We performed a broad search of publications paralleling the QUOROM statement 13
related to bariatric surgery, obesity, and CV risk factors using MEDLINE (1950 to cutoff date April 8th, 2008) using the following search terms: obesity surgery, gastroplasty, gastric bypass, bariatric surgery, obesity/su [surgery], anastomosis, roux-en-Y gastric bypass. We independently combined these terms, and obesity with either gastric banding, biliopancreatic diversion, or jejunoileal bypass, thereby providing us with a total of 12,018 citations. We limited these studies to any of the following criteria for type of publication: retrospective studies, randomized controlled trials, longitudinal studies, prospective studies, cohort studies, case-control studies, clinical trial [publication type], comparative study, or follow-up studies, resulting in 4,346 citations. The authors subsequently combined these citations with any of the following keywords: diabetes mellitus, glucose intolerance, metabolic syndrome, hyperlipidemia, hypertension, hypertriglyceridemia, hypercholesterolemia, risk factors, cardiovascular risk factors, or comorbidity. We excluded non-human studies, pediatric studies and limited our search to the English literature, leaving 656 citations.
Of these 656 citations, the primary author (JAB) reviewed the individual abstracts and further excluded case reports, letters, reviews, or commentaries that may have eluded our initial search (n=43). Studies with <6 months follow-up (n=252), those unrelated to bariatric surgery (n=113), and studies with <100 patients (n=104) were omitted. We excluded follow-up studies performed at the same institution (n=4). Of the 140 remaining citations, we excluded those without mention of CV risk factors: diabetes, hypertension, hypercholesterolemia, or weight change (n=64).
We reviewed the entire manuscripts of the remaining 75 citations and excluded studies for the following reasons: study data presented in descriptive or non-numeric form (n=30); cost studies (n=5); no data on CV risk factors (n=4); studies dated earlier than 1980 (n=7); brief reports including a letter, a commentary and an editorial each (n=3); quality of life study (n=3); studies with <100 patients (n=7); 2 studies that utilized the FRS in their analysis; one study each with a follow-up of less than one year and another with >20% of patients with an incomplete follow-up; and 3 studies were re-analyzed data of previously published studies. Four studies had missing numeric data for hypertension or hypercholesterolemia. We contacted these authors in an attempt to obtain data from the primary source. One author supplied this missing data, one responded but did not have the data available and two authors did not respond to our request. Six studies fulfilled all of our inclusion criteria, including our own population-based study () 10,12,14–17
Studies Examined Using Systematic Review
Selection of the validation cohort has been previously published 12
. Briefly, we performed a historical, population-based study examining all Olmsted County residents referred for Roux-en-Y gastric bypass to the Mayo Clinic Nutrition Center from 1990–2003 using the Rochester Epidemiology Project, a medical record-linkage system. All medical records are available for review allowing for complete ascertainment of patient’s history. Our final study cohort consisted of 197 surgical patients and 163 non-operative patients. To validate our previously published risk model 12
we applied the FRS and PROCAM risk scores 1,7
to this cohort. We included only patients with complete data to calculate risk scores. For the FRS, complete data were available on 182 surgical and 158 control patients, whereas for the PROCAM, complete data were available for 173 surgical and 141 control patients.
The FRS is the most commonly used CV risk tool in the USA. Many other risk scores are derived from the FRS and hence were not considered in this study. The predicted score is based on a community-based cohort of 5345 patients, aged 30–74 years at the time of the initial Framingham examination. Follow-up was 12 years with a total of 610 patients experiencing a cardiac event. The latest version 1
was used to compute 10-year risk of fatal or non-fatal coronary events using the following variables in this model: sex, age, total and high density cholesterol, systolic blood pressure, smoking (yes/no), and a diagnosis of diabetes mellitus (yes/no).
The PROCAM score 7
was utilized because many of our included studies were of European origin. The study was performed among German government and company workers between 1979 and 1985 with 96% follow-up. These authors developed a 10-year CV prediction score to estimate the global risk of both fatal and non-fatal coronary events based on an actual 325 acute coronary events in the 5389 men followed-up between 35 to 65 years of age at time of recruitment. Variables included consisted of age, low density lipoprotein, high density lipoprotein, triglycerides, systolic blood pressure, smoking (yes/no), diabetes (yes/no), and a family history of a myocardial infarction (yes/no).
We decided not to use other commonly known risk assessments. The UK Prospective Diabetes Mellitus study and its risk equation 18
is limited specifically to patients with only diabetes mellitus and would not be applicable in our cohort. The Systematic Coronary Evaluation Risk Score 19
focuses on CV and non-CV deaths and not events. Its computed score in younger patients (age<65) based on the risk assessment charts would be very low and therefore was not used.
The FRS and PROCAM risk tables were used to compute 10-year risk for both our validation cohort and the individual studies. For the validation cohort, we calculated 10-year risks separately by gender on the overall cohort, and when the patient’s age was standardized at age 55 years. We utilized the mean values for the required risk function variables to calculate the FRS and PROCAM scores. A composite score was obtained and converted into 10-year CV risk using their respective data. The purpose was to delineate directionality of CV risk after surgical intervention, using studies obtained from our systematic review and not to assess 10-year risk precisely. The Friedewald formula was used in studies (n=3) that did not measure low density lipoprotein directly. For studies that did not provide information about specific CV risk factors including smoking status or family history of myocardial infarction, the authors used data previously published from the risk model’s original study cohort to impute these values. For the FRS, we assumed that 39% of all patients were smokers at both baseline and follow-up in both sexes. We assumed that 21.6% had a positive family history in calculating the PROCAM score, in studies with missing variables. The proportion of diabetics in the Swedish Obesity study was calculated using the reported % of patients recovered from diabetes along with the number of diabetics at follow-up10
. For the individual studies, the proportion of diabetics or smokers was multiplied by the number of risk points for that entity. Separate estimates were calculated by sex. Because the FRS consists of two separate tables by gender, we estimated the scores separately, first assuming that all patients were male, and subsequently considered all patients to be females. As all manuscripts contained the demographic mix by sex, we calculated an overall score using the proportional mix of sexes and each individual score. In studies with a control group, the delta between baseline and follow-up was obtained to determine the difference in CV risk for patients who have undergone bariatric surgery compared to control patients.
For the validation cohort, continuous data are presented as mean ± standard deviation. For comparisons within each cohort between baseline and follow-up, we used a two-sided paired t-tests and Wilcoxon Signed Rank. We compared the changes between groups with a two-sample t-test of unequal variances and the Wilcoxon Rank Sum test. A p-value <0.05 was considered statistically significant. Descriptive statistics were provided only to describe the risks from the mean scores. All analyses were performed using JMP for SAS (Windows version 7.0.0, SAS Institute Inc, Cary, NC).