Obesity poses a major challenge for modern medicine, exacerbated by the rapid global spread of Western-pattern lifestyles characterized by overnutrition and low physical activity [1
]. Among potential treatments for obesity, bariatric surgery is the single most effective, leading to rapid and durable amelioration of both obesity and its comorbidities such as type 2 diabetes (T2D) [2
]. Over the past decade, sleeve gastrectomy has become one of the most popular and effective bariatric surgical techniques, and its central mechanism of action relies upon calorie restriction [4
]. Although a growing body of evidence has demonstrated bariatric surgery’s broad efficacy across patient populations, it has also introduced uncertainty with respect to sex-specific effects [6
]. Historically, women have sought this intervention more frequently than men with 80% of bariatric surgeries performed on female patients, and this imbalance has persisted for the past decade [7
]. However, obesity incidence is nearly equal in both sexes [10
], and as the number of bariatric surgeries continues to accelerate [11
], it is increasingly urgent to understand whether and how differential responses may occur. Previous studies have yielded conflicting results: Some find significant differences in particular parameters while others either fail to replicate these results or find opposite response profiles [2
]. Most such studies examine restricted sets of metabolic or anthropometric parameters, often, for example, measuring only the concentration of glycated hemoglobin (HbA1c) as a metric of glycemic health. However, potential sex-specific responses to bariatric surgery likely emerge from complex interactions among hormone signaling axes, anatomical, and physiological differences between the sexes. This implies that many traits must be simultaneously measured to understand systemic, sex-specific responses. Here, we performed deep, longitudinal phenotyping for one year on a balanced cohort of men and women to understand how they differed in response to bariatric surgery.
Of previous studies that have detected significant differences in bariatric outcomes between sexes, the majority have found that males respond more favorably than females. This difference occurs in diverse parameters, such as heart rate variability [15
], serum leptin levels [16
], fat-free mass loss [17
], and absolute and relative weight loss [15
]. However, these studies are relatively small (n ≈ 102
), and fewer but much larger studies (n ≈ 104
) have uncovered conflicting evidence showing that women improve more in cardiovascular and metabolic responses [13
]. The majority of these studies consist of retrospective or prospective cohort observations, and thus likely suffer from confounding between the sexes [23
]. Indeed, several confounding variables have been widely detected and replicated in independent patient cohorts such as the greater age, alcohol consumption, smoking history, and severity of pre-existing conditions in men seeking bariatric surgery [8
]. The greater age of men is likely an extremely important mediating variable in explaining men’s superior responses since many obesity-related morbidities covary with age [25
]. Thus, there is an outstanding need to examine sex-specific responses to bariatric surgery in an age-matched cohort.
Two primary ways to account for potential mediating variables and to control for confounders have been implemented: (1) Increased sample size, assuming that a random distribution of confounding variables is more likely to be captured in a larger cohort, and (2) adjusting estimates of response effects by explicitly modeling confounders [23
]. The first method is limited by the accuracy of its underlying assumption—in the case of sex, confounding variables may covary as strongly in large samples as in small. The second method is limited by the comprehensiveness of phenotyping so that sufficient variables must be modeled to ensure that the effects of the predictor of interest (in our case, sex) can be isolated from covariates. In evaluating sex-specific responses to bariatric surgery, an important covariate is differences in nutritional habits between men and women. These differences emerge from complex psychosocial processes, emerging in the twin trends that women show greater concern about body image [29
] and greater anxiety about bodyweight [30
]. This has been specifically confirmed in bariatric surgical candidates with women showing lower psychological well-being and more depressive symptoms related to body image and diet [14
]. This, in turn, can lead to more frequent and extreme alterations in dietary habits in women to control weight (for example, more frequent low-carbohydrate diet attempts [31
]), which may impact the effect of bariatric surgery [32
]. Assessing these dietary and nutritional habits is an important aspect of a comprehensive analysis of sex-specific responses to bariatric surgery, and we carefully track and model these habits in this study.
Finally, the mechanistic causes of sex-specific differences in bariatric surgical treatment remain controversial since most investigations of bariatric surgery’s effects have been exclusively conducted in male animal models despite the predominance of female bariatric patients [34
]. However, recent sex-inclusive studies have traced causal differences between male and female subjects to the gene-expression level, finding major differences in metabolic regulatory pathways [35
]. Together, these findings highlight the importance of understanding systemic metabolism in gauging the efficacy of bariatric surgery. We use deep metabolic phenotyping to accomplish this goal, focusing on metrics of dysglycemia. This is especially relevant in bariatric surgical candidates since T2D is a severe and prevalent comorbidity of obesity, occurring in 28% of obese patients [37
]. Bariatric surgery generally, and sleeve gastrectomy specifically, provides rapid, highly effective relief for most diabetes features with complete remission of 78% of T2D patients within two years of surgery [2
]. However, few studies examine multiple metabolic parameters that change upon bariatric surgery, how they differ between men and women, and how they affect dysglycemia status.
We answer four questions in this study: (1) How does sleeve gastrectomy affect metabolic and anthropometric parameters across sexes? (2) What is the most informative set of covariates to isolate predictor effects? (3) Do responses to bariatric surgery depend upon sex, and if so, do men or women respond better? (4) Are there sex-specific trends in dysglycemia resolution?
In this study, we evaluate sex-specific differences in the efficacy of sleeve gastrectomy to improve glucose homeostasis and anthropometric health. SG is the most frequently performed bariatric surgery worldwide, having surpassed RYGB in popularity, since its primary outcome (long-term weight loss) is equally efficacious, leads to fewer postoperative complications, and is technically less challenging and cheaper than RYGB [46
]. This has been amply demonstrated by major randomized clinical trials including both the Swiss Multicenter Bypass or Sleeve Study [50
] and the Finnish Sleeve versus Bypass Study [51
]. These showed that total and excess weight loss at five years were not significantly different between SG and RYGB and that both procedures were similarly effective for diabetes resolution, sleep apnea and quality of life improvement. However, RYGB was also shown to be marginally more effective in patients with very high BMI (>50 kg/m2
) and with poorly managed T2D, producing slightly greater weight loss and diabetes remission in these groups, respectively.
Furthermore, SG is an extremely safe surgical procedure, although not necessarily the safest bariatric procedure. For example, Chang et al. showed that adjustable gastric banding (AGB) was safest, followed by SG and RYGB (peri- and post-operative mortality rates, respectively: AGB = 0.07% and 0.21%, SG = 0.29% and 0.34% and RYGB = 0.38% and 0.72%, complications rates: AGB = 7.80%, SG = 8.90% and RYGB = 12.00% [11
]). However, in our study, the complication rate calculated as the percentage of patients who experienced post-operative bleeding, stomal stenosis, leak, vomiting, reflux, gastrointestinal symptoms, or nutritional and electrolyte abnormalities was only 5.20%. This lower complication rate might be explained by our clinic’s strict implementation of guidelines to decrease post-operative morbidity, which have been shown to greatly reduce bariatric surgical complications [52
]. For example, our patients were discharged with prescriptions for proton pump inhibitors (which reduce the risk of ulcer development and gastroesophageal reflux) so that these complications did not arise after the surgery. Taking together SG’s cost-effectiveness, procedural simplicity, and high rate of favorable outcomes, it is currently the first-line bariatric surgery, and this universality and popularity made it an ideal procedure to examine in this study.
The majority of previous studies that have sought to understand whether differences in SG efficacy are correlated with sex are based upon cohorts with unbalanced cross-sections of men and women (approximately 20% male and 80% female) [13
]. Here, on the other hand, our sample is very close to sex balance (47% male and 53% female). Although this former proportion reflects the sex ratio of patients undergoing bariatric surgery in most clinics, it has little statistical utility when the goal is to estimate sex-specific effects—it simply allows more precise estimates of uncertainty about female effects [53
]. Our balanced sample, on the other hand, eliminated bias that might arise from better estimates of uncertainty within subsamples of our cohort [54
Furthermore, most studies examine cohorts in which men’s age is greater than women’s so that age may be an important mediating and potentially confounding variable in evaluations of sex-specific effects [13
]. For example, a greater age is known to be associated with diminished weight loss after bariatric surgery [55
]. In older people, adipose tissue deposition is increased and lipid metabolism is decreased, together constituting a metabolic sink which can be compensated by increasing food intake [56
]. This increased consumption could greatly attenuate the benefit of bariatric surgery, and, if men are older, may appear to be related to sex. In this study, we decouple the potential confounding between age and sex since, although men consumed more total calories than women, they were not older. In fact, age was statistically significantly different only in a one-way non-parametric Mann–Whitney test which found that men were slightly younger than women (but no statistical difference was detected by the median test (Table 2
). Furthermore, this difference was only 3.5 years (men = 44.3, women = 47.8), which likely has little clinical relevance. Since men and women in our cohort are largely matched for age and since we control for age in statistical analyses, we have a great deal of confidence that our sex-specific results are not confounded by age. This age matching was fortuitous since we did not implement any age filter in the inclusion or exclusion criteria, and we do not attempt to speculate as to why younger men sought bariatric surgery at our center than at many others.
Confounding, in general, is a significant issue in prospective studies and especially in studies that focus on sex since a constellation of cultural habits and physiological differences are correlated to sex and gender expression [57
]. Previous studies have grappled with this issue by extensive exercises in model specification, adding and dropping terms and reporting how significance estimates change [13
]. Here, we adopted a model selection approach by reporting all possible combinations of model covariates (see Supplementary Appendices S1 and S2
), but our significance estimates were largely stable (e.g., time terms remained significant with different covariate combinations of age, diet, sex and physical activity). Although effect size magnitudes changed, their directions did not, and we base our conclusions upon the direction of change in response to bariatric surgery. Furthermore, we used formal model selection techniques (REML comparisons and likelihood ratio tests) to choose the most informative model, and this proved to be the most parameterized model for most response variables. This shows that covariates are vital to our analysis and that any inferences we draw from them must be, and were, adjusted for confounding differences between the sexes.
By accounting for this confounding, we found that although men had greater absolute anthropometric metrics like total body mass and weight, there was no statistically significant difference in BMI. This implies that scaling by gross body size may largely equalize anthropometric morbidity in men and women. On the other hand, there was no statistically significant difference in fat mass between men and women. Since women were on average smaller than men (e.g., total body mass 23.5 kg less than men), women’s baseline fat content slightly higher than men’s (52.3% versus 45.6% fat mass as a percent of total body mass, respectively). Men also showed considerably greater derangement in glucose regulation with higher index values of insulin resistance and greater glucose load throughout OGTT, though no significant difference in HbA1c. This is in line with other studies that have found that men present with greater obesity-related comorbidities when undergoing surgery [23
]. Men’s greater morbidity at baseline relative to women may account for previous findings that men improve more, since they have “further to go” when bariatric surgery normalizes deranged health axes.
To understand whether and how men and women responded differently to bariatric surgery, we tested for significant interactions between time and sex, examining both linear and quadratic time terms. A linear interaction term indicates that men increased or decreased more than women averaging over all follow-up visits, while the quadratic interaction term indicates whether there was significant acceleration or deceleration in this trend (i.e., is there significant curvature in the response). We detected 11 significant linear interactions with six also showing significant quadratic interactions (Table 3
). All of these interactions indicated that men improved more than women over the study period with greater decreases in glucose and insulin loads, improved insulin sensitivity and beta-cell function and greater body mass and fat reduction. Significant quadratic interactions indicated that the deceleration or acceleration in response was more pronounced in men than women. For example, men’s response in mean insulin concentration curved more strongly (Figure 3
a), implying that men’s downward trend may have continued more durably than the deceleration in women’s response. However, this is a tentative conclusion that requires further follow-up to determine whether men’s greater improvement is sustained over a period longer than one year.
Another way to examine responses is to scale within individual patients and evaluate means from scaled differences (Figure 2
), and this showed remarkable divergence in the response of mean insulin and glucose concentration between men and women. Indeed, men showed much greater decreases in these means, although evaluating each timepoint individually within the four-point OGTT by the GLMM revealed a sex-specific effect only at 30′ (Table 3
). This is the most proximate measurement to the introduction of the oral glucose bolus to the patient, representing the peak glucose load experienced by the patient. Following bariatric surgery, men’s ability to clear this peak glucose apparently improved more than women’s, and this translated into an overall greater decrease in mean glucose load. Interestingly, there was no significant difference in HbA1c, which gauges average blood glucose levels over three months. Together, this shows that bariatric surgery may resolve men’s acute glucose derangement (occurring on the scale of minutes after glucose exposure) more than women’s, but that their longer-term responses may not be as divergent.
There were several limitations to our study. Our patient cohort consisted entirely of Polish Caucasians, and was not large (n = 154). On the other hand, by limiting our sample to a specific national group we eliminated confounding variables that may covary with nationality. Similarly, there were relatively few patients with diabetes at baseline (n = 47, 31%), so that our analysis of sex-specific remission differences is not as robust as across-diagnosis analyses. Finally, our nutritional data are based upon self-reported food diaries analyzed by technicians and thus may suffer from the systematic biases of self-reported consumption. However, all subjects were educated by trained dieticians in how to maintain detailed, accurate, and unbiased diet diaries.