OBES SURG DOI 10.1007/s11695-014-1197-y
ORIGINAL CONTRIBUTIONS
Use of Bariatric Outcomes Longitudinal Database (BOLD) to Study Variability in Patient Success After Bariatric Surgery Stephen C. Benoit & Tina D. Hunter & Diane M. Francis & Nestor De La Cruz-Munoz
# Springer Science+Business Media New York 2014
Abstract Background This study was conducted to determine the contributions of various predictors to the large variations in absolute weight loss and percent body mass index (BMI) loss after bariatric surgery. Methods The data source was the Bariatric Outcomes Longitudinal Database S M by the Surgical Review Corporation. Eligibility criteria included a first bariatric surgery for adjustable gastric band (AGB), Roux-en-Y gastric bypass (RYBG), or sleeve gastrectomy (SG) between January 2007 and February 2010; age 21 years or older; presurgery BMI>30 kg/m2; and at least one preoperative visit within 6 months and at least one postoperative visit 30 days or more after surgery. Potential predictor variables included procedural details, patient demographics, comorbidities, and prior surgical history. Linear regression models of absolute weight loss and %BMI loss were fitted at 12, 18, and 24 months. The 12month absolute weight loss endpoint was then chosen for a
more in-depth analysis of variability through variable transformations and separate models by procedure. Results A total of 31,443 AGB, 40,352 RYGB, and 2,194 SG patients met all inclusion criteria. Regression models explained 37 to 55 % of the variability in %BMI loss and 52 to 65 % of variability in absolute weight loss. The key predictors for absolute weight loss at 12 months were procedure (44.8 %) and baseline weight (18.5 %), with 34.2 % of the variability unexplained. Other significant predictors, each of which accounted for 30 kg/m2 Age≥21 years on the day of bariatric surgery At least one preoperative visit within 6 months prior to the bariatric surgery and at least one postoperative visit
recorded at least 30 days after the index surgery, both with valid BMI and/or height and weight recorded Subpopulations with postoperative visits between 3 and 9, 9 and 15, 15 and 21, and 21 and 27 months after surgery were identified for 6-, 12-, 18-, and 24-month endpoints of percent BMI loss and absolute weight loss. If a patient had multiple visits within a time period, data from the visit nearest the midpoint of that time period was captured in the analyzable dataset. General linear models were fitted for the outcomes of absolute weight loss and percent BMI loss to explore which of these dependent variables could best be modeled, as measured by the percent of variance explained. Each outcome variable was modeled at three time points (12, 18, and 24 months), using those subsets of patients with follow-up visits in the respective time period. Three different methods of selecting predictor variables were used for each combination of outcome and time period. These selection methods included (1) saturated, wherein all potential predictor variables remained in the model; (2) least angle regression (LAR) selection; and (3) stepwise selection. The 45 potential predictor variables included type of procedure (RYGB, AGB, or SG), surgical approach (laparoscopic, laparoscopic with robotic assist, open, converted to open, hand-assisted, or other), surgeon (surgical resident participated, surgical fellow participated, no residents or fellows), age at date of surgery, sex (male or female), race (African American, Asian, Caucasian, Hispanic, Native American, Pacific Islander/Hawaiian, or other), height, weight, BMI, American Society of Anesthesiologists (ASA) classification (1–5), prior surgery (abdominal, breast cancer, cardiovascular, joint, any nonbariatric), baseline comorbidities (alcohol usage, substance abuse, anxiety, back pain, chronic heart failure, depression, diabetes, fibromyalgia, hypertension, hyperlipidemia, liver disease, musculoskeletal disorder, sleep apnea), number of baseline medications, baseline smoking (yes, no), employment status (disabled, full time, homemaker, not specified, part time, retired, self-employed, student, unemployed), and postoperative group support (monthly, no attendance, not known, weekly, yearly). After fitting all 18 models above (2 endpoints×3 time points×3 selection methods), the 12-month absolute weight loss endpoint was explored in further depth. In particular, this endpoint was utilized to understand whether transformations of key predictor variables would increase the percent of variability in the outcome that could be explained and also to explore differences in the statistical significance of individual predictors among the three types of bariatric procedures. Separate models were fitted for each type of bariatric procedure, and the proportion of the total population variability in weight loss explained by each predictor in each model was tracked through the sums of squares to determine how much
OBES SURG
of the total variability could be explained. Finally, simplified models were created to estimate and plot expected 12-month absolute weight loss as a function of the type of bariatric procedure and baseline weight.
Results The analysis dataset included a total of 73,989 patients, of whom 55 % (40,352) had RYGB, 42 % (31,443) had AGB,
Table 1 Patient characteristics Category
Total N Age at index procedure 50 % EWL (e.g., 51 and 100 % EWL were treated equally as success endpoints, while 0 and 49 % EWL were treated equally as failure endpoints). In reality, there are varying degrees of success and many alternate definitions, including different levels of absolute weight or BMI loss, and resolution or reduction of comorbid conditions associated with morbid obesity, all of which are related to the amount of weight/BMI loss achieved by a patient. Thus, explaining the underlying patient variability in the amount of absolute weight loss or %BMI loss as a result of bariatric surgery is an important undertaking. In contrast, this detail is lost when a success outcome is dichotomized at a single cutoff level. Multivariable regression analysis of absolute weight loss for the cohort of 26,921 patients with 12-month outcomes in this study showed that the type of bariatric procedure explained 44.8 % of the variability in weight loss results among patients. This finding is consistent with the results from several meta-analyses which found that mean weight loss varied
OBES SURG Fig. 2 Estimated weight loss at 12 months by bariatric procedure and baseline weight. AGB adjustable gastric band, RYGB Roux-en-Y gastric bypass, SG sleeve gastrectomy
by procedure type, with the greatest weight loss occurring after RYGB, intermediate weight loss after gastroplasty, and the lowest weight loss after AGB [8, 9, 16]. In addition, all systematic reviews and meta-analyses comparing %EWL achieved with different types of bariatric procedures have reported higher %EWL following RYGB compared with AGB [5, 8, 9, 17–19] or SG [17, 20]. The finding of a mean weight loss of 46.2 kg at 12 months after RYGB in the current study was consistent with metaanalyses which reported mean weight losses from 43.5 to 47.1 kg at 6 to 36 months of follow-up [8, 9, 16]. However, our finding of only 21.4 kg mean weight loss at 12 months after AGB is considerably less than the weight losses of 27.4 to 32.4 kg reported in the same meta-analyses (Table 4) and the mean weight loss of 42.9 kg at 3 years of postoperative follow-up reported in a meta-analysis of AGB studies [21]. These differences could be due in part to varying durations of follow-up among the studies because we saw continuing weight loss with increasing time after AGB, with a mean
weight loss of 25.5 kg in a smaller population that had 24 months of follow-up. Other contributing factors to this difference could be differences in inclusion/exclusion criteria, such as starting BMI and/or comorbid conditions. Finally, greater levels of support, follow-up, or band adjustment in smaller clinical trial groups or different standards of care at the time of the earlier studies vs. the current study may be likely contributors to the differences in mean weight loss. Our analysis found that baseline weight accounted for 18.5 % of the total variability in 12-month absolute weight loss outcomes, while factors including age, race, sex, height, diabetes, number of medications, and the log of days after surgery (within the 12-month window of 9 to 15 months) accounted for only 2.5 % of the variability. By type of bariatric procedure, preoperative weight explained 45.1 and 45.5 % of the variability for RYGB and SG, respectively, but only 14.1 % of the variability for AGB. The importance of preoperative weight for outcomes after RYGB is consistent with the findings from previous studies
Table 4 Comparison of absolute weight loss outcomes in this study and published meta-analyses Study
Follow-up Mean absolute weight loss from baseline (months) Total proceduresa RYGB AGB
Current study 12
Buchwald et al. [8]
6-36
Maggard et al. [16]
12
Buchwald et al. [9]
ORIGINAL CONTRIBUTIONS
Use of Bariatric Outcomes Longitudinal Database (BOLD) to Study Variability in Patient Success After Bariatric Surgery Stephen C. Benoit & Tina D. Hunter & Diane M. Francis & Nestor De La Cruz-Munoz
# Springer Science+Business Media New York 2014
Abstract Background This study was conducted to determine the contributions of various predictors to the large variations in absolute weight loss and percent body mass index (BMI) loss after bariatric surgery. Methods The data source was the Bariatric Outcomes Longitudinal Database S M by the Surgical Review Corporation. Eligibility criteria included a first bariatric surgery for adjustable gastric band (AGB), Roux-en-Y gastric bypass (RYBG), or sleeve gastrectomy (SG) between January 2007 and February 2010; age 21 years or older; presurgery BMI>30 kg/m2; and at least one preoperative visit within 6 months and at least one postoperative visit 30 days or more after surgery. Potential predictor variables included procedural details, patient demographics, comorbidities, and prior surgical history. Linear regression models of absolute weight loss and %BMI loss were fitted at 12, 18, and 24 months. The 12month absolute weight loss endpoint was then chosen for a
more in-depth analysis of variability through variable transformations and separate models by procedure. Results A total of 31,443 AGB, 40,352 RYGB, and 2,194 SG patients met all inclusion criteria. Regression models explained 37 to 55 % of the variability in %BMI loss and 52 to 65 % of variability in absolute weight loss. The key predictors for absolute weight loss at 12 months were procedure (44.8 %) and baseline weight (18.5 %), with 34.2 % of the variability unexplained. Other significant predictors, each of which accounted for 30 kg/m2 Age≥21 years on the day of bariatric surgery At least one preoperative visit within 6 months prior to the bariatric surgery and at least one postoperative visit
recorded at least 30 days after the index surgery, both with valid BMI and/or height and weight recorded Subpopulations with postoperative visits between 3 and 9, 9 and 15, 15 and 21, and 21 and 27 months after surgery were identified for 6-, 12-, 18-, and 24-month endpoints of percent BMI loss and absolute weight loss. If a patient had multiple visits within a time period, data from the visit nearest the midpoint of that time period was captured in the analyzable dataset. General linear models were fitted for the outcomes of absolute weight loss and percent BMI loss to explore which of these dependent variables could best be modeled, as measured by the percent of variance explained. Each outcome variable was modeled at three time points (12, 18, and 24 months), using those subsets of patients with follow-up visits in the respective time period. Three different methods of selecting predictor variables were used for each combination of outcome and time period. These selection methods included (1) saturated, wherein all potential predictor variables remained in the model; (2) least angle regression (LAR) selection; and (3) stepwise selection. The 45 potential predictor variables included type of procedure (RYGB, AGB, or SG), surgical approach (laparoscopic, laparoscopic with robotic assist, open, converted to open, hand-assisted, or other), surgeon (surgical resident participated, surgical fellow participated, no residents or fellows), age at date of surgery, sex (male or female), race (African American, Asian, Caucasian, Hispanic, Native American, Pacific Islander/Hawaiian, or other), height, weight, BMI, American Society of Anesthesiologists (ASA) classification (1–5), prior surgery (abdominal, breast cancer, cardiovascular, joint, any nonbariatric), baseline comorbidities (alcohol usage, substance abuse, anxiety, back pain, chronic heart failure, depression, diabetes, fibromyalgia, hypertension, hyperlipidemia, liver disease, musculoskeletal disorder, sleep apnea), number of baseline medications, baseline smoking (yes, no), employment status (disabled, full time, homemaker, not specified, part time, retired, self-employed, student, unemployed), and postoperative group support (monthly, no attendance, not known, weekly, yearly). After fitting all 18 models above (2 endpoints×3 time points×3 selection methods), the 12-month absolute weight loss endpoint was explored in further depth. In particular, this endpoint was utilized to understand whether transformations of key predictor variables would increase the percent of variability in the outcome that could be explained and also to explore differences in the statistical significance of individual predictors among the three types of bariatric procedures. Separate models were fitted for each type of bariatric procedure, and the proportion of the total population variability in weight loss explained by each predictor in each model was tracked through the sums of squares to determine how much
OBES SURG
of the total variability could be explained. Finally, simplified models were created to estimate and plot expected 12-month absolute weight loss as a function of the type of bariatric procedure and baseline weight.
Results The analysis dataset included a total of 73,989 patients, of whom 55 % (40,352) had RYGB, 42 % (31,443) had AGB,
Table 1 Patient characteristics Category
Total N Age at index procedure 50 % EWL (e.g., 51 and 100 % EWL were treated equally as success endpoints, while 0 and 49 % EWL were treated equally as failure endpoints). In reality, there are varying degrees of success and many alternate definitions, including different levels of absolute weight or BMI loss, and resolution or reduction of comorbid conditions associated with morbid obesity, all of which are related to the amount of weight/BMI loss achieved by a patient. Thus, explaining the underlying patient variability in the amount of absolute weight loss or %BMI loss as a result of bariatric surgery is an important undertaking. In contrast, this detail is lost when a success outcome is dichotomized at a single cutoff level. Multivariable regression analysis of absolute weight loss for the cohort of 26,921 patients with 12-month outcomes in this study showed that the type of bariatric procedure explained 44.8 % of the variability in weight loss results among patients. This finding is consistent with the results from several meta-analyses which found that mean weight loss varied
OBES SURG Fig. 2 Estimated weight loss at 12 months by bariatric procedure and baseline weight. AGB adjustable gastric band, RYGB Roux-en-Y gastric bypass, SG sleeve gastrectomy
by procedure type, with the greatest weight loss occurring after RYGB, intermediate weight loss after gastroplasty, and the lowest weight loss after AGB [8, 9, 16]. In addition, all systematic reviews and meta-analyses comparing %EWL achieved with different types of bariatric procedures have reported higher %EWL following RYGB compared with AGB [5, 8, 9, 17–19] or SG [17, 20]. The finding of a mean weight loss of 46.2 kg at 12 months after RYGB in the current study was consistent with metaanalyses which reported mean weight losses from 43.5 to 47.1 kg at 6 to 36 months of follow-up [8, 9, 16]. However, our finding of only 21.4 kg mean weight loss at 12 months after AGB is considerably less than the weight losses of 27.4 to 32.4 kg reported in the same meta-analyses (Table 4) and the mean weight loss of 42.9 kg at 3 years of postoperative follow-up reported in a meta-analysis of AGB studies [21]. These differences could be due in part to varying durations of follow-up among the studies because we saw continuing weight loss with increasing time after AGB, with a mean
weight loss of 25.5 kg in a smaller population that had 24 months of follow-up. Other contributing factors to this difference could be differences in inclusion/exclusion criteria, such as starting BMI and/or comorbid conditions. Finally, greater levels of support, follow-up, or band adjustment in smaller clinical trial groups or different standards of care at the time of the earlier studies vs. the current study may be likely contributors to the differences in mean weight loss. Our analysis found that baseline weight accounted for 18.5 % of the total variability in 12-month absolute weight loss outcomes, while factors including age, race, sex, height, diabetes, number of medications, and the log of days after surgery (within the 12-month window of 9 to 15 months) accounted for only 2.5 % of the variability. By type of bariatric procedure, preoperative weight explained 45.1 and 45.5 % of the variability for RYGB and SG, respectively, but only 14.1 % of the variability for AGB. The importance of preoperative weight for outcomes after RYGB is consistent with the findings from previous studies
Table 4 Comparison of absolute weight loss outcomes in this study and published meta-analyses Study
Follow-up Mean absolute weight loss from baseline (months) Total proceduresa RYGB AGB
Current study 12
Buchwald et al. [8]
6-36
Maggard et al. [16]
12
Buchwald et al. [9]
