Study Overview
Objective. To determine the association between bariatric surgery and long-term mortality rates among patients with severe obesity.
Design. Retrospective cohort study.
Setting and participants . This analysis relied upon data from Veteran’s Administration (VA) patients undergoing bariatric surgery between 2000 and 2011 and a group of matched controls. For this data-only study, a waiver of informed consent was obtained. Investigators first used the VA Surgical Quality Improvement Program (SQIP) dataset to identify all bariatric surgical procedures performed at VA hospitals between 2000 and the end of 2011, excluding patients who had any evidence of body mass index (BMI) less than 35 kg/m 2 and those with certain baseline diagnoses that would be considered contraindications for surgery, as well as those who had prolonged inpatient stays immediately prior to their surgical date. No upper or lower age limits appear to have been specified, and no upper BMI limit appeared to have been set.
Once all surgical patients were identified, the investigators attempted to find a group of similar control patients who had not undergone surgery. Initially they pulled candidate matches for each surgical patient based on having the same sex, age-group (within 5 years), BMI category (35-40, 40-50, >50), diabetes status (present or absent), racial category, and VA region. From these candidates, they selected up to 3 of the closest matches on age, BMI, and a composite comorbidity score based on inpatient and outpatient claims in the year prior to surgery. The authors specified that controls could convert to surgical patients during the follow-up period, in which case their data was censored beginning with the surgical procedure. However, if a control patient underwent surgery during 2012 or 2013, censoring was not possible given that the dataset for identifying surgeries contained only procedures performed through the end of 2011.
Main outcome measures. The primary outcome of interest was time to death (any cause) beginning at the date of surgery (or baseline date for nonsurgical controls) through the end of 2013. The investigators built Cox proportional hazards models to evaluate survival using multivariable models to adjust for baseline characteristics, including those involved in the matching process, as well as others that might have differentially impacted both likelihood of undergoing surgery and mortality risk. These included marital status, insurance markers of low income or disability, and a number of comorbid medical and psychiatric diagnoses.
In addition to the main analyses, the investigators also looked for effect modification of the surgery-mortality relationship by a patient’s sex and presence or absence of diabetes at the time of surgery, as well as the time period in which their surgery was conducted, dichotomized around the year 2006. This year was selected for several reasons, including that it was the year in which a VA-wide comprehensive weight management and surgical selection program was instituted.
Results. The surgical cohort was made up of 2500 patients, and there were 7462 matched controls. The surgical and control groups were similar with respect to matched baseline characteristics, tested using standardized differences (as opposed to t test or chi-square). Mean (SD) age was 52 (8.8) years for surgical patients versus 53 (8.7) years for controls. 74% of patients in both the surgical and control groups were men, and 81% in both groups were white (ethnicity not specified). Mean (SD) baseline BMI was 47 (7.9) kg/m 2 in the surgical group and 46 (7.3) kg/m 2 for controls.
Some between-group differences were present for baseline characteristics that had not been included in the matching protocol. More surgical patients than controls had diagnoses of hypertension (80% surgical vs. 70% control), dyslipidemia (61% vs. 52%), arthritis (27% vs. 15%), depression (44% vs. 32%), GERD (35% vs.19%), and fatty liver disease (6.6% vs. 0.6%). In contrast, more control patients than surgical patients had diagnoses of alcohol abuse (6.2% in controls vs. 3.9% in surgical) and schizophrenia (4.9% vs. 1.8%). Also, although a number of different surgical types were represented in the cohort, the vast majority of procedures were classified as Roux-en-Y gastric bypasses (RYGB). 53% of the procedures were open RYGB, 21% were laparoscopic RYGB, 10% were adjustable gastric bands (AGB), and 15% were vertical sleeve gastrectomies (VSG).
Mortality was lower among surgical patients than among matched controls during a mean follow-up time of 6.9 years for surgical patients and 6.6 years for controls. Namely, the 1-, 5- and 10-year cumulative mortality rates for surgical patients were: 2.4%, 6.4%, and 13.8%. Unadjusted mortality rates for nonsurgical controls were lower initially (1.7% at 1 year), but then much higher at years 5 (10.4%), and 10 (23.9%). In multivariable Cox models, the hazard ratio (HR) for mortality in bariatric patients versus controls was nonsignificant at 1 year of follow-up. However, between 1 and 5 years after surgery (or after baseline), multivariable models showed an HR (95% CI) of 0.45 (0.36–0.56) for mortality among surgical patients versus controls. For those with more than 5 years of follow up, the HR was similar (0.47, 95% CI 0.39–0.58) for death among surgical versus control patients. The investigators found that the year during which a patient underwent surgery (before or after 2006) did impact mortality during the first postoperative year, with those who had earlier procedures (2000-2005) exhibiting a significantly higher risk of death in that year relative to non-operative controls (HR 1.66, 95% CI 1.19–2.33). No significant sex or diabetes interactions were observed for the surgery-mortality relationship in multivariable Cox models. There was no information provided as to the breakdown of cause of death within the larger “all-cause mortality” outcome.
Conclusion. Bariatric surgery was associated with significantly lower all-cause mortality among surgical patients in the VA over a 5- to 14-year follow-up period compared with a group of severely obese VA patients who did not undergo surgery.
Commentary
Rates of severe obesity (BMI ≥ 35 kg/m 2) have risen at a faster pace than those of obesity in the United States over the past decade [1], driving clinicians, patients and payers to search for effective methods of treating this condition. Bariatric surgery has emerged as the most effective treatment for severe obesity; however, the existing surgical literature is predominated by studies with short- or medium-term postoperative follow-up and homogenous participant populations containing large numbers of younger non-Hispanic white women. Research from the Swedish Obesity Study (SOS), as well as smaller US-based studies, has suggested that severely obese patients who undergo bariatric surgery have better long-term survival than their nonsurgical counterparts [2,3].Counter to this finding, a previous medium-term study utilizing data from VA hospitals did not find that surgery conferred a mortality benefit among this largely male, older, and sicker patient population [4].The current study, by the same group of investigators, attempts to update the previous finding by including more recent surgical data and a longer follow-up period, to see whether or not a survival benefit appears to emerge for VA patients undergoing bariatric surgery.
A major strength of this study was the use of a large and comprehensive clinical dataset, a strength of many studies utilizing data from the VA. The availability of clinical data such as BMI, as well as diagnostic codes and sociodemographic variables, allowed the authors to match and adjust for a number of potential confounders of the surgery-mortality relationship. Another unique feature of VA data is that members of this health care system can often be followed regardless of their location, as the unified medical record transfers between states. This is in contrast to many claims-based or single-center studies of surgery, where patients are lost to follow-up if they move or transfer insurance providers. This study clearly benefited from this aspect of VA data, with a mean postoperative follow-up period of over 5 years in both study groups, much longer than is typically observed in bariatric surgical studies, and probably a necessary feature for examining more of a rare outcome such as mortality (as opposed to comparing weight loss or diabetes remission). Another clear contribution of this study is that it focused on a group of patients not typical of bariatric cohorts—this group was slightly older and sicker, with far more men than women, and therefore at a much higher risk of mortality than the typically younger females that are part of most studies.
Although the authors did adjust for many factors when comparing the surgical and nonsurgical groups, it is possible, as with any observational study, that unmeasured confounders may have been present. Psychosocial and behavioral features that may be linked both to a person’s likelihood of undergoing surgery, and to their mortality risk are of particular concern. It is worth noting, for example, that far more patients in the nonsurgical group were identified as schizophrenic, and that the rate of schizophrenia in that severely obese group was much higher than that of the general population. This pattern may have some relationship to the weight-gain promoting effect of antipsychotic medications and the unfortunate reality that patients with severe obesity and severe mental illness may not be as well equipped to seek out surgery (or viewed as acceptable candidates) as those without severe mental illness. One possible limitation mentioned by the authors was that control group patients who underwent surgery in 2012 or 2013 would not have been recognized (and thus had their data censored in this study), possibly leading to incorrect categorization of exposure category for some amount of person-time during follow-up. In general, though, there is a low likelihood of this phenomenon impacting the findings, given both the relative infrequency of crossover observed in the cohort prior to 2011, and the relatively short amount of person-time any later crossovers would have contributed in the later years of the study.
Although codes for baseline disease states were adjusted for in multivariable analyses, the surgical patients were in general a medically sicker group at baseline than control patients. As the authors point out, if anything, this should have biased the findings in favor of seeing higher mortality rate in the surgical group, the opposite of what was found. Further strengthening the finding of a correlation between survival and surgery is the mix of procedure types included in this study. Over half of the procedures were open RYGB surgeries, with far fewer of the more modern and lower risk procedures (eg, laparoscopic RYGB) represented. Again, this mix of procedures would be expected to result in an overestimation of mortality in surgical patients relative to what might be observed if all patients had been drawn from later years of the cohort, as surgical technique evolved.
Applications for Clinical Practice
This study adds to the evidence that patients with severe obesity who undergo bariatric surgery have a lower risk of death up to 10 years after their surgery compared with patients who do not have these procedures. The findings of this work should provide encouragement, particularly for managaing older adults with more longstanding comorbidities. Those who are strongly motivated to pursue weight loss surgery, and who are deemed good candidates by bariatric teams, may add years to their lives by undergoing one of these procedures. As always, however, the quality of life experienced by patients after surgery, and a realistic expectation of the ways in which surgery will fundamentally change their lifestyle, must be a critical part of the discussion.
—Kristina Lewis, MD, MPH