Obesity Surgery, 10, 391-401
Meta-Analysis
Bariatric Surgery for Morbid Obesity
Mark J. Monteforte, MD; Charles M. Turkelson, PhD
ECRI, Plymouth Meeting, Pennsylvania, USA*
Background: Bariatric surgery is a treatment for
severely obese patients. We examined the efficacy of
bariatric surgery, addressing three questions: 1)
“What is the overall weight reduction following
bariatric surgery?” 2) “What complications are asso-
ciated with bariatric surgery?” 3) “What impact does
weight loss have on obesity-related comorbidity?”
Methods: Fixed and random effects meta-analy-
ses were used to determine the amount of weight
reduction following bariatric surgery. The influence
of a variety of co-variates that could affect study
results was examined. Information from evidence-
based sources was used to explore the impact of
weight loss on comorbidities.
Results: Meta-analyses results were affected by loss
to follow-up, and within-study heterogeneity of vari-
ance. Therefore, results were pooled from studies with
complete patient follow-up. Meta-analysis of six stud-
ies reporting weight loss at 1 year and four studies
with mean follow-up of 9 months to 7 years demon-
strated BMI reductions of 16.4 kg/m2 and 13.3 kg/m2,
respectively. Weight reduction following bariatric
surgery may be associated with improvements in risk
factors for cardiac disease including hypertension,
type 2 diabetes and lipid abnormalities, and may
decrease the severity of obstructive sleep apnea.
Conclusion: Bariatric surgery is appropriate for
obese patients (BMI >40 kg/m2 or ³35 kg/m2 with obe-
sity-related comorbidity) in whom non-surgical treat-
ment options were unsuccessful. Additional
research is needed to examine the long-term bene-
fits of weight loss following bariatric surgery, partic-
ularly with respect to obesity-related comorbidities.
*ECRI is a non-profit health services research organization
that provides technical assistance to hospitals, health plans,
and government agencies worldwide.
Reprint requests to: Charles M Turkelson, PhD, Chief Research
Analyst, Health Technology Assessment Group, ECRI, 5200
Butler Pike Plymouth Meeting, PA 19462-1298, USA. Tel: (610)
825-6000; fax: (610) 834-1275; email: eturkelson@ecri.org
© FD-Communications Inc.
Key Words: Anastomosis, Roux-en-Y; gastric bypass;
gastroplasty; meta-analysis; obesity, morbid; bariatric
surgery; weight loss
Introduction
Obesity is a major health concern. Obese individu-
als are at increased risk of mortality and morbidity,
including hypertension, type 2 diabetes, coronary
heart disease, stroke, gallbladder disease
(cholelithiasis and cholecystitis), osteoarthritis,
sleep apnea, respiratory problems, and many types
of cancer (including endometrial, breast, prostate,
and colon). Obesity is also associated with compli-
cations of pregnancy, menstrual irregularities, hir-
sutism, stress incontinence, and psychosocial impair-
ments (e.g., binge eating, body image perceptions,
depression, and social stigmatization). 1 Prevalence
estimates for U.S. adults with a body mass index
(BMI) ³35 kg/m2 and ³40 kg/m2 are 8.0% and 2.8%,
respectively,2 and the relative risk for all-cause mor-
tality is increased at BMI levels ³30 kg/m2.3
Nonsurgical approaches to treatment of clinical-
ly severe (morbid) obesity (BMI >40 kg/m2)
include various combinations of low-calorie or
very low calorie diets, behavioral modification,
exercise, and pharmacologic agents.4 A major
drawback to nonsurgical approaches is their failure
to maintain reduced body weight in most obese
patients. 4
Partly for this reason, bariatric surgery is a treat-
ment option often recommended for severely obese
patients (BMI >40 kg/m2 or BMI ³35 kg/m2 with
obesity-related comorbidity) who have had unsuc-
Obesity Surgery, 10, 2000 391
Monteforte and Turkelson
cessful results with nonsurgical methods for weight capacity as mechanisms for weight reduction.
reduction. The extent to which surgical procedures Roux-en-Y gastric bypass (RYGBP) and distal
are used for the morbidly obese is suggested by (extended) Roux-en-Y gastric bypass (RYGBP-E)
data from the International Bariatric Surgery are examples of combination procedures.
Registry (IBSR). Malabsorptive procedures are those that reduce
In 1997, the IBSR reported data submitted by 50 nutrient absorption, typically by bypassing a part
surgical practices involving 75 surgeons on proce- of the small intestine. Measures of the effective-
dures performed from 1986 through 1995 on 14,641 ness of malabsorptive procedures were not ana-
patients. According to IBSR data, 84.7% of patients lyzed in this article.
received one of four surgical procedures: vertical Bariatric procedures may alternatively be classi-
banded gastroplasty (36.3%), Roux-en-Y gastric fied on an anatomic basis, depending on whether
bypass (29.5%), silastic ring gastroplasty (9.9%), or the operation involves the stomach, small intestine,
distal Roux-en-Y gastric bypass (9.0%). The authors or both. If one were to use such an anatomic clas-
did note, however, that there had been a recent sification system, then restrictive procedures would
increased use of gastric banding operations.5 be nearly synonymous with operations involving
The most obvious reported benefit of bariatric the stomach, malabsorptive procedures with opera-
surgery is weight loss. Published evidence for such tions involving the small intestine, and combina-
weight loss is largely from case series and uncon- tion procedures with operations involving both the
trolled clinical trials. Differences among these tri- stomach and the small intestine. The use of this
als mean that it is not simple to determine the aver- classification system would not greatly alter the
age weight loss that can be expected to result from results of our analysis.
surgery and, therefore, whether the surgery has a There is currently no uniform measure for report-
significant impact on obesity-related co-morbidi- ing weight loss outcomes. Studies of bariatric surgi-
ties. To overcome these difficulties, we used meta- cal procedures have reported weight loss in a num-
analysis to determine the overall amount of weight ber of different ways, including change in absolute
reduction following bariatric surgery. We also weight (pounds and/or kilograms), change in BMI
sought to determine whether the meta-analytically (kg/m2), percentage of excess body weight loss
determined weight loss following bariatric surgery (EBWL), or a change in ideal body weight (IBW).
has an impact on mortality and several obesity- In 1998, the National Institutes of Health (NIH)
related comorbidities, including hypertension, type recommended that BMI be used to classify over-
2 diabetes, lipid abnormalities, and sleep apnea. weight and obesity and to estimate relative risk for
This present analysis is not limited to any single disease compared to normal weight.1 The NIH
surgical procedure, but, rather, it covers those pro- adopted a classification system of overweight and
cedures described as restrictive or as a combination
of restrictive and malabsorptive. In the present arti- Table 1. NIH and ASBS Classification of Obesity
cle, bariatric procedures were divided into groups
NIHa ASBS b
depending on the mechanism by which weight loss BMI Classification BMI Classification
is promoted. Restrictive surgical procedures (kg/m2) (kg/m2)
decrease the capacity of the stomach, thereby lim- <18.5 Underweight
18.5-24.9 Normal <25 Normal
iting the volume of food consumed before the feel- 25.0-29.9 Overweight 25-<27 Overweight
ing of satiety occurs. Gastroplasty, including verti- 27-<30 Mild obesity
cal banded gastroplasty (VBG) and silastic ring 30.0-34.9 Obesity 30-<35 Moderate
(Class I) obesity
vertical gastroplasty (SRVG) and gastric banding 35.0-39.9 Obesity 35-<40 Severe obesity
procedures such as adjustable silicone gastric (Class II)
banding (ASGB) are examples of restrictive proce- ³40 Extreme obesity >40 Morbid obesity
(Class III)
dures. Combination surgical procedures are those >50 Super obesity
that bypass part of the digestive tract, usually with a
Adapted from “Clinical Guidelines on the Identification, Evaluation, and
a decrease in stomach capacity. Such procedures Treatment of Overweight and Obesity in Adults.”1
combine malabsorption and diminished stomach bAdapted from “Guidelines for Reporting Results in Bariatric Surgery”6

392 Obesity Surgery, 10, 2000


obesity based on BMI (Table 1). The basis for this
BMI classification scheme stems from observa-
tional and epidemiological studies that relate BMI
to risk of morbidity and mortality.1
In 1997, the Standards Committee of the
American Society for Bariatric Surgery (ASBS)
adopted a classification of obesity, based on BMI
(Table 1). The NIH and ASBS standards are simi-
lar, the primary exception being that the NIH terms
those individuals with a BMI of 30 kg/m2 or more
as obese, whereas the ASBS standards classify
those individuals with a BMI of 27 to <30 kg/m2 as
mildly obese. For the purposes of this article, the
terms extreme obesity, morbid obesity, and severe-
ly obese are used interchangeably. The different
definitions do not affect the analysis in this article.
All of the studies retrieved for analysis, except for
one, included patients with a mean preoperative
BMI ³40 kg/m2.
Materials and Methods
Literature Searches. To prepare this article, we
conducted a systematic and comprehensive search
for information that involved consulting a number
of databases (Table 2). In addition, other mecha-
nisms were used to retrieve additional relevant infor-
mation, including review of bibliographies/refer-
ence lists from peer-reviewed and gray literature.
(Gray literature includes reports, studies, etc., pro-
duced by local government agencies, private organi-
zations, educational facilities, corporations, etc.,
which do not appear in the peer-reviewed literature.)
Inclusion/Exclusion Criteria. We included in our
analysis all articles published in English during the
last 5 years. This 5-year period was chosen so that
the analysis would de-emphasize those bariatric
procedures (such as jejunoileal bypass) that are no
longer widely performed. For one bariatric proce-
dure, silastic ring vertical gastroplasty, for which
there is relatively little information, older articles
were considered. All randomized controlled trials
were included, regardless of their publication date.
Case reports were not included. Case series and all
forms of controlled trials with ³10 patients that
reported patient weight before and after surgery
were included. Case reports and case series with
Meta-Analysis of Bariatric Surgery
Table 2. Summary of search strategies
Databases searched:
The Cochrane Database of Systemic Reviews (through
1999 Issue 2)
The Cochrane Registry of Clinical Trials (through 1999
Issue 2)
The Cochrane Review Methodology Database (through
1999 Issue 2)
Combined Health Information Database (CHID) (through
November 30, 1998)
CRISP (through April 1999)
Current Contents (through April 1999)
The Database of Reviews of Effectiveness (Cochrane
Library) (through 1999 Issue 2)
DIRLINE (through December 1998)
ECRI Library Catalog (through April 1999)
EMBASE (Excerpta Medica) (1975 through April 16, 1999)
HCFA CD-ROM (Health Care Financing Administration)
(through March 1999)
HCFA Web site (Health Care Financing Administration)
(through March 1999)
Health Devices Alerts (1977 through April 1999)
Healthcare Standards (1975 through April 1999)
Health Devices Sourcebase (through April 1999)
HealthSTAR (Health Services, Technology, Administration,
and Research) (1975 through May 10, 1999)
HSRProj (through May 10, 1999)
International Health Technology Assessment (IHTA) (1990
through April 1999)
Medline (1965 through April 16, 1999)
National Guideline Clearinghouse (through 1999)
Nursing and Allied Health (NAHL/CINAHL) (1982 through
November 3, 1998)
PsycINFO (1980 through December 3, 1998)
TARGET (Technology Assessment Research Guide for
Emerging Technologies) (through April 1999)
The search strategies employed a number of freetext key-
words as well as controlled vocabulary terms including (but
not limited to) the following concepts:
Controlled Trials: Randomized controlled trials; controlled
clinical trials; meta-analysis; random allocation; single-blind
method; double-blind method, evidence-based medicine
Gastric Surgery: Gastric bypass; gastroplasty; biliopancre-
atic diversion; Roux-en-Y anastomosis; vertical banding;
jejunoileal bypass; obesity/surgery; surgical stapling; stom-
ach/surgery; gastric banding; stomach bypass; gastroje-
junostomy; silastic bands
Weight: Obesity; morbid obesity; diabetic obesity; weight
reduction; weight loss; weight control; appetite depressants;
diet, reducing; anti-obesity agents; obesity/therapy
Miscellaneous: Quality of life; QOL; comorbidity; social
adjustment; patients/psychology; activities of daily living;
epidemiology; prevalence; statistics; data; demographics;
population; prediction
Co-morbidities: dyslipidemia; hyperlipidemia; lipoproteins;
lipoproteins, HDL; lipoproteins, LDL; triglycerides; choles-
terol; diabetes mellitus, non-insulin-dependent; NIDDM;
type II diabetes; blood pressure, high; hypertension; sleep
apnea syndromes
Exclusions: wasting syndrome; cachexia; obesity/dt In
general, the searches were restricted to human. Case
reports were excluded.
Obesity Surgery, 10, 2000 393
Monteforte and Turkelson
fewer than 10 patients were not included, because
the generalizability of their results, given the possi-
bility of unique patients and/or methods, is ques-
tionable. Articles that did not specify the type of
bariatric procedure and articles that co-mingled
data for more than one bariatric procedure were not
included. In order to avoid double counting, when
articles presented data from the same patients in
more than one publication, we used only the data
from the largest and most recent report.
We included only articles that expressed their
results as (or that contained enough information to
allow us to calculate) BMI. Use of BMI is consis-
tent with suggestions in the “Guidelines for
Reporting Results in Bariatric Surgery” issued by
the Standards Committee of the ASBS in 1997.6
When a study reported outcomes in terms of BMI
and percentage of ideal body weight, only the BMI
results were used in our analysis. This avoids “dou-
ble counting” of studies and allows for expression
of results in terms of BMI wherever possible.
Meta-Analytic Methods. For meta-analysis of the
effects of bariatric surgery on weight loss, we
employed the methods described by Hedges and
Olkin (fixed effects and random effects models) and
meta-analyses of data in their original metric
(BMI).7,8 Because our analysis was on studies that
reported pre- and post-surgery data, and because sta-
tistical techniques for meta-analysis of such “repeat-
ed measures” data are not well studied, we treated
each study’s pre- and post-surgical data as if they
were obtained from a contemporaneous control and
experimental group. This means that our analysis is
somewhat conservative (i.e., it underestimates the
true degree of weight loss that follows bariatric
surgery). Another consequence of the meta-analytic
technique that we employed is that studies needed to
report pre- and post-operative measures of weight
along with some corresponding measure of disper-
sion (or provide us with sufficient data to calculate
these quantities to be included in the analysis).
In order to help better understand the magnitude
of the effect sizes obtained from meta-analyses
using the Hedges and Olkin methods, we also con-
ducted meta-analyses of data in their original metric
(in this case, meta-analysis conducted in the original
metric results, in effect sizes in BMI units). As
above, pre- and post-surgical data reported by any
given study were treated as if they were obtained
394 Obesity Surgery, 10, 2000
from contemporaneous, separate groups. Some cau-
tion is required in interpreting the results of meta-
analyses, because these statistical techniques are not
well studied.
We performed separate meta-analyses using out-
comes reported at 12-months and those reported at
each study’s longest follow-up duration. Twelve-
month outcome data were used because this was
the most frequently reported duration of follow-up
among the studies that met our inclusion criteria.
To ensure that the results of each study were
expressed in accordance with the intent-to-treat
principle, data from studies that were unable to fol-
low all of their patients to the completion of the
study were analyzed in two ways. First, we
assumed that the post-operative mean and standard
deviation of the BMIs of any study’s patients who
were not followed to completion equaled the pre-
operative mean and standard deviation of all
patients. This assumption is likely to be conservative
because it is tantamount to assuming that, on aver-
age, surgery was unsuccessful in all patients not fol-
lowed to a study’s completion. Secondly, we
assumed that the pre-operative mean and standard
deviation of the BMIs of any study’s patients who
were not followed to completion equaled the mean
postoperative BMI of patients who were followed to
study completion. This assumption is likely to be
anti-conservative because it is tantamount to assum-
ing that, on average, surgery was successful in all
patients who were not followed to the end of the
study. Thus, our two assumptions to counteract the
inability to follow all patients to study completion
yield meta-analyses providing results that likely
bracket the true effectiveness of bariatric surgery.
Treating pre- and post-surgical data as if they
were obtained from separate groups of patients can
result in violation of the statistical assumptions
underlying the meta-analytic methods we
employed. Further, because the only controlled
studies on bariatric surgery are studies that mea-
sured patient outcomes before and after surgery (as
opposed to controlled trials with contemporaneous
groups), the reporting in the surgical literature
could give rise to information loss. This informa-
tion loss arises because currently available studies
do not explicitly examine whether patients respond
differently to bariatric surgery. For example, it is
entirely possible that, within the time-frame of any

given study, patients with less favorable BMIs
experience greater benefit than do patients with
more favorable BMIs. When differential responses
to surgery occur, the average treatment effect does
not describe the response of all patients to treat-
ment. In many (but not all) cases, a differential
response to treatment will cause the variance around
the pre-treatment mean to be significantly different
from the variance around the post-treatment mean.
For example, a significantly larger post-treatment
variance indicates some patients showed a substan-
tial change (i.e., weight loss) while another subset of
patients showed little or no change (or even gained
weight). We therefore tested whether the variances
around the pre- and post-surgical means were differ-
ent using Bartlett’s test for heterogeneity of vari-
ance.9 We stress that because this test would not
detect all differential responses to treatment, the
presence of statistically significant heterogeneity of
variance can be taken to indicate the presence of dif-
ferential responses, but the absence of such statisti-
cal significance cannot be taken as proof that no
such differential responses occurred.
When performing a meta-analysis, it is important
to test for the presence of statistically significant
differences among the results of the studies in the
analysis and, if such differences are found, to
determine the cause of that heterogeneity. To test
for among-study heterogeneity, we employed two
statistics, the Q-statistic and a test of each study’s
standardized residual (both tests are described by
Hedges and Olkin.7,8 To determine the cause of het-
erogeneity, we employed stepwise multiple meta-
regression. Regression analyses were conducted
for both assumptions regarding “missing” patients
(i.e., patients who were not followed to study com-
pletion) and for both the latest follow-up and 12-
month follow-up data. The variables included in
the regression analysis were percentage of patients
lost to follow-up, type of surgery (i.e., restrictive or
combination), preoperative BMI, length of follow-
up (except for 12-month outcome data), and publi-
cation year. Ratio of female to male patients and
average age of patients were also considered but
were not included, because this information was
not available for all studies. Variables were added
in a stepwise fashion in order of their z-scores (data
available upon request). The variables with the
highest z-scores were added first. We stopped
Meta-Analysis of Bariatric Surgery
adding variables when the p-value for the coeffi-
cient of the added variable was not significantly
related to the predicted variable.
Results
Using the above inclusion/exclusion criteria, 62 arti-
cles were identified, including 14 randomized con-
trolled trials (RCTs). All of these RCTs compared
one treatment to another, and none employed an
unoperated control group. For these trials, we con-
structed a synthetic control group according to the
two assumptions described in Methods. Twenty of
these 62 articles reported BMI and contained suffi-
cient information to be considered for meta-analysis
(Table 3; a list of studies that did not contain suffi-
cient information for meta-analysis is available from
the authors upon request).
Of the 20 studies reporting BMI, 11 were unable to
follow all of their patients for the full study (study
length ranged from 3 months to a mean of 98.9
months). Eight of the 14 studies reporting 12-month
follow-up data did not follow all patients for the
entire 12-month follow-up period. For these studies,
the percentage of patients lost to follow-up ranged
from 13% to 99%.
Meta-analyses were conducted for studies with
complete 12-month and latest follow-up data
(Figure 1 and Figure 2). The Q-statistic for hetero-
geneity was statistically significant for both the lat-
est follow-up data and the 12-month follow-up data.
Thus, some studies had significantly different results
from others, so that the average result of all the stud-
ies is not a particularly meaningful description of the
efficacy of any individual technique.
The standard residuals (described above) were
used to identify studies that were outliers. Applying
the method of standardized residuals to those stud-
ies included in meta-analysis of the latest and 12-
month follow-up data identified three outliers
(Figure 1) and two outliers (Figure 2), respectively.
Bartlett’s test for heterogeneity was statistically
significant for two10,11 of the three outliers by
meta-analysis of latest follow-up data and one10 of
two outliers by meta-analysis of 12-month follow-
up data. A statistically significant Bartlett’s test
suggests the presence of within study variance (i.e.,
Obesity Surgery, 10, 2000 395
Monteforte and Turkelson
Table 3. Studies considered for meta-analysis

Author Intervention Author Intervention

Zorrilla et al., 199921 VBG-GBPa Hsu et al., 199711 RYGBP
VBG-GBPb
Ballesta-Lopez et al., 199822 Laparoscopic Taskin et al., 199723 VBG
NAGB ASGB
Cowan and RYGBP-E Capella and Capella, 199624 VBG
Buffington, 199816 (Females) VBG-RYGBP
(Males)
Karayiannakis et al., 199810 VBG Ryden et al., 199619 VBG
(Successes) c
VBG
(Failures)c
Melissas et al., 199825 VBG Busetto et al., 199518 ASGB
Sapala et al., 199812 NTGBP Kolanowski, 199514 VBG
Seymour et al., 199813 VBG Carson et al., 199426 RYGBP
(hypertensive)
RYGBP
(normotensive)
van Gemert et al., 199817 VBG (M)d Letiexhe et al., 199420 SRVG
VBG (D*)d
Freeman et al., 199727 RYGBPe Brolin et al., 199228 RYGBP-1g
RYGBP-Ef RYGBP-2h
Gahtan et al., 199729 VBG Zimmerman et al., 199215 RYGBP
SRVG
ASGB: adjustable silicone gastric banding NTGBP: near total gastric bypass
RYGBP-E: extended Roux-en-Y gastric bypass RYGBP: Roux-en-Y gastric bypass
GBP: gastric bypass SRVG: Silastic ring vertical gastroplasty
NAGB: nonadjustable gastric banding VBG: vertical banded gastroplasty
a
VBG-GBP without interposition of a limb of jejunum between gastric pouch and the excluded stomach
bVBG-GBP with interposition of a limb of jejunum between gastric pouch and the excluded stomach
c
Success: >50% excess body weight loss (EBWL); Failure: <50% EBWL
d
Marlex (Bard, Billerica, MA) band; Dacron (Maedox, Boston, MA) band
e
(45-135 cm Roux limbs)
f
(180-225 cm Roux limbs)
g75 cm of jejunum bypassed
h
150 cm of jejunum bypassed

1.65 (Cowan and Buffington)

2.0.2 (Cowan and Buffington)
3.2 (Karayiannakis et al.)*
3.2 (Karayiannakis
2.08 (Seymour et al.)
2 (van Gemert et al.) et al.)*
Study

Study

2.14 (van Gemert et al.) 1.93 (Sapala et al.)

2.01 (van Gemert et al.) 2.08 (Seymour et al.)
2.6 (Hsu et al.)
2.16 (Ryden et al.) 1.71 (Busetto et al.)
0.97 (Ryden et al) 2.76 (Kolanowski, et al.)*
1.71 (Busetto et al.)
12.93 2.45 (Zimmerman et al.)
2.45 (Zimmer et al.) (Letiexhe et al.) 1.7 (Zimmerman et al.)
1.7 (Zimmerman et al.)
1.88 (Overall d)**
1.82 (Overall d**)
-5 0 5 10 15 20 0 1 2 3 4 5
Hedges’ d Hedges’ d
Figure 1. Meta-analysis of studies with complete latest Figure 2. Meta-analysis of studies with complete 12-
follow-up data. month follow-up data.
*Outlier by method of standardized residuals7 *Outlier by method of standardized residuals7
**Outliers excluded **Outliers excluded
Q-statistic = 7.3 (p = 0.696652); not statistically significant Q-statistic = 2.65 (p = 0.617755); not statistically significant

396 Obesity Surgery, 10, 2000


that some patients responded to surgery signifi-
cantly differently than other patients), and provides
further evidence that inclusion of such studies
might lead to erroneous conclusions about effect
size. Therefore, outlier studies were excluded from
further analysis.
There were no statistically significant differences
among the remaining studies that were included in
meta-analyses (i.e., the Q-test for heterogeneity was
nonsignificant and there were no outliers identified
by their standardized residuals). Meta-analysis of the
6 studies reporting latest follow-up data demonstrat-
ed an overall effect size (i.e., overall d) with 95%
confidence interval of 1.82 (1.63-2.02) (Figure 1).
Meta-analysis of the four studies reporting 12-month
follow-up data demonstrated an overall effect size
(i.e., overall d) with 95% confidence interval of 1.88
(1.68-2.07) (Figure 2).
The results of meta-analyses conducted in the
original metric are shown in Figure 3 and Figure 4.
Meta-analysis of latest follow-up data demonstrat-
ed an overall effect size (i.e., BMI reduction) with
95% confidence interval of 13.3 kg/m2 (12.1-14.6
kg/m2). Meta-analysis of 12-month follow-up data
demonstrated an overall BMI reduction of 16.4
kg/m2 (15.0-17.7 kg/m2).
The results of these meta-analyses are from stud-
ies that were able to follow all of their patients for
the entire study length. When we attempted to
include all studies in meta-analysis, there was a
significant amount of heterogeneity (i.e., a signifi-
Study
4 (Ryden et al.)
0 5
Figure 3. Meta-analysis of latest follow-up data (original metric).
*Outlier by method of standardized residuals7
**Outliers excluded
Q-statistic = 5.48 (p = 0.704982); not statistically significant
Meta-Analysis of Bariatric Surgery
cant Q-test). Significant heterogeneity was present
using either method of correcting for missing data
(previously described).
Stepwise multiple regression was performed in
order to try to identify the source of heterogene-
ity. Regression analyses were conducted for both
12-month and latest follow-up data using either
correction for missing data. Significant hetero-
geneity was observed in three of our four multi-
variate models. However, the multiple regression
analysis, in which we assumed that missing data
was equal to the pre-treatment mean appeared to
leave no statistically significant amount of unac-
counted-for variance. Unfortunately, the percent-
age of patients lost to follow-up was a significant
predictor of outcome in this model, which could
indicate that our assumption created an artifact.
Therefore, we considered this model unreliable
and conducted meta-analyses that were limited to
only studies that reported complete follow-up.
Discussion
The goal of the present article was to examine the
effects of bariatric surgery on reduction in BMI in
morbidly obese persons and to determine whether
weight reduction results in clinically significant
improvements in obesity-related comorbidities.
Meta-analyses were conducted for studies of
13 (Cowan and Buffington)
17.2 (Cowan and Buffington)
21 (Karayiannakis et al.)*
12.8 (Seymour et al.)
13 (van Gemert et
16.7 (van Gemert et al.)
15 (van Gemert et al.)
18.9 (Hsu et al.)
11.5 (Ryden et al.)
12.4 (Busetto et al.)
11.4 (Letiexhe et al.)
15.8 (Zimmerman et al.)
14 (Zimmerman et al.)
13.3 (Overall effect size)**
10 15 20 25 30
Effect Size (BMI units)
Obesity Surgery, 10, 2000 397
Monteforte and Turkelson
NIH classification of overweight and obesity, this
21 (Karayiannakis et al.) weight loss translates to a decrease from obesity
17.4 Sapala et al.) class III (BMI ³40 kg/m2) to overweight (BMI
12.8 (Seymour et al.)
25.0-29.9 kg/m2), obesity class I (BMI 30.0-34.9
12.4 (Busetto et al.)*
Study

16.04 (Kolanowski et al.) kg/m2), or obesity class II (BMI 35.0-39.9 kg/m2).

15.8 (Zimmerman et al.)
14 (Zimmerman et al.)
16.4 (Overall effect size)**
0 5 10 15 20 25 30
Effect Size (BMI units)
Figure 4. Meta-analysis of 12-month follow-up data
(original metric).
*Outlier by method of standardized residuals7
**Outliers excluded
Q-statistic = 3.97 (p = 0.410080); not statistically significant
bariatric surgery that reported BMI and had com-
plete patient follow-up. From the results of meta-
analysis (conducted in their original metric), the
following conclusions can be made. Based upon
the four studies12-15 reporting BMI results with
complete 12-month follow-up, morbidly obese
patients had an overall reduction in BMI of 16.4
kg/m2. Based upon the six studies with complete
follow-up for the final reported BMI,13,15-19 mor-
bidly obese patients had an overall reduction in
BMI of 13.3 kg/m2. Mean duration of follow-up for
the six studies included in the latter meta-analysis
ranged from 9 months to 7 years.
For patients included in the meta-analysis, this
weight reduction resulted in post-operative BMIs
ranging from 30 to 34.1 kg/m2 at 1 year and 27.7 to
39.0 kg/m2 at latest follow-up. According to the
Although these results are based upon a relative-
ly small number of studies, they are based on stud-
ies that contained the most completely reported
information with the highest percentage of follow-
up. As such, these estimates may underestimate
effect size, because, for statistical reasons, the
analysis does not contain some of the studies that
reported more favorable results. It may be possible
that some (but not necessarily all) patients actually
lose more weight (i.e., have lower BMI scores)
after bariatric surgery.
These findings also suggest that the initial reduc-
tion in BMI (as measured by 12-month results)
might be greater than BMI achieved when patients
are followed for longer periods of time. This would
be consistent with the findings of the 1991 NIH con-
sensus development conference on gastrointestinal
surgery for severe obesity.4 This conference noted
that substantial weight loss generally occurs, with
the weight nadir occurring in 18 to 24 months.4 The
committee also noted that some regain of weight is
common by 2 to 5 years after operation.
In 1997, the Standards Committee of the
American Society for Bariatric Surgery recom-
mended that ideal long-term follow-up for weight
loss should be 5 years or longer and discouraged
reporting weight loss with less than 2 years of fol-

Table 4. Summary of reported complicationsa by procedure type

Complication Restrictive Procedures (n = 3,568) Combination Procedures (n = 3,626)
Perioperative Deaths 5 (0.14%) 14 (0.39%)
(£ 30 days of surgery)
Bleeding 16 (0.45%) 32 (0.88%)
Dumping Syndrome 10 (0.28%) 531 (14.64%)
Infection 111 (3.11%) 191 (5.27%)
Splenic Injury 8 (0.22%) 29 (0.8%)
Stomal Stenosis/
Gastric Outlet Obstruction 79 (2.21%) 97 (2.68%)
Gastric Pouch/Stoma
Dilation 86 (2.41%) 17 (0.47%)
Staple-Line Failure 55 (1.54%) 216 (5.96%)
Stomach Erosion/
Ulceration 43 (1.21%)* 42 (1.16%)*
*20 (0.56%) of these were band erosions **2 (0.06%) of these were band erosions
Vitamin/Mineral 58 (1.63%) 398 (10.98%)
Deficiency
Vomiting 303 (8.49%) 93 (2.56%)
aComplications reported as number and percentage of patients.

398 Obesity Surgery, 10, 2000


low-up.6 Duration of follow-up was 12 months or
less in four of the six studies included in meta-
analysis.13,15,16,18 Therefore, overall reduction in
BMI reflected by these studies may underestimate
maximum weight-loss potential for these patients.
Furthermore, this overall reduction in BMI may
not reflect the final BMI if this group of patients
were followed for longer periods of time.
Although we included studies that used a variety
of surgical procedures in our meta-analysis, our
results suggest that there is little difference in effi-
cacy among these procedures. Among the six stud-
ies included in the meta-analysis, five different sur-
gical procedures were represented: three restrictive
procedures (ASGB, SRVG, and VBG) and two
combination procedures (RYGBP and RYGBP-E).
As such, it is difficult to draw conclusions about
either type of surgery or about any particular
bariatric procedure. However, the lack of a signifi-
cant Q-test for heterogeneity or any outliers suggests
that, within this data set, there is no difference
between procedures. Stated another way, there were
no significant differences in effect size resulting
from any of the five different bariatric operations
included in the meta-analysis. However, more stud-
ies are needed to determine whether any clinically
meaningful difference among operations exists.
In order to ascertain whether the results of stud-
ies included in meta-analysis are representative, we
performed a regression analysis of change in BMI
as a function of initial BMI for both included and
non-included studies. Regression analysis demon-
strated that the relationship between initial BMI
and change in BMI is linear (i.e., greater reductions
in BMI were observed for patients with a greater
initial BMI). The correlation (r2) between initial
BMI and change in BMI for included studies was
0.58. The correlation (r2) between initial BMI and
change in BMI for non-included studies was 0.25
(difference not statistically significant).
There have been a number of reported complica-
tions associated with bariatric surgery (Table 4). In
the present article, complications reported by the
62 studies that met initial inclusion criteria were
totaled and grouped by surgery category (i.e.,
restrictive vs. combination procedures). The total
number of complications reported by each study
were tabulated as opposed to the number of
patients who developed complications. Therefore,
Meta-Analysis of Bariatric Surgery
some patients may have experienced more than one
complication while other patients may have experi-
enced none. It should also be noted that the report-
ed outcomes might underestimate the actual num-
ber of complications because complications were
either not provided by a particular study or a par-
ticular complication may not have been reported.
Because it is not clear whether there is complete
reporting of complications in the literature, we did
not attempt to compare complication rates among
the different surgical procedures.
Morbid obesity is accompanied by a reduction in
life expectancy, which is due in large part to signifi-
cant comorbid associations in the form of metabolic
abnormalities and several severe cardiopulmonary
disorders.4 In addition, significant psychosocial and
economic problems are frequently experienced by
persons with severe (morbid) obesity.4
Evidence of the effect that weight loss following
bariatric surgery has on various comorbid condi-
tions stems largely from case series reports. One of
the key problems in evaluating reports of case
series in surgical therapy is the lack of standards
for comparison.4 The practice of comparing post-
operative indicators of comorbidity to the same
patient’s own pre-operative status is insufficient for
evaluation of long-term effects and of survival.4
The difficulty in evaluating the effects of weight
loss on various obesity-related comorbidities is
compounded by the lack of complete follow-up.
Twenty of the 62 studies that met the initial criteria
for inclusion in the present article provided com-
plete follow-up data. Of these 20 studies, only
three studies reported the effects of weight loss fol-
lowing bariatric surgery on one or more obesity-
related comorbidities.10,16,20 Therefore, a quantita-
tive analysis of this data was not performed.
To ascertain the impact of BMI on mortality and
to determine the effects of weight loss on several
major obesity-related comorbidities, other evi-
dence-based sources were consulted. The magni-
tude of weight loss following bariatric surgery, as
determined through meta-analysis, was used to
examine whether such weight loss is clinically
meaningful.
The relationship between BMI and mortality was
recently described by the American Cancer
Society.3 In this large, prospective study, the lowest
rates of death from all causes were found at BMIs
Obesity Surgery, 10, 2000 399
Monteforte and Turkelson
between 22.0 kg/m2 and 23.4 kg/m2 in women and
23.5 kg/m2 and 24.9 kg/m2 in men.3 Among healthy
patients who had never smoked, the relative risk for
all-cause mortality increased with BMI levels of
approximately 30 kg/m2 or more. The reported rela-
tive risks of death from all causes among adults (age
30-64 years) are listed in Table 5.
Data regarding the effect of weight loss on major
obesity-related comorbidities were derived from the
1998 NIH “Clinical Guidelines on the Identification,
Evaluation, and Treatment of Overweight and
Obesity in Adults”.1 This evidence-based report sys-
tematically reviewed the relevant published scientif-
ic literature. The available evidence was derived
largely from randomized controlled trials that exam-
ined the impact of weight loss achieved through
nonsurgical measures. The findings of the NIH
report are summarized in Table 6.
The mean pre-operative BMI for those patients
from the six studies included in meta-analysis of
latest outcome data ranged from 41 kg/m2 to 52.3
kg/m2. Meta-analysis of these studies demonstrat-
ed a 13.3 kg/m2 reduction in BMI. Applying this
BMI reduction to this group of patients would
result in post-operative BMIs ranging from 27.7 to
39 kg/m2 (25%-32% reduction).
According to prevalence data from the National
Health and Nutrition Survey (NHANES III, 1988 to
Table 5. Reported relative-risks of death from all causes among obese adults (age 30-64 years)
Men
30.0-31.9
Relative risk (95% confidence interval) 1.62
(1.34-1.97)
Women
30.0-31.9
Relative risk (95% confidence interval) 1.51
(1.28-1.79)
Data from Calle et al.3
Table 6. Changes in comorbidity associated with weight reductiona
Comorbidity Weight Reduction
Hypertension 10 kg
Type 2 diabetes 2.4 to 5 kg
Lipids 5% to 13%
Obstructive sleep apnea 10%
aSource: NIH, “Clinical Guidelines on the Identification, Evaluation, and Treatment of Overweight and Obesity in Adults.”1
400 Obesity Surgery, 10, 2000
1994), the average height for men and women more
than 20 years of age is 1.68 and 1.62 m, respective-
ly.2 Applying the 13.3 to kg/m2 reduction in BMI to
this population translates to a weight reduction of
37.5 kg for men and 34.9 kg for women.
Based on the findings of the NIH report (Table
6), weight reduction following bariatric surgery in
morbidly obese patients may be associated with
improvements in risk factors associated with cardiac
disease, including hypertension, type 2 diabetes, and
lipid abnormalities. For patients with obstructive
sleep apnea, weight reduction may result in decreas-
ing the severity of this condition. Because weight
reduction following bariatric surgery is greater than
the weight loss reported by studies included in the
NIH report, significant improvements in these
comorbidities are expected. However, the magnitude
of these improvements can only be inferred.
Likewise, it appears that reduction in BMI following
bariatric surgery would decrease patients’ relative
risk for all-cause mortality.
Further research is needed to examine the long-
term benefits of bariatric surgery, particularly with
respect to obesity-related comorbidities. Research
in this area might consist of RCTs that compare the
levels of morbidity and mortality for patients who
receive surgery to those of patients who do not.
Additional research comparing the various surgical
BMI (kg/m2)
32.0-34.9 >35
2.05 2.30
(1.66-2.53) (1.72-3.06)
32.0-34.9 35.0-39.9 ³40
1.53 1.86 2.70
(1.27-1.84) (1.51-2.30) (2.03-3.60)
Associated Change in Comorbidity
Systolic blood pressure: -7 mm Hg
Diastolic blood pressure: -3 mm Hg
HbA1c: -2% to -2.4%
Cholesterol: 0% to -18%
Triglycerides: -2% to -44%
LDL cholesterol: -3% to -22%
HDL cholesterol: -7% to 22%
50% reduction in severity

procedures used to treat obesity is needed.
ECRI, a non-profit independent healthcare research organization,
is a Collaborating Center of the World Health Organization.
References
1. National Heart, Lung, and Blood Institute, NHLBI Obesity
Education Initiative Expert Panel. Clinical guidelines on the
identification, evaluation, and treatment of overweight and
obesity in adults. The evidence report. Bethesda (MD):
National Institutes of Health, National Heart, Lung and
Blood Institute (NHLBI). 1998 June, p. 228.
2. Kuczmarski RJ, Carroll MD, Flegal KM et al. Varying
body mass index cutoff points to describe overweight
prevalence among U.S. adults: NHANES III (1988 to
1994). Obes Res 1997;5: 542-8.
3. Calle EE, Thun MJ, Petrelli JM et al. Body-mass index and
mortality in a prospective cohort of U.S. adults. N Engl J
Med 1999; 341: 1097-105.
4. Gastrointestinal surgery for severe obesity: National Institutes
of Health Consensus Development Conference Statement. Am
J Clin Nutr 1992;55: 615S-9S.
5. Mason EE, Tang S, Renquist KE et al. A decade of change in
obesity surgery. National Bariatric Surgery Registry (NBSR)
Contributors. Obes Surg 1997;7:189-97.
6. American Society for Bariatric Surgery, Standards
Committee. Guidelines for reporting results in bariatric
surgery. Obes Surg 1997;7: 521-2.
7. Hedges LV, Olkin I. Statistical methods for meta-analysis.
Boston: Academic Press, Inc. 1985. p 369.
8. Cooper H, Hedges LV, eds. The Handbook of Research
Synthesis. New York: Russell Sage Foundation: 1994. p.
573.
9. Armitage P, Berry G. Statistical Methods in Medical
Research. 3rd ed. Oxford: Blackwell Scientific: 1994. p 620.
10. Karayiannakis AJ, Syrigos KN, Zbar A et al. The effect of
vertical banded gastroplasty on glucose-induced beta- endor-
phin response. J Surg Res 1998; 80: 123-8.
11. Hsu LK, Sullivan SP, Benotti PN. Eating disturbances and
outcome of gastric bypass surgery: a pilot study. Int J Eat
Disord 1997;21: 385-90.
12. Sapala JA, Wood MH, Sapala MA et al. Marginal ulcer
after gastric bypass: a prospective 3-year study of 173
patients. Obes Surg 1998; 8: 505-16.
13. Seymour K, Mackie A, McCauley E et al. Changes in
esophageal function after vertical banded gastroplasty as
demonstrated by esophageal scintigraphy. Obes Surg
1998; 8: 429-33.
14. Kolanowski J. Gastroplasty for morbid obesity: the
Meta-Analysis of Bariatric Surgery
internist’s view. Int J Obes 1995;19 (Suppl 3): S61-5.
15. Zimmerman V, Campos CT, Buchwald H. Weight loss
comparison of gastric bypass and Silastic ring vertical gas-
troplasty. Obes Surg 1992; 2: 47-9.
16. Cowan GS Jr, Buffington CK. Significant changes in blood
pressure, glucose, and lipids with gastric bypass surgery.
World J Surg 1998; 22: 987-92.
17. van Gemert WG, Adang EM, Greve JW et al. Quality of
life assessment of morbidly obese patients: effect of
weight-reducing surgery. Am J Clin Nutr 1998; 67: 197-
201.
18. Busetto L, Valente P, Pisent C et al. Eating pattern in the
first year following adjustable silicone gastric banding
(ASGB) for morbid obesity. Int J Obes 1996; 20: 539-46.
19. Ryden O, Hedenbro JL, Frederiksen SG. Weight loss after
vertical banded gastroplasty can be predicted: a prospective
psychological study. Obes Surg 1996; 6: 237-43.
20. Letiexhe MR, Scheen AJ, Gerard PL et al. Insulin secre-
tion, clearance and action before and after gastroplasty in
severely obese subjects. Int J Obes 1994; 18: 295-300.
21. Zorrilla PG, Salinas RJ, Salinas-Martinet AM. Vertical
banded gastroplasty-gastric bypass with and without the
interposition of jejunum: Preliminary report. Obes Surg
1999; 9: 29-32.
22. Ballesta-Lopez C, Bastida-Vila X, Catarci M et al.
Laparoscopic gastric banding for morbid obesity with
expanded PTFE: technique and early results in the first 100
consecutive cases. Hepatogastroenterology 1998; 45: 2447-
52.
23. Taskin M, Apaydin BB, Zengin K et al. Stoma adjustable
silicone gastric banding versus vertical banded gastroplas-
ty for the treatment of morbid obesity. Obes Surg 1997; 7:
424-8.
24. Capella JF, Capella RF. The weight reduction operation of
choice: vertical banded gastroplasty or gastric bypass. Am
J Surg 1996;171: 74-9.
25. Melissas J, Christodoulakis M, Spyridakis M et al.
Disorders associated with clinically severe obesity: signif-
icant improvement after surgical weight reduction. South
Med J 1998;91: 1143-8.
26. Carson JL, Ruddy ME, Duff AE et al. The effect of gastric
bypass surgery on hypertension in morbidly obese patients
[published erratum appears in Arch Intern Med 1994; 154:
1770]. Arch Intern Med 1994; 154: 193-200.
27. Freeman JB, Kotlarewsky M, Phoenix C. Weight loss after
extended gastric bypass. Obes Surg 1997; 7: 337-44.
28. Brolin RE, Kenler HA, Gorman JH et al. Long-limb gas-
tric bypass in the superobese. A prospective randomized
study. Ann Surg 1992; 215: 387-95.
29. Gahtan V, Goode SE, Kurto HZ et al. Body composition
and source of weight loss after bariatric surgery. Obes Surg
1997; 7: 184-8.
(Received March 21, 2000; accepted June 16, 2000)
Obesity Surgery, 10, 2000 401