Introduction

Dr. George Bray, pioneer of obesity research, aptly titled his editorial that accompanied two publications associating long-term mortality with bariatric surgery [1, 2], “The Missing Link—Lose Weight, Live Longer” [3]. Indeed, it appears unclear whether significant and sustained intentional weight loss results in improved mortality when equivocal results from observational studies demonstrate an increase [4, 5], decrease [6], or no change [7] in mortality following nonsurgical, voluntary weight loss [8]. While increasing body mass index (BMI) has been plainly associated with increasing disease risk [9–14] and long-term mortality [15], with an even a greater mortality risk when BMI is equal to or greater than 35 kg/m2 compared to class I obesity (BMI equal to 30–34.9 kg/m2) [8, 16–18], most clinical studies (observational in nature) have failed to show decreased mortality for cardiovascular disease subsequent to nonsurgical weight reduction [19] and have, in fact, demonstrated an increased risk following nonsurgical weight loss [19–22]—even if participants were overweight or obese at pre-weight loss intervention [19, 21, 23]. A similar unclear outcome exists for the association of cancer risk and voluntary weight loss. Although multiple studies have demonstrated cancer risk to be positively associated with BMI [24–27], studies linked to voluntary weight loss have reported reduced cancer risk (self-reported, voluntary weight loss) [19, 28, 29] as well as increased cancer risk [4, 19, 30].

Speculation for this apparent paradox (i.e., voluntary, nonsurgical weight loss resulting in equivocal mortality improvement) has centered on a number of methodological issues such as bias and confounding [8, 9, 19, 31]. The task of accurately assigning whether the study participant’s weight loss was voluntary has been problematic, especially in observational studies [8, 19]. The inclusion of subjects with occult disease, tobacco use, or patient’s whose weight loss therapy has been recommended for disease management such as diabetes has also been recognized as possible confounders to truly assessing intentional weight loss outcomes [31]. Perhaps the primary factor that has limited large population intentional weight loss studies (nonsurgical) has been the challenge of participants achieving and sustaining clinically meaningful weight loss over an extended period of time [8, 31–33]. Nonsurgical treatment of obesity, especially for the severely obese (i.e., BMI ≥ 35 kg/m2), has typically been unsuccessful in achieving substantial and sustained weight reduction [9, 34–36], with traditional approaches to weight loss such as behavioral counseling, physical activity, and diet and pharmacological therapy now considered inadequate for treating severe obesity [37–39]. Further, intense medical therapy employed in recent randomized control trials has resulted in only a 5 % weight loss in severely obese participants at 1 to 2 years follow-up [40–42]. Lifestyle-based therapy employed for diabetes prevention has not demonstrated significant prevention of events related to cardiovascular disease over a 10- to 20-year follow-up period [19, 43, 44], and lifestyle therapy plus weight loss medication have not resulted in reduced cardiovascular disease endpoints [19, 45, 46]. In conclusion, quoting Dr. Lars Sjöström, “Taken together, trials of nonsurgical weight loss in obese participants have failed to demonstrate a benefit of reduced mortality or decreased cardiovascular disease event rates” [19].

The only medical intervention (to date) that has demonstrated substantial effects on weight loss in the severely obese population, short- and long-term, is bariatric surgery [1, 47]. Prospectively controlled cohort studies [39, 48], randomized control trials [40, 41], and meta-analysis findings [49] have reported percentages of initial weight loss among severely obese patients undergoing gastric bypass or sleeve gastrectomy procedures to be 25–35 % at 1 to 2 years of follow-up and 25–28 % at 6 to 10 years following surgery. In addition, limited long-term studies have demonstrated clinically relevant improvements in obesity-related conditions such as T2DM remission and improved blood pressure and lipids, with some degree of recurrence of these comorbidities over time [39, 48, 50–53]. Therefore, in contrast to conventional weight loss therapy, bariatric surgery represents a voluntary/intentional intervention resulting in a significant body weight reduction percentage which has been sustained for a meaningful period of time. These factors are likely the primary reasons bariatric surgery has facilitated successful research associating intentional weight loss and long-term mortality, the key that eventually provided the missing link—lose weight and live longer.

The remainder of this review highlights the number of longer-term mortality post-bariatric surgical studies. Longer-term is defined as at least 2 years of postoperative follow-up after bariatric surgery. Variation in methodological aspects of long-term mortality studies are discussed as well as physiological mechanisms that may explain reduced longer-term mortality following bariatric surgery. Increased risk for externally caused deaths following bariatric surgery such as suicides, accidents and poisonings of undetermined intent, mortality risk prediction, and decision modeling to predict life expectancy (LE) are briefly discussed in this review. Although meaningful effort has been undertaken to perform an exhaustive review of past and current manuscripts relating long-term mortality following bariatric surgery, one or more studies may have been inadvertently omitted. If this is the case, apologies are extended to authors of such scientific reports.

Methodological Characteristics

Since the first published study by MacDonald et al. in 1997 [54] reporting long-term mortality associated with bariatric surgery, there have been at least 28 studies reporting mortality rates at least 2 years after bariatric surgery [1, 2, 54–79]. While the majority of these papers relate to all-cause mortality, deaths caused by cardiovascular disease(s), cancer, and suicide are frequently highlighted. As expected, these studies have included a wide variety of methodological approaches. However, common to all but one of these studies has been the use of a retrospective cohort. Only the Swedish Obesity Subjects study [1] has followed a prospective cohort design. Highlighted in Table 1 is a brief description of these bariatric surgical studies, listed in order by the year in which the research was published.

Table 1 Brief description of bariatric surgical studies
Full size table

As detailed in Table 1, bariatric surgical cohorts have ranged in number from 154 [54] to 18,972 patients [58]. Inclusion of control cohorts has been employed by 71 % of the studies (n = 20) with one study comparing the clinical data of patients 2 years prior to bariatric surgery to their post-surgical data [78]. The only study to have actual recruitment of control participants was the SOS study, which included 2037 control patients matched on a number of variables to the 2010 surgical patients [1]. The source of nonsurgical control subjects have included: 1) mortality rates of the general population (state and national) [62, 66–68, 76]; 2) non-bariatric surgical patient data extracted from hospitals, including Veterans Affairs medical centers, health clinics, and bariatric surgical centers; 3) national diabetes registry [79]; and 4) driver license applicants who had not had bariatric surgery [2]. Briefly, other characteristics of interest regarding the selection of control cohorts have included: 1) severely obese subjects without bariatric surgery; 2) control patients with similar cardiovascular events [71] or hospitalizations [73]; 3) severely obese patients seeking bariatric surgery but did not undergo the surgery [54, 60]; severely obese (≥35 kg/m2) male Medicare patients [66]; patients who were diagnosed with T2DM at the time of matching to bairatirc surgical patients [79]; and 5) veterans, representing a generally higher health-risk and predominantly male [65, 69, 77]. Methodological selection of severely obese controls from hospitals, clinics, and medical centers has generally included the use of CPT codes corresponding to obesity or morbid obesity. Mortality rates of the general population have been used in at least six studies [62, 66–68, 76]. In addition to inclusion of the general population mortality rates, two of these studies also involved obese population-based controls [66, 76]. Telem et al. obtained from the Center for Disease Control and Prevention the actuarial mortality rates for the state of New York, and in addition to general mortality rates, they assumed one-third of New York to be obese, estimating the hazard ratio (HR) of the obese population compared to the general population [76, 80]. To obtain both general- and obese-population mortality rates for control comparisons, Marsk et al. accessed data from the Military Service Conscription Registry which included records of height and weight (i.e., BMI) and conscription dates [66]. One motivation for using general population-based control cohorts as opposed to hospital patient-obtained controls is that the hospital patients may naturally have a greater morbidity profile when compared to the general population, possibly introducing a bias (referred to as the Berksonian bias [81]) [62]. Accessing the Swedish National Diabetes Registry, Eliasson et al. matched only bariatric surgical patients who were T2DM prior to gastric bypass surgery with patients who were diagnosed with T2DM at the time of matching [79].

Of the long-term mortality reports, only the study reported by Marsk et al. [66] included only male surgical patients. Two unique study aspects were found in the studies by Johnson et al. [71] and Scott et al. [73]. To explore the safety and long-term survival of bariatric surgical patients who have diagnosed cardiovascular disease prior to their bariatric surgery, Johnson et al. retrospectively compared all-cause mortality of 349 bariatric patients to 903 morbidly obese surgical controls (i.e., non-bariatric surgery), where all patients (bariatric and controls) had a history of a cardiovascular event (myocardial infarction or stroke, which ever occurred first) prior to their bariatric or non-bariatric surgeries [71]. Scott et al. compared long-term cardiovascular events and all-cause mortality of post-bariatric surgical patients to patients who had undergone surgery for either orthopedic- or gastrointestinal-related procedures [73]. As previously indicated, Eliasson et al. matched 6132 RYGB patients with T2DM patients prior to surgery with 6132 control patients diagnosed with T2DM at the time of matching [79]. The first reported long-term mortality following bariatric surgery of MacDonald et al. was limited to type 2 diabetic patients who were seeking bariatric surgery [54]. As highlighted in Table 1, the types of bariatric surgical procedures employed in long-term mortality studies have primarily focused on Roux-en-Y gastric bypass, but have also incorporated gastric banding, biliopancreatic diversion, gastrojejunostomy, and sleeve. Finally, while all but one of the studies have reported on all-cause mortality (study by Tindle et al. [67] reported only on suicide-related mortality), many of the studies have also reported on deaths related to other causes such as cardiovascular disease, cancer, and suicides. Matching-related methods for mortality studies also included a variety of schemes. Due to the prospective nature of the SOS study, Sjöström et al. employed 18 variables to match surgical and nonsurgical participants [1, 82]. The Utah mortality study matched the year of surgery, date-of-birth, gender, BMI, and surgical date of the bariatric surgical patient to driver license applicants using the date of surgery to match with the year that the driver license was obtained [2]. As detailed in Table 1, matching techniques have included group matching with various control group schemes, propensity matching in two of the reported studies [69, 75], and matching by diabetic [54] or prior-to-surgery cardiovascular events [71] in bariatric and non-bariatric participants.

Long-Term Mortality Findings

Of particular interest is the fact that of all studies of long-term mortality following bariatric surgery where bariatric surgical patients have been compared to severely obese control subjects, the bariatric group has a reduced mortality when compared to the non-bariatric control group. The range of reduction in mortality rates of the obese patients in the bariatric surgery groups when compared to the obese control groups not undergoing surgery varied from 24 to 89 %. Obese patients undergoing bariatric surgery reduced their mortality rate or risk of death by approximately 51 %. Although Maciejewski et al. reported a 36 % reduction in mortality among veterans undergoing bariatric surgery compared to non-operated veteran patients (HR: 0.65, 95 % CI, 0.51–0.80), the difference in mortality between bariatric and control groups was no longer significant (HR: 0.83, 95 % CI, 0.64–1.39) when propensity matching was applied (1694 propensity matched patients). Four years after this report study by Maciejewski et al., Arterburn et al. expanded upon this previous research and matched (non-propensity) 2500 post-bariatric surgical veterans with 7462 non-operated veterans and reported a reduced mortality among the bariatric group of 55 % (HR, 0.45 95 % CI, 0.36–0.5) 1 to 5 years post-surgery and 53 % (HR, 0.47, 95 % CI, 0.39–0.58) for 5 to 14 years post-surgery when compared to obese controls [77].

The three studies investigating surgical and control patients with diagnosed pre-surgical diabetes [54; 79] or a major cardiovascular event [71] reported reduced mortality in the pre-surgical diabetic patients when compared to the severely obese diabetic controls (HR, 0.60 (surgical mortality rate 9 % vs. nonsurgical rate 28 %)) [54] (HR, 0.60 (surgical mortality rate 9 % vs. nonsurgical rate 28 %) [54] and HR, 0.42 (95% CI, 0.30-0.57; p<0.0001) [79] and an equally impressive reduction in mortality among the pre-surgical cardiovascular event patients compared to the previous cardiovascular event obese controls (HR, 0.60 (95 % CI, 0.36–0.99) [71]. As previously cited, Scott et al. also focused on disease-related mortality by comparing patients who had undergone orthopedic- or gastrointestinal-related surgeries to postoperative bariatric surgical patients. Following appropriate adjustments, the reported HR was 0.72 (95 % CI, 0.58–0.89) and 0.48 (95 % CI, 0.39–0.61) comparing bariatric to orthopedic and gastrointestinal surgical patients, respectively. With the exception of one study [76], when longer-term mortality of post-bariatric surgical patients has been compared with mortality rates of the general population, there is no significant long-term mortality reduction in the bariatric vs. control subjects. These studies have reported that the bariatric surgery cohort has a greater mortality when compared to the mortality rates of the general population [62, 66–68], with a HR of 1.4 to 1.5 when comparing the bariatric surgical group to obese control groups. The exception is the recently reported study by Telem et al. [76] where long-term observed mortality rate of bariatric surgical patients in New York state were reported to be 1.5 % compared to the predicted general population mortality rate (New York) of 2.1 % (p = 0.005).

Systematic Review and Meta-Analysis

In response to interest in the association of long-term mortality and bariatric surgery, systematic reviews and meta-analysis studies have more recently been reported [83, 84], as well as a review commentary [85] and a systematic review of completed suicide after bariatric surgery [86]. Pontiroli and Morabito, reviewing long-term mortality related to gastric banding and gastric bypass, conducted a meta-analysis of eight mortality studies whose publication year ranged from 1997 to 2007 [1, 2, 54, 55, 57, 59–61]. They reported reduced odds of all-cause mortality (odds ratio [OR], 0.55, 95 % CI, 0.49–0.63), of cardiovascular mortality (OR, 0.58, 95 % CI, 0.46–0.73) and of all-cause mortality (OR: 0.70, 95 % CI, 0.59–0.84) [84]. A systematic review and meta-analysis reported by Kwok et al. [83] included 14 bariatric surgery-related mortality studies [1, 2, 39, 52, 53, 55, 57, 59–61, 69, 70, 73, 87], two of which did not appear to be long-term in nature [53, 87] (assuming long-term to be mortality occurring at least 2 years post-surgery). The reduction in mortality among bariatric surgical patients compared to non-operated controls was OR, 0.48, 95 % CI, 0.35–0.64. In addition, they reported mortality reduction in myocardial infarction (OR, 0.46, 95 % CI, 0.30–0.69) and stroke (OR, 0.49, 95 % CI, 0.32–0.75) bariatric surgical patients compared to nonsurgical controls [83].

Externally Caused Deaths

One of the early reported long-term mortality studies (2007) reported a significant increase in externally caused deaths [2]. Externally caused deaths are defined as deaths such as suicides, accidents, poisonings of undetermined intent, and homicides. Published the same year as the aforementioned Utah study, Omalu et al. reported on death rates and causes of death following bariatric surgery among Pennsylvania residents [62]. Although cardiovascular-related deaths were the primary cause of long-term mortality, the authors noted a meaningful number of suicides and drug overdose-related deaths (45 deaths from 16,683 bariatric operations). Following this study reported by Omalu et al., Tindle et al. published the results of long-term follow-up of bariatric surgery and its association with risk of suicide [67]. Over a 10-year period (1995–2004), Tindle et al. reported “a substantial excess of suicides among all patients who had bariatric surgery in Pennsylvania” [67]. The suicide rate for post-bariatric surgical men was reported to be 13.7/10,000 compared to the Pennsylvania suicide rate of 2.5/10,000 and for post-bariatric surgical women the suicide rate was 5.2/10,000, compared to the Pennsylvania suicide rate of 0.6/10,000 [67]. Peterhänsel et al., further pursuing this topic, published a systematic review of completed suicide following bariatric surgery [86]. From a total of 36 bariatric surgery studies where completed suicide was reported, the authors used data from 28 of these studies to estimate suicide rate of 4.1/10,000 person-years (95 % CI, 3.2–5.1/10,000 person-years) [86], a rate higher than the general population. Although the reason for associated increased suicide and bariatric surgery is not clearly understood, there is an obvious need to better understand this important risk.

Life Expectancy

We identified three papers that estimated life expectancy (LE) gains or losses associated with bariatric surgery [88, 89, 90•]. The overall findings suggest that bariatric surgery is associated with increased life expectancy. Yet, there are several inconsistencies between recently-published and previous findings, suggesting a need to address some of the methodological inconsistencies and limitations that plausibly explain these inconsistencies. Nevertheless, the direct interpretability of gains/losses in life expectancy as an outcome measure, especially when reporting the effectiveness of bariatric surgery, is beneficial in a clinical setting. Many earlier studies of mortality risk associated with bariatric surgery have reported HRs as outcomes instead of using life expectancy. Correct interpretation of HRs is often difficult to communicate outside scientific community.

A recent paper by Schauer et al. (2015) estimated bariatric-associated life expectancies in obese patients with diabetes [90•]. They estimated changes in LE by estimating the LE differences of an obese person with diabetes undergoing surgery vs. a similar person not undergoing surgery. Gains of 6.7 and 6.5 years for a 45-year-old woman and man both with a BMI of 45 and no history of hypertension, coronary artery diseases (CAD), or congestive heart failure (CHF) were estimated, respectively. The absolute gains in LE for a 45-year-old woman and man continued to be as high as 6.7 and 5.4 years even when the patient had a history of hypertension, CAD, and CHF. The LE of a bariatric surgery patient with a history of hypertension, CAD, and CHF, however, remained shorter compared to a bariatric surgery patient with no history of these diseases. For example, the estimated LE of a 45-year-old woman undergoing surgery with no history of hypertension, CAD, or CHF was 38.4 years vs. only 22.3 years when the patient had a history of these diseases. Hence, while the absolute LE gains associated with bariatric surgery were nearly the same for patients with and without history of hypertension, CHF and CAD; the relative LE gains were higher for patients with history of these additional comorbidities. Another interpretation of the shorter LE of an obese diabetic bariatric surgery patient with history of additional comorbidities is that the surgery is unable to eliminate the excess mortality burden of the additional comorbidities when compared to an obese diabetic patient who does not have additional comorbidities.

The authors also reported that the gains in LE decreased as the BMI increased, and after the BMI of 62, a nonsurgical option was expected to achieve greater LE. Some of these findings of Schauer et al. 2015 [90•] seem inconsistent with their own findings published in 2010 [89]. In 2010, Schauer and colleagues had inferred that higher BMIs were associated with greater gains in LE [89]. Further, the magnitude of the LE gains estimated from their previous paper in 2010 was smaller (2.95 years for a 42-year-old woman with BMI of 45 and 2.57 years of gains for a 44-year-old man with BMI of 45). Interestingly, Schauer et al.’s 2010 findings are similar to findings from Pope et al. (2006) [88]. Pope and colleagues estimated bariatric surgery-associated LE gains of 2.6 years and 3.3 for a 40-year-old woman and man, respectively, with a BMI of 40.

Findings from all of these three papers were based on the Markov-based state transition models. In a Markov state transition, modeling framework outcomes or health states for cohorts of patients are simulated over a fixed period of time using at least two more data sources. In all of these papers where the mortality is the final health state, simulation ends when all the patients in the simulation have died. The different pieces of information used to develop these simulations include: 1) efficacy of bariatric surgery; 2) BMI-mortality association; and 3) the short-term 30-day mortality risk associated with bariatric surgery. Some of the inconsistencies in the findings may be attributed to methodological differences. For example, Schauer et al. 2015 [90•] estimated the efficacy of bariatric surgery on obese diabetic patients using data from three HMO research network sites, while in 2010, they estimated the efficacy based on findings of Adams et al. (2007) [2]. This led to at least two analytical differences. First, unlike their 2015 efficacy estimate, the estimated efficacy of bariatric surgery was based on an analysis that did not stratify for diabetic status of the patients. Second, the average length of follow-up while estimating the efficacy for the 2010 paper was 7.1 years while it was only 2 years for the 2015 paper. In addition to pursuing future analyses that may clarify some of these analytical inconsistencies, there may be opportunities to further refine these models and pursue more advanced/direct approaches to estimate LE gains [91, 92].

Conclusion and Research Gaps

Based upon multiple studies associating long-term mortality following bariatric surgery, the following findings have emerged. Bariatric surgical patients have: 1) significantly reduced long-term all-cause mortality when compared to severely obese non-bariatric surgical control groups; 2) greater mortality when compared to the general population, with the exception of one study; 3) reduced cardiovascular-, stroke-, and cancer-caused mortality when compared to severely obese non-operated controls; and 4) increased risk for externally caused death such as suicide.

The exact causal relationship between bariatric surgery and extended mortality when compared to the severely obese, non-bariatric surgery population is not known. However, multiple studies have reported on the long-term benefit/reduction of comorbidities following bariatric surgery [19, 39, 50, 93–99, 100•]. The link between comorbidity reduction in cardiovascular events and long-term mortality, for example, has been hypothesized and possible physiologic mechanisms have been proposed [99]. The degree to which these mechanisms are related to the significant and sustained weight loss achieved through bariatric surgery or to non-weight reduction effects are not clearly understood. Future investigation into the long-term weight loss and clinical outcome durability of bariatric surgery in association with long-term mortality are likely to provide additional understanding of the link between bariatric surgery and reduced long-term mortality. Finally, listed below are a number of gaps related to this topic for which research findings would be beneficial.

  • Is duration of severe obesity prior to surgery more important than age at surgery in relation to post-surgical mortality?

  • Do long-term mortality outcomes vary in relation to bariatric surgical procedures?

  • Does improved mortality equal improved long-term quality of life?

  • Can robust animal models (or well-designed human studies) help disentangle energy-restricted weight loss from physiologic consequences of gastric bypass surgery on reduced mortality?

  • Why do post-surgical bariatric surgery patients have higher mortality from suicide, accidents, and poisonings? (Are current pre-surgical psychological evaluations sufficient to exclude at-risk patients? Are post-surgical counseling strategies needed?)