Patient exclusions are shown in the boxes on the right.
eFigure. Age and BMI Interaction (N = 726 RYGB Patients)
Customize your JAMA Network experience by selecting one or more topics from the list below.
Wood GC, Benotti PN, Lee CJ, et al. Evaluation of the Association Between Preoperative Clinical Factors and Long-term Weight Loss After Roux-en-Y Gastric Bypass. JAMA Surg. 2016;151(11):1056–1062. doi:10.1001/jamasurg.2016.2334
Which preoperative clinical factors are associated with long-term postoperative weight loss after Roux-en-Y gastric bypass (RYGB)?
This retrospective analysis of 726 RYGB patients 7 or more years after surgery found that preoperative insulin use, history of smoking, and use of 12 or more medications were significantly associated with greater long-term postoperative weight loss. Preoperative hyperlipidemia, older age, and higher body mass index were significantly associated with poorer long-term postoperative weight loss.
Preoperative identification of patients at higher risk for long-term suboptimal outcomes can help to guide clinical care, and identification of potentially modifiable risk factors has the potential to improve surgical outcomes.
Weight loss after bariatric surgery varies, yet preoperative clinical factors associated with long-term suboptimal outcomes are not well understood.
To evaluate the association between preoperative clinical factors and long-term weight loss after Roux-en-Y gastric bypass (RYGB).
Design, Setting, and Participants
From June 2001 to September 2007, this retrospective cohort study followed up RYGB patients before surgery to 7 to 12 years after surgery. The setting was a large rural integrated health system. Of 1033 eligible RYGB patients who consented to participate in longitudinal research and completed surgery before October 2007, a total of 726 (70.3%) had a weight entered in the electronic medical record 7 or more years after surgery and were included in the analyses after exclusions for pregnancy and mortality. Date of the long-term weight measurement was recorded between August 2010 and January 2016.
Main Outcomes and Measures
The primary outcome was percentage weight loss (%WL) at 7 to 12 years after surgery. Preoperative clinical factors (>200) extracted from the electronic medical record included medications, comorbidities, laboratory test results, and demographics, among others.
Among the 726 study participants, 83.1% (n = 603) were female and 97.4% (n = 707) were of white race, with a mean (SD) preoperative body mass index (calculated as weight in kilograms divided by height in meters squared) of 47.5 (7.4). From the time of surgery to long-term follow-up (median, 9.3 postoperative years), the mean (SD) %WL was 22.5% (13.1%). Preoperative insulin use, history of smoking, and use of 12 or more medications before surgery were associated with greater long-term postoperative %WL (6.8%, 2.8%, and 3.1%, respectively). Preoperative hyperlipidemia, older age, and higher body mass index were associated with poorer long-term postoperative %WL (−2.8%, −8.8%, and −4.1%, respectively).
Conclusions and Relevance
Few preoperative clinical factors associated with long-term weight loss after RYGB were identified. Preoperative insulin use was strongly associated with better long-term %WL, while preoperative hyperlipidemia, higher body mass index, and older age were associated with poorer %WL. Our findings provide additional insight into preoperative identification of RYGB patients at higher risk for long-term suboptimal outcomes.
Bariatric surgery patients are expected to lose 30% to 40% of their body weight and up to 67% of the excess body weight, depending on the type of surgery.1-4 However, weight loss trajectories after bariatric surgery are not uniform, and some patients do not achieve or are unable to maintain expected weight losses.3,5,6
Courcoulas et al5 recently identified 5 weight loss trajectories among Roux-en-Y gastric bypass (RYGB) patients during the first 3 postoperative years. In their study, approximately 23% of patients did not achieve or did not maintain greater than 25% of the body weight loss at postoperative year 3. Coleman and colleagues3 also evaluated 3-year weight outcomes in a large cohort and found high variability in RYGB weight loss at postoperative year 3.
In addition, longer-term studies7-9 indicate that weight loss after surgery may be more challenging for some patients, although study sample sizes were smaller than cohorts in the short-term studies.3,5 In the Swedish Obese Subjects (SOS) trial,7 RYGB patients maintained approximately 25% of their body weight loss at postoperative year 10 after initially losing 38% at year 1; however, only 34 RYGB patients were included in the study at year 10. de Hollanda et al9 followed up RYGB and sleeve gastrectomy patients for a median of 4.5 years after surgery and found that 24% of patients failed to achieve or maintain 50% excess weight loss. Adams and colleagues8 reported that at 6 years after surgery only 76% of RYGB patients maintained 20% or more of their total body weight loss. Finally, a recent study10 found the mean percentage excess weight loss (%EWL) to be 58.9% at 10 to 13 years after RYGB, although only 45.6% (134 of 294) of eligible study patients were included for follow-up.
Collectively, these studies indicate that patients may respond differently to the same surgical intervention for severe obesity and find that most will regain weight. Identifying patient characteristics associated with suboptimal weight loss has the potential to improve patient and procedure selection, as well as invite adjuvant perioperative treatments and interventions designed to modify these risk factors. Several preoperative clinical factors associated with suboptimal weight loss after RYGB are already well established for the first 5 postoperative years. Consistently, higher preoperative body mass index (BMI) and type 2 diabetes are associated with poorer postoperative weight loss.9,11 A comprehensive review12 of preoperative clinical factors also identified preoperative weight loss as positively associated with postoperative weight loss in some studies. In a previous study11 of preoperative clinical factors in the short term, our group identified age (>50 years), higher preoperative BMI, use of diabetes medication, liver fibrosis, and iron deficiency, among other factors, as being associated with poorer weight loss nadir at approximately 36 months after surgery.
However, little is known about how preoperative clinical factors relate to longer-term weight loss trajectories, particularly after patients have reached their weight loss nadir, making informed discussions between health care professionals and surgical candidates about long-term outcomes challenging. The primary objective of the present study was to evaluate the association between a range of preoperative clinical factors and long-term weight loss (7-12 years after surgery) in a large cohort (N = 726) of RYGB patients using weight loss metrics (percentage weight loss [%WL] and %EWL). In addition, we aimed to determine whether preoperative clinical factors previously identified as associated with short-term weight change11 also relate to long-term weight change using the electronic medical record (EMR) to obtain data from a real-world patient population. Finally, we sought to address the limited availability in the current literature regarding long-term health outcomes after RYGB by evaluating changes in comorbidities before surgery to 7 to 12 years after surgery for diabetes, hypertension, and hyperlipidemia.
From June 2001 to September 2007, this retrospective study followed up RYGB patients before surgery to 7 to 12 years after surgery. All patients who entered the bariatric surgery program at a large integrated rural health system (Geisinger Health System, Danville, Pennsylvania) were offered participation in an ongoing research program focused on bariatric surgery outcomes. Participants were recruited at their preoperative bariatric surgery clinic appointments, and clearance for primary bariatric surgery was considered the study inclusion criterion. The bariatric surgery program consisted of a 6-month preoperative program that included a diet-induced weight loss target of 10% of body weight. Patients were scheduled for follow-up visits at 1 week, 2 weeks, 2 months, 5 months, 8 months, and 12 months after RYGB and then yearly thereafter. All participants provided written informed consent. This study was approved by the Geisinger Institutional Research Review Board. Date of the long-term weight measurement was recorded between August 2010 and January 2016.
Patients who underwent RYGB gastric bypass surgery before October 2007 were potential candidates (n = 1145) for inclusion in the present retrospective analysis. All potential participants in this study completed at least 8 postoperative years at the time of data extraction and had at least 1 weight measurement in the EMR at postoperative year 7 or beyond. Of these 1145 possible participants, we excluded 55 who died within 7 years after surgery and another 57 who had a revision surgery within 7 years after surgery, leaving 1033 available for review for the presence of a long-term weight measurement. Among these 1033 patients, 275 were excluded because they did not have any weight measurement 7 or more years after surgery, and another 32 were excluded because of pregnancy at the time of their follow-up weight measurements at 7 years or beyond. These exclusions left 726 participants (70.3% of the 1033 eligible) who had a long-term weight measurement at least 7 postoperative years after surgery and were included in analyses (Figure 1).
Participants with a long-term weight measurement were compared with the 275 participants without a long-term weight measurement. Participants with a long-term weight measurement were more likely to be female (83.1% [603 of 726] vs 76.4% [210 of 275], P = .02) and had a lower mean BMI (calculated as weight in kilograms divided by height in meters squared) (47.5 vs 48.6, P = .04) but were similar in age (45.2 vs 44.6 years, P = .46).
For the 726 participants included, we used their most recent weight measurement from the EMR for the analysis (median, 9.3 postoperative years; range, 8.0-12.0 postoperative years). For this study, we calculated 2 weight outcomes for comparison of preoperative clinical factors. First was %EWL. This method has been used by our group in a prior publication11 of short-term clinical predictors of weight change after bariatric surgery. To calculate the total amount of excess weight, this method compares a presurgery measure (weight at surgery) with an ideal body weight (given a height, the weight at which BMI is 25). The %EWL is recorded for each postoperative weight measurement. When using this measure, we adjusted for BMI at the time of surgery. Second was %WL. Weight at the time of surgery was used for total body weight. The %WL is recorded for each postoperative weight measurement.
Data were obtained from several clinical sources and entered into a Geisinger Institutional Research Review Board–approved database of RYGB patients enrolled in a longitudinal obesity research program. Clinical data were extracted from the EMR (EpicCare; Epic Systems Corporation) and exported to a data warehouse, Geisinger Health System’s Clinical Decision Intelligence System. The database also included results of intraoperative liver biopsy histopathological analysis and blood samples. This database and the methods for data extraction were described previously.11,13 Additional self-administrated survey data were obtained during the preoperative period. Preoperative clinical variables (n = 206) evaluated in this analysis were previously published11 and spanned medication use (92 classes), comorbidity diagnoses (50 International Classification of Diseases, Ninth Revision, codes), and baseline laboratory test results (45 laboratories), along with 19 other factors (including demographics, anthropometrics, smoking and alcohol use history, surgical access, preoperative weight loss, time since initial consultation, dietary and psychiatric status, number of medications, number of diagnoses, and liver biopsy histopathologic findings).
The primary study outcome was weight change up to 10 years after surgery. This variable was expressed as %EWL and %WL after adjusting for baseline BMI.14
Descriptive statistics, such as means (standard deviations) and percentages, were used to describe the study population. The distributions of %EWL and %WL were graphically presented using histograms or box-and-whisker plots. Clinical variables associated with each weight loss outcome were identified using Pearson product moment correlation (continuous variables), analysis of variance (categorical variables with ≥3 groups), linear trend test (ordinal variables), and 2-sample t test (categorical variables with 2 groups). Multiple linear regression models were built for each weight loss outcome by considering the subset with bivariate P < .10. Continuous covariates were checked for nonlinearity by categorizing the data into groups (eg, using quartiles of the distribution or scientifically valid cutoff values). Effect modifiers were evaluated by testing for significance of interaction terms between age, sex, and BMI with the other clinical variables.
Each of the clinical variables that were significant in a univariate analysis was considered for entry into multiple linear regression models (age, sex, and BMI were included in the model regardless of significance level). The model was built with the goal of identifying a set of clinical variables that independently predict weight loss. The model variables were selected by adding factors after considering correlations with variables in the model, strength of associations, clinical intuition, and, in the case of categorical predictor variables, frequency of occurrence. The model was constructed using a forward stepwise approach, starting with the variables with the strongest bivariate association and adding one variable at a time until the addition of a new variable was not significant. To examine effect modification, interactions between selected variables were considered and retained when significant. McNemar test was used to evaluate changes in comorbidity status. All tests were 2-sided, and P < .05 was considered significant. A software program (SAS, version 9.4; SAS Institute Inc) was used for statistical analysis.
The mean (SD) age of the cohort at the time of surgery was 45.2 (10.7) years, and the mean (SD) BMI was 47.5 (7.4). The cohort was 83.1% (603 of 726) female, and 97.4% (707 of 726) were of white race. Characteristics of participants at the time of surgery are listed in Table 1.
From the time of surgery to long-term follow-up (median, 9.3 postoperative years), RYGB patients had a mean (SD) %WL of 22.5% (13.1%) and a mean (SD) %EWL of 50.2% (30.8%). The distributions of %WL and %EWL are shown in Figure 2.
Preoperative clinical factors significantly associated with long-term %WL after RYGB using multivariable models are listed in Table 2. Preoperative insulin use (6.8%; 95% CI, 3.8%-9.8%; P < .001), history of smoking (2.8%; 95% CI, 0.7%-4.9%; P = .009), and use of 12 or more medications (3.1%; 95% CI, 0.9%-5.3%; P = .005) were associated with greater long-term postoperative %WL. Preoperative hyperlipidemia (−2.8%; 95% CI, −0.7% to −4.9%; P = .01), older age (per 10-year increase, −8.8%; 95% CI, −3.1% to −14.4%; P = .002), and higher BMI (per 5-point increase, −4.1%; 95% CI, −1.5% to −6.6%; P = .002) were associated with poorer long-term postoperative %WL.
However, we found an interaction between age and BMI (eFigure in the Supplement), with higher BMI in younger patients yielding poorer long-term weight loss and lower BMI in older patients also yielding poorer long-term weight loss. All preoperative clinical factors that were significant in modeling using %WL were also significant using %EWL.
The percentage of patients receiving diabetes treatment significantly (P < .001) decreased from 46.1% (335 of 726) before surgery to 18.2% (132 of 726) at follow-up (median, 9.3 postoperative years). The percentage of patients receiving hypertension treatment also significantly decreased from 65.6% (476 of 726) before surgery to 58.7% (426 of 726) at follow-up (P < .001). The percentage of patients receiving dyslipidemia treatment significantly decreased from 35.0% (254 of 726) before surgery to 27.7% (201 of 726) at long-term follow-up (P < .001).
Preoperative insulin users experienced a %WL of 24.8% to 29.2% by year 2, which they largely maintained through 10 years after surgery (Table 3). Non–insulin users experienced a %WL of 31.1% by postoperative year 2 and maintained 22.0% through postoperative year 7 and beyond. Non–insulin users before or after surgery (n = 596) lost weight (%WL of approximately 22.0%) comparable to those who used insulin before surgery but discontinued use after surgery (n = 46). Participants using insulin both before and after surgery (n = 38) experienced a %WL of 30.1% through postoperative year 7 and beyond.
The present study identified several preoperative clinical factors associated with long-term weight change after RYGB (median, 9 postoperative years). The same preoperative clinical factors were significant regardless of which weight loss outcome was used (%WL or %EWL), suggesting that either outcome measure, if preoperative BMI is properly accounted for, may be appropriate for such analyses.
Of more than 200 potential preoperative clinical factors, only 3 were associated with greater long-term weight loss, including preoperative insulin use (6.8% greater weight loss), history of smoking (2.8% greater weight loss), and use of 12 or more medications before surgery (3.1% greater weight loss). Our results indicated that specific preoperative medication use in addition to insulin (eg, tricyclic antidepressants) was significantly associated with weight loss after bariatric surgery, but we chose to categorize the number of medications in this study as a proxy for overall health. Participants taking the most medications (≥12) before surgery had better weight loss outcomes. Possible explanations for this finding are their greater interaction with health care professionals needed to manage multiple conditions or perhaps unintentional weight loss related to health conditions. Additional studies are needed to evaluate these medications individually in relation to long-term weight loss.
Few preoperative clinical factors were associated with poorer weight loss after RYGB, including a diagnosis of hyperlipidemia (−2.8% less weight loss), older age (−8.8% less weight loss for every 10 years), and higher BMI (−4.1% less weight loss for every 5 kg/m2). A previous study11 and a systematic review12 have also identified higher preoperative BMI as a risk factor for poorer outcomes after bariatric surgery. Numerous factors previously identified as correlating with short-term weight loss,11 including preoperative iron deficiency, liver fibrosis, and preoperative weight loss, did not correlate with long-term weight loss in our study. In addition, one factor (insulin use) seemed to reverse the direction of association from our group’s previous study11 of this cohort. The use of insulin before surgery was found to be strongly associated with greater weight loss in the long term. This finding was contrary to what was anticipated because preoperative use of diabetes medications was associated with poorer weight loss in the earlier postoperative period (approximately 0-36 months after surgery)11 and poorer weight loss nadir11 in this cohort. In addition, diabetes is widely reported to be associated with suboptimal weight loss after RYGB.15-19
To further explore the effect of insulin use, a post hoc analysis was conducted by examining early weight loss (2 years after surgery), long-term weight loss (≥7 years), and weight regain after stratifying by patients who never used insulin, used insulin before but not after surgery, and used insulin both before and after surgery (Table 3). Early %WL was highest in patients with no insulin use and was lowest in patients using insulin before and after surgery. However, this association was in the opposite direction for long-term %WL, and patients using insulin before and after surgery continued to lose weight (an additional %WL of 5.3%). When evaluating weight regain, patients never using insulin regained approximately 9.1% of their %WL, while patients using insulin before surgery only had minimal weight regain (1.6%). Overall, patients who used insulin both before and after surgery lost the most weight up to 10 years after surgery, and this finding requires further study.
It is unclear why preoperative insulin users seem to manifest a weight loss trajectory that deviates from the weight loss typically reported after bariatric surgery.5,11 It is possible that insulin use necessitates greater interaction with the health care system, leading to better adherence and ultimately better weight loss. There is some evidence that greater adherence to follow-up appointments is positively associated with short-term weight loss.20 However, a post hoc analysis evaluating the number of health care visits among insulin users and non–insulin users in our cohort from surgery through postoperative year 7 revealed no differences in the mean number of visits to health care professionals (46.4 visits for non–insulin users, 49.2 visits for insulin users before but not after surgery, and 54.9 visits for insulin users before and after surgery; P = .20). In addition, we evaluated the possible influence of newer weight loss–promoting diabetes medications used in conjunction with insulin on weight loss (eg, glucagon-like peptide 1 agonists), but only 2 participants in our cohort were using a combination of these medications. Finally, we performed a sensitivity analysis that excluded 5 patients believed to have type 1 diabetes (based on EMR review) given the possibility of a differing effect of type 1 vs type 2 diabetes on an individual’s postoperative weight loss and weight regain. Results of the sensitivity analysis were similar for preoperative clinical factors and insulin-specific findings. Therefore, mechanisms underlying our insulin findings remain unclear. Other explanations that we were unable to explore included the possibility that insulin users paid greater attention to and demonstrated greater adherence to dietary recommendations, as well as changes in insulin dosing related to weight loss.
Finally, RYGB was associated with significant improvements in hyperlipidemia and diabetes, which has been previously reported.10,16,21 We also found long-term improvement in hypertension, which differs from previous findings by Sjöström and colleagues.22 The differing long-term weight loss outcomes, sample sizes, and surgical procedures included may account for our divergent findings. More long-term studies of the incidence of hypertension after RYGB in large cohorts are needed to understand this potential postoperative benefit.
This study has several strengths. Our sample represents one of the largest bariatric surgery cohorts, to our knowledge, with weight outcome data available up to 12 years after surgery, as well as one of the highest long-term follow-up rates (70.3% [726 of 1033]). The long-term data were extracted from the EMR and therefore reflect how our postoperative patients actually fared in clinical practice. Using the EMR allowed for the assessment of a wide range of preoperative clinical factors, as well as detailed comorbidity information, in a large sample of patients. However, our sample was predominantly of white race; therefore, we were unable to evaluate differences by race in our cohort. A previous study3 found race/ethnicity to be significantly associated with weight loss outcomes after RYGB, and studies at more diverse sites could explore this finding in the long term.
Overall, few preoperative clinical factors were associated with weight change in the long-term postoperative course. Future studies are needed to replicate these findings, particularly surrounding insulin use. Comprehensive investigations of potential preoperative psychosocial and behavioral factors or other modifiable preoperative or early postoperative factors that may influence weight in the long term could also help to identify patients at risk for suboptimal outcomes. These results can help to guide clinical care and improve patient-directed informed consent discussions about bariatric surgery.
Accepted for Publication: May 14, 2016.
Corresponding Author: Michelle R. Lent, PhD, Geisinger Obesity Institute, Geisinger Clinic, 100 N Academy Ave, Danville, PA 17822 (email@example.com).
Published Online: August 10, 2016. doi:10.1001/jamasurg.2016.2334
Author Contributions: Mr Wood had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Study concept and design: Wood, Benotti, Mirshahi, Still.
Acquisition, analysis, or interpretation of data: Wood, Lee, Gerhard, Lent.
Drafting of the manuscript: Wood, Benotti, Lent.
Critical revision of the manuscript for important intellectual content: Wood, Lee, Mirshahi, Still, Gerhard.
Statistical analysis: Wood.
Administrative, technical, or material support: Still, Gerhard.
Study supervision: Benotti, Still, Lent.
Conflict of Interest Disclosures: None reported.
Funding/Support: This work was supported by funds from Geisinger Clinic, The Sigfried and Janet Weis Center for Research, and Geisinger Obesity Institute, as well as by grant P30DK072488 from the Nutrition Obesity Research Center and grant KL2 5KL2TR001077-02 (Dr Lee) as part of an Institute for Clinical and Translational Research grant from the National Institutes of Health/National Center for Advancing Translational Sciences.
Role of the Funder/Sponsor: The funders had no role in the design or conduct of the study, including the collection, management, analysis, and interpretation of the data; the preparation, review, or approval of the manuscript; or the decision to submit the manuscript for publication.
Create a personal account or sign in to: