were obese and eligible for

bariatric surgery (as defined above).

The quarterly monetized savings in absenteeism and presen-

teeism postbanding were then combined with the direct medical

cost savings to estimate total quarterly savings resulting from an

LAGB procedure. These savings were combined with the direct

and indirect costs of the procedure to estimate time to breakeven

and net costs at the end of 5 years. Direct costs for an LAGB pro-

cedure were taken directly from Finkelstein et al. [12]. Work-loss

estimates were taken from Fisher [13] and monetized by using the

approach presented in Finkelstein et al. [11].

Fisher reports that the average number of days prior to return-

ing to normal activity was 7.2 and that 15.8 days were required for

a full recovery. These were converted into 5.2 days of absenteeism

(7.2 minus 2 weekend days) and 3.3 days of presenteeism. The

presenteeism estimate was calculated by taking 15.8 days and

subtracting out 7.2 days of absenteeism, two additional weekend

days, and assuming that patients are working at 50% productivity

for the 6.6 days prior to full recovery.

Confidence intervals for the net cost estimates were generated

by combining each of 1000 bootstrapped iterations from the direct

cost estimates with a random draw from the preferred model

specification from the absenteeism and presenteeism estimates

assuming a bivariate normal distribution with means equal to the

estimated elasticities, variances equal to the square of the esti-

mated standard errors from the corresponding regression model,

and covariance based on the estimated standard errors and an

assumed correlation of r ? 0.5. We also conducted additional sen-

sitivity analyses for the time to breakeven by using the extreme

valuesof?and?generatedfromthealternativeregressionmodels

and using the primary estimates but assuming the two elasticity

estimates are 1) uncorrelated or 2) perfectly correlated (r ? 1.0).

Results

Table 1 presents summary statistics for the MEPS and NHWS sam-

ples used in the primary analysis. The two samples have similar

age and gender profiles. Likely because it is an Internet panel, the

NHWSsampleincludesmorewhitesandfewerblacksandHispan-

ics. The average BMI among the surgery-eligible sample is also

slightly larger in MEPS than in NHWS, 44.1 (standard error [SE] ?

0.4) versus 42.9 (SE ? 0.1). MEPS respondents who are eligible for

bariatric surgery report $4050 (SE ? 790) in annual medical expen-

ditures. They also report, on an annualized basis, an average of 7.7

days (SE ? 1.9) where at least a half-day or more of work was

missed because of health problems. However, 12% and 39% of the

observations included zeros for medical expenditures and work

loss, respectively.

Annualized estimates of absenteeism among the surgery-eligi-

ble NHWS sample revealed 14.7 days (SE ? 1.0) of work loss on

average, with 37% of individuals missing zero days. This estimate

is more than double the MEPS estimate. This difference likely re-

sults from differences in how the question was asked. NHWS cap-

turesworklossoflessthanhalfaday,whereasMEPSdoesnot.The

average number of presenteeism days for a surgery-eligible indi-

vidual in NHWS was 58.0 (SE ? 1.4), with 55% reporting zero pre-

senteeism days. Using the approach presented in Finkelstein et al.

[11], when limited to obesity-attributable work loss among the

surgery-eligible population, the annual obesity-attributable ab-

senteeism and presenteeism estimates are reduced to 7.0 days

and 21.0 days, respectively. When monetized, this equates to $620

per quarter for absenteeism and $1870 per quarter for presentee-

ism.

Column 1 of Table 2, reprinted from Finkelstein et al. [12], pres-

ents average quarterly savings post-LAGB. Dividing these savings by

costs for the control group in the corresponding quarter generated

the estimated ?is. The average percentage savings (relative to the

control group) post–gastric banding are 38%, with estimates ranging

from 27% (quarter 3) to 63% (quarters 15 through 20 pooled).

Table 3 presents estimates ?, ?, and ? times ?. The estimate of ?,

which represents the percentage change in absenteeism for a given

percentage change in medical expenditures, from the primary spec-

ification is 0.31, meaning that, for example, a 10% decrease in medi-

calexpendituresinagivenquarterwouldgeneratea3.1%decreasein

quarterly absenteeism costs. Various model specifications and sam-

ples lead to estimates of ? between 0.24 and 0.57.

The estimate of ?, the percentage change in presenteeism for a

given percentage change in absenteeism, from the primary spec-

ification was 0.49, with estimates ranging from 0.12 to 0.58. Multi-

plying ? times ? provides an estimate of the percentage change in

presenteeism for a given percentage change in medical expendi-

tures. Using the primary estimates of ? times ? yields an estimate

of 0.15 (0.31 ? 0.49), revealing that, following the example above, a

10% reduction in quarterly medical expenditures would yield a

1.5% reduction in quarterly presenteeism costs.

Finkelstein et al. [12] reported that the direct medical cost of a

gastricbandingprocedureis$20,030.Monetizingtheindirectcosts

reported in Fisher [13] and applying the assumptions presented

above to split the costs between absenteeism and presenteeism

Table 1 – Summary statistics for individuals eligible for

LAGB in the Medical Expenditure Panel Survey (MEPS)

and the National Health and Wellness Survey (NHWS).

MEPS (n ? 134) NHWS (n ? 2164)

Age (y), mean (SE)

Male (%)

Race (%)

White [reference]

Black

Hispanic

Asian

Other

Body mass index (kg/m2),

mean (SE)

Percentage with zero

medical

expenditures

Absenteeism days,* mean

(SE)

Percentage with zero

absenteeism days

Presenteeism days, mean

(SE)

Percentage with zero

presenteeism days

43.2 (1.1)

49.3

44.7 (0.2)

49.1

61.5

20.1

14.5

0.0

3.9

67.8

16.4

9.9

1.0

4.9

44.1 (0.4) 42.9 (0.1)

12 N/A

7.7 (1.9)14.7 (1.0)

39 37

N/A58.0 (1.4)

N/A55.0

LAGB, laparoscopic adjustable gastric banding; N/A, not applicable/

available; SE, standard error.

* Absenteeism is measured differently in MEPS and NHWS. MEPS

includes the following question in each interview round that al-

lows for quantifying annual work loss due to illness or injury:

“How many days did [respondent] miss a half day or more of work

due to health problems?” Absenteeism is measured in NHWS us-

ing the following question: “During the past seven days, how

many hours did you miss from work because of your health prob-

lems?” These estimates were then annualized.

301

V A L U E I N H E A L T H 1 5 ( 2 0 1 2 ) 2 9 9 – 3 0 4

Page 4

generated indirect cost estimates of $1900 and $1280, respectively,

suggesting that the total cost associated with a gastric banding

procedure is $23,210. Table 4 combines these cost estimates with

estimates of the ?is, ?, and ? and the obesity-attributable costs in

the absence of the procedure to generate net savings in each quar-

ter resulting from gastric banding.

Column 1 of Table 4, reproduced from Finkelstein et al. [12],

presents the net costs of LAGB when focusing solely on direct

medical expenditures. These data reveal that the costs of the pro-

cedure are recovered in 16 quarters (4 years) when focusing solely

on direct medical expenditures. Column 2 provides results for

medical and absenteeism costs combined. When focusing on

these cost drivers only, the time to breakeven remains at 16 quar-

ters; reductions in absenteeism over this time period are exactly

offset by the number of days missed required to obtain the proce-

dure. Column 3 presents results for medical, absenteeism, and

presenteeism costs combined. Using all three cost categories, the

time to breakeven is reduced by half a year, from 16 to 14 quarters.

Best-case estimates of ? and ? reduce the time to breakeven to 13

quarters. Using the Fisher et al. estimates of work loss resulting

from the procedure, worst-case estimates suggest that the time to

breakeven remains at 16 quarters.

Although inclusion of absenteeism and presenteeism costs

has only a modest effect on time to breakeven, beyond the

breakeven period estimated savings are much larger when in-

direct costs are considered. Focusing on 5-year savings, Finkel-

stein et al. [12] reported net savings in medical expenditures

from a gastric banding procedure to be $4970 (?$3090). Includ-

ing absenteeism costs increases net savings to $6180 (?$3550).

Savings are further increased to $10,960 (?$5864) when all three

cost categories are included.

Discussion

As noted in the Introduction, indirect costs are the single largest

driverofthecostsofobesity,yet,becauseofdatalimitations,these

costs are typically omitted from cost-effectiveness and cost-ben-

efit analyses of obesity interventions. Given that employers bear a

large share of the indirect costs of obesity and are ultimately re-

Table 2 – Quarterly savings in medical expenditures post-LAGB.

Quarter relative to

band placement

Quarterly reduction in medical

expenditures* (95% CI)†

Percentage reduction in medical expenditures relative to

control group (?i)‡(95% CI)

2

3

4

5

6

7

8

9

?590 (?910 to ?270)

?380 (?690 to ?80)

?670 (?1000 to ?340)

?1210 (?2160 to ?260)

?1510 (?2050 to ?970)

?810 (?1390 to ?230)

?880 (?1550 to ?210)

?1160 (?1870 to ?450)

?2170 (?3550 to ?790)

?3090 (?4560 to ?1620)

?1590 (?2790 to ?390)

?1780 (?3510 to ?50)

?2820 (?5640 to 0)

?1140 (?1510 to ?770)

36 (30–42)

27 (20–34)

31 (24–38)

35 (19–51)

37 (27–47)

23 (10–36)

24 (9–39)

28 (12–44)

43 (22–64)

50 (33–67)

42 (21–63)

37 (7–67)

53 (20–86)

63 (36–90)

10

11

12

13

14

15–19 (pooled)

CI, confidence interval; LAGB, laparoscopic adjustable gastric banding.

* LAGB procedure occurs on day 1 of quarter 1.

†Confidence intervals based on 1000 bootstrapped iterations of the direct cost data.

‡Percentage reduction in medical expenditures relative to control group is generated from results presented in Hammond [9].

Table 3 – Estimates of the elasticity between medical expenditures and absenteeism from the Medical Expenditure

Panel Survey (MEPS) and the elasticity between absenteeism and presenteeism from the National Health and Wellness

Survey (NHWS).

Estimate Percentage change in

absenteeism for a

given percentage

change in medical

expenditures (?)

Percentage change in

presenteeism for a

given percentage

change in

absenteeism (?)

Percentage change in

presenteeism for a

given percentage

change in medical

expenditures (? ? ?)

Preferred model specification (95% CI)

Range based on alternate model specifications

0.31 (0.22–0.39)*

[0.24–0.41]†

0.49 (0.43–0.55)*

[0.12–0.58]‡

0.15 (0.005–0.33)*

[0.03–0.33]

CI, confidence interval; OLS, ordinary least squares.

* Estimate is based on 1000 simulations drawn from the sample mean and variance of ? and ? assuming a bivariate normal distribution with a

correlation of 0.5 between ? and ?.

†The lower bound estimate is from the fixed effects full sample regression after adjusting zero values as in the primary specification. The upper

bound estimate is from the cross-sectional OLS model after similar adjustment for zero values.

‡The lower bound estimate is from the OLS model on the surgery-eligible sample without adjusting zero values. The upper bound estimate is

from the OLS model on the full sample after adjusting zero values as in the primary specification.

302

V A L U E I N H E A L T H 1 5 ( 2 0 1 2 ) 2 9 9 – 3 0 4

Page 5

sponsible for making coverage decisions for their employees, this

omission represents a significant limitation.

This analysis presented a novel strategy for estimating indirect

cost savings in the absence of longitudinal data pre- and postint-

ervention. The approach relied on estimating the relationship be-

tween changes in medical expenditures and corresponding

changes in absenteeism and presenteeism. Implementation is-

sues aside, the logic behind the approach is straightforward. In-

terventions that improve health should simultaneously reduce

medical expenditures, which can be directly assessed via longitu-

dinal claims data, and generate indirect cost savings through re-

ductions in absenteeism and presenteeism, which can be indi-

rectly quantified by estimating the relationship between changes

in medical expenditures and changes in absenteeism and presen-

teeism.

Application of this approach to estimate the net costs of gastric

banding revealed a high correlation between medical expendi-

tures, absenteeism, and presenteeism among the surgery-eligible

obesepopulation.Finkelsteinetal.[12]estimatedanaveragequar-

terly savings postbanding of 38% ($1400). Based on this value and

the elasticity estimates presented in the preceding section, these

results reveal average quarterly indirect cost savings of 12.2%

($390) and 6.0% ($70) for absenteeism and presenteeism, respec-

tively. When indirect costs are included, the estimated time to

breakeven is reduced from 4 to 3.5 years and the potential savings

to employers beyond this period are greatly increased. These re-

sultsshowa222%increaseinthe5-yearsavingsalone,from$4,950

to $10,960, highlighting the importance of incorporating indirect

costs into the analysis.

As noted in the direct cost manuscript upon which these esti-

mates are based, much of the savings in direct costs were gener-

atedthroughloweruseofinpatientservicesand,toalesserextent,

lower payments for prescription drugs partly as a result of reduc-

tions in the use of diabetes medications [12]. Although the rela-

tionship between medical expenditures and absenteeism and

presenteeism was estimated by using cross-sectional data,

lower use of inpatient services would be expected to lead to less

work loss, suggesting that the estimated relationships are in-

ternally consistent. The net savings are also consistent with the

medical literature. A recent review of several prior meta-anal-

yses of LAGB reports that all studies reporting on comorbidities

showed significant resolution or improvement of type 2 diabe-

tes mellitus, hypertension, and dyslipidemia. Sleep apnea was

also significantly improved [14].

Although this study has much strength, there are some limita-

tions. One limitation is that the results hinge on obtaining unbi-

ased estimates of quarterly savings in direct medical expendi-

tures. These estimates are based on data obtained from the

MarketScan®Commercial Claims and Encounters Database be-

tween January 1, 2003, and March 31, 2008 [15]. The database in-

cluded full claims from approximately 100 large payers represent-

ing millions of covered lives. Moreover, a supplemental database

allowed for an identification of a subset of individuals who had a

self-reported BMI of more than 35 kg/m2in a health risk assess-

ment. This supplement was used to identify a sample of surgery-

eligible patients who, after propensity score matching, were used

as a control group. The longitudinal nature of the data and the

ability to merge BMI data make this one of the few data sets avail-

able for conducting this type of analysis. The results, however,

may not generalize beyond members of these health plans. More-

over, any biases in the direct cost estimates will be exacerbated

when indirect costs are included because the indirect cost savings

are a function of the savings in direct costs. In addition, both ab-

senteeism and presenteeism were based on self-report; objective

measures of these values would be preferable.

It would also be preferable to estimate the relationship be-

tween medical expenditures, absenteeism, and presenteeism by

using longitudinal data; however, only 2 years of MEPS data and 1

year of NHWS data were available. Moreover, because the two

elasticity estimates ? and ? were estimated from separate data

sets, we could not directly estimate their covariance. In reality, the

elasticities are likely to be positively correlated because medical

expenditures,absenteeism,andpresenteeismarealldrivenbythe

underlying health of the individual. To account for this, we as-

sumed a correlation of 0.5 between ? and ? when estimating the

standard errors for the 5-year net savings estimates. This assump-

tion does not affect the point estimates, but it does affect the

standard errors. If the estimates of ? and ? were perfectly corre-

Table 4 – Net costs and time to breakeven post-LAGB.

Quarter Medical expenditures

only (95% CI)

Medical expenditures ?

absenteeism

(95% CI)*

Medical expenditures ?

absenteeism ?

presenteeism (95% CI)*

?1

1

2

3

4

5

6

7

8

9

10

11

12

13

14

15

16

$1030 (750–1310)

$20,030 (19,750–20,310)

$19,440 (18,940–19,940)

$19,060 (18,400–19,720)

$18,390 (17,600–19,180)

$17,180 (16,220–18,140)

$15,670 (14,460–16,880)

$14,860 (13,390–16,330)

$13,980 (12,250–15,710)

$12,820 (10,760–14,880)

$10,650 (7860–13,440)

$7560 (3800–11,320)

$5970 (1640–10,300)

$4190 (?790 to 9170)

$1370 (?4720 to 7460)

$230 (?5880 to 6340)

$910 (?7030 to 5210)

$1080 (770–1390)

$21,930 (21,310–22,550)

$21,200 (20,410–21,990)

$20,710 (19,770–21,650)

$19,920 (18,820–21,020)

$18,570 (17,394–19,750)

$16,910 (15,460–18,360)

$16,010 (14,310–17,710)

$15,040 (13,050–17,030)

$13,770 (11,480–16,070)

$11,440 (8460–14,420)

$8160 (4270–12,050)

$6420 (2020–10,820)

$4500 (?560 to 9560)

$1480 (?4840 to 7800)

$110 (?6290 to 6510)

$1250 (?7769 to 5269)

$1190 (800–1580)

$23,210 (22,550–23,870)

$22,200 (21,330–23,070)

$21,500 (20,360–22,640)

$20,470 (19,010–21,930)

$18,850 (17,000–20,700)

$16,920 (14,580–19,260)

$15,850 (13,140–18,560)

$14,690 (11,550–17,830)

$13,220 (7860–18,580)

$10,570 (6100–15,040)

$6930 (1360–12,500)

$4880 (?1380 to 11,140)

$2690 (?4370 to 9750)

$700 (?9110 to 7710)

$2510 (?11,370 to 6350)

$4310 (?13,680 to 5060)

* Confidence intervals were generated by combining each of 1000 bootstrapped iterations from the direct cost estimates with a random draw

from the preferred model specification from the absenteeism and presenteeism estimates assuming a bivariate normal distribution with

means equal to the estimated elasticities, variances equal to the square of the estimated standard errors from the corresponding regression

model, and covariance based on the estimated standard errors and a correlation equal to 0.5.

303

V A L U E I N H E A L T H 1 5 ( 2 0 1 2 ) 2 9 9 – 3 0 4