Effect of bariatric surgery on cancer risk: results from an emulated target trial using population-based data

Abstract

Background

The impact of weight loss induced by bariatric surgery on cancer occurrence is controversial. To study the causal effect of bariatric surgery on cancer risk from an observational database, a target-trial emulation technique was used to mimic an RCT.

Methods

Data on patients admitted between 2010 and 2019 with a diagnosis of obesity were extracted from a national hospital discharge database. Criteria for inclusion included eligibility criteria for bariatric surgery and the absence of cancer in the 2 years following inclusion. The intervention arms were bariatric surgery versus no surgery. Outcomes were the occurrence of any cancer and obesity-related cancer; cancers not related to obesity were used as negative controls.

Results

A total of 1 140 347 patients eligible for bariatric surgery were included in the study. Some 288 604 patients (25.3 per cent) underwent bariatric surgery. A total of 48 411 cancers were identified, including 4483 in surgical patients and 43 928 among patients who did not receive bariatric surgery. Bariatric surgery was associated with a decrease in the risk of obesity-related cancer (hazard ratio (HR) 0.89, 95 per cent c.i. 0.83 to 0.95), whereas no significant effect of surgery was identified with regard to cancers not related to obesity (HR 0.96, 0.91 to 1.01).

Conclusion

When emulating a target trial from observational data, a reduction of 11 per cent in obesity-related cancer was found after bariatric surgery.

Introduction

Obesity is a fast-growing global health issue¹. In 2016, the WHO² estimated that 1.9 billion adults were overweight and, of these, 650 million were affected by obesity, which corresponds to 13 per cent of the adult population.

Obesity increases the risk of death from cardiovascular, respiratory, and liver diseases³. Moreover, a strong association between obesity and cancer has been shown for hormone-related cancers (breast, corpus uteri, and prostate), colorectal cancer, oesophageal cancer, and gastric cancer⁴. Several mechanisms have been proposed to explain the risk of obesity on cancer development, in particular insulin resistance, chronic inflammation, and alterations in the metabolism of adipokines and sex hormones⁵.

Bariatric surgery is considered as the most effective treatment for sustained weight loss and remission of obesity-related conditions⁶. Several studies^7,8 have also shown a reduction in the incidence of cancer after bariatric surgery. Nevertheless, most of these studies have methodological limitations. The control for indication bias represents a major issue, as patients who undergo bariatric surgery may not be comparable to those who do not. In addition, small sample sizes resulted in limited power to assess cancer risks according to tumour type. As a result, existing studies^7,9–11 have shown conflicting results (in particular for colorectal cancer), or counterintuitive effects with regard to the reduction of cancer risks after bariatric surgery for cancers not related to obesity.

The standard for evaluating the causal effect of an intervention is an RCT. Randomization provides groups that differ only in the intervention applied and, as such, allow identification of causal effects with limited or no bias. Emulating a target randomized trial from observational databases allows, under specific assumptions, identification of the causal effects of an intervention¹². In this study, the effect of bariatric surgery on cancer incidence was estimated by emulating a target trial using nationwide observational data.

Methods

Study design

Data were extracted from a national discharge database, the Programme De Médicalisation des Systèmes d’Information (PMSI), which collects discharge reports from all hospitals in France, irrespective of facility ownership or academic affiliation. Reporting discharge reports is mandatory and represents the basis for hospital funding. Hence, the database is comprehensive for all reimbursed surgical interventions in the country^13,14.

The target trial was designed to compare a candidate for bariatric surgery who underwent bariatric surgery with patients who did not. Three assumptions are needed to conceptualize an observational study as a randomized experiment: consistency (defined as the proper definition of treatment under comparison corresponding to the version of treatment in the data), exchangeability (the conditional probability of receiving every value of treatment depends only on the measured co-variables), and positivity (the conditional probability of receiving every value of treatment is greater than zero)¹⁵.

Participants

To minimize potential bias and ensure comparability (exchangeability and positivity assumptions), patient selection criteria were applied that would be used in a target trial. Eligible patients were 18–65 years old, and admitted with a diagnosis of obesity between 2010 and 2017, with a BMI of at least 40 kg/m², or with BMI 35–40 kg/m² associated with an obesity-related co-morbidity (hypertension, type 2 diabetes, dyslipidaemia, or obstructive sleep apnoea syndrome). The identification of obesity and co-morbidities was made on the basis of ICD-10 codes. Patients with a history of alcohol abuse and psychoactive substance use were excluded, as addictions represent a contraindication to bariatric surgery. Also excluded were patients with a history of cancer at baseline and those diagnosed with cancer in the 2 years after inclusion to avoid detection bias (misdiagnosis of cancer in preoperative assessment for bariatric surgery). Patients were then divided into two groups: those who underwent bariatric surgery and those who did not. Patients in the surgical group were identified through the medical acts classification (Classification Commune des Actes Médicaux, 11th edition), which is a national standardized classification of surgical, endoscopic, and interventional procedures¹³. Patients who underwent adjustable gastric banding, sleeve gastrectomy, or gastric bypass, irrespective of surgical approach (open or laparoscopic), were included. The unambiguous identification of the intervention permitted researchers to assure the consistency assumption.

The inception time for the bariatric cohort was the hospital stay for bariatric surgery. The inception time for the control group was the date of admission when a diagnosis of obesity appeared for the first time, irrespective of the reason for admission. All included patients were followed from the first diagnosis of obesity until cancer appearance, death or censoring in December 2019.

Outcomes

The main outcome was cancer incidence, for any type of cancer and for subgroups of obesity-related cancers. Obesity-related cancers included 14 types of cancer with sufficient evidence for being obesity-related as reported by the International Agency for Research on Cancer Working Group. These were oesophageal adenocarcinoma, postmenopausal breast cancer, and cancers of the kidney, colon, rectum, gastric cardia, liver, gallbladder, pancreas, ovary, corpus uteri and thyroid, and meningioma and multiple myeloma⁴.

To verify whether the relationship between bariatric surgery and cancer risk was inferred correctly, the risk of cancer not related to obesity was defined as a negative-control outcome¹⁶. Negative controls are used in epidemiological studies to detect confounding, selection, and measurement biases¹⁷. The formal approach is repetition of the experiment under conditions in which it is expected to produce a null result. In the present study, for patients affected by cancers that are not related to obesity, intentional weight loss was not expected to provide any effect. Hence, a null effect on negative controls would suggest that unmeasured confounders had been ruled out.

Incident cancers were identified in the database through ICD-10 codes, which in general are based on the topography/anatomical sector rather than on tumour histology. For instance, an oesophageal cancer could be identified, but information about the histological type (adenocarcinoma or squamous cell carcinoma) was not available. For identification of postmenopausal breast cancer, the definition of cancers diagnosed at age 55 years or older was adopted, as reported previously in studies using electronic data sources^11,18.

In the PMSI database, each patient is assigned a unique identifier, which remains unchanged over time. Thus, linkage between different hospital stays in different hospitals is possible. As the identifier is anonymous, patient consent was not required.

Statistical analysis

Weighting

Using a propensity score approach, inverse probability of treatment weighting (IPTW) was applied to balance baseline characteristics between exposure groups. The propensity score was derived using a logistic regression model, employing the following pretreatment variables: patient’s age, sex, BMI, duration of follow-up, co-morbidities, and tobacco consumption. Each observation was weighted by the inverse of the probability of a patient receiving bariatric surgery, given observed confounders identified on the index date. The IPTW approach aimed to create a pseudo-population in which the exposure group (bariatric surgery versus no bariatric surgery) was independent of measured confounders under the conditional exchangeability hypothesis¹⁹. To avoid the impact of extreme weights on estimations, weights were truncated at a maximum threshold of 15. No other truncation was applied. The balance among patient co-variables was assessed using the standardized mean difference (SMD); a difference of 0.100 or less was considered a well balanced result²⁰.

Multivariable analysis

To compare the risk of cancer in patients receiving versus not receiving bariatric surgery, a Cox proportional hazards regression model adjusted for IPTW was used. The model was calculated to assess the risk of any cancer, of obesity-related cancer, and of cancer not related to obesity. The risk of type-specific cancers presenting at least 50 cases was also assessed. As the weighting did not allow for perfectly similar groups, the adjustment made it possible to take into account residual differences between the two cohorts.

Sensitivity analysis

Several sensitivity analyses were conducted to assess the robustness of the findings. First, patients who underwent adjustable gastric banding were excluded from the analysis, as this procedure has become rare among bariatric procedures²¹. Second, an alternative 1 : 2 nearest-neighbours propensity score-matched design was applied, in which one patient receiving bariatric surgery was associated with two patients not receiving bariatric surgery based on the propensity score. A Cox proportional hazards regression model was used to assess cancer incidence.

All statistical analyses were performed with R software (R Foundation for Statistical Computing, Vienna, Austria)²². Data are reported according to the RECORD statement²³. Access to the PMSI database was submitted to authorization by the National Commission on Informatics and Liberty (authorization number 01947391).

Results

Participants

Between 2010 and 2019, 3 633 019 patients with a diagnosis of obesity were admitted. Among these, 2 764 636 patients (76.1 per cent) were not diagnosed with cancer after at least 2 years of follow-up. Patients aged 18–65 years, with a BMI ≥ 40 kg/m², or BMI 35–40 kg/m² and at least one obesity-related co-morbidity, were selected, leaving 1 140 347 patients eligible for the study. Some 288 604 patients (25.3 per cent) underwent bariatric surgery (bariatric group) and 851 743 (74.7 per cent) did not (control group) (Fig. S1). Baseline characteristics of the patients are shown in Table 1. There were several differences between the bariatric and the control groups, including sex ratio (81.6 versus 55.2 per cent women respectively), age (mean(s.d.) 39.8(11.5) versus 51.8(11.1) years), and BMI over 40 kg/m² (58.7 versus 26.6 per cent). Mean follow-up was 5.7(2.2) years in the bariatric group and 6.5(2.3) years in the control group.

Table 1

Patient characteristics at baseline and in the pseudo-population of inverse probability of treatment weighting analysis

	Study groups			Pseudo-population
	Bariatric group	Control group	SMD	Bariatric group	Control group	SMD
	(n = 28 8604)	(n = 85 1743)		(n = 897 674)	(n = 1 169 710)
Men	53 194 (18.4)	381 997 (44.8)	0.592	270 254.8 (30.1)	434 774.7 (37.2)	0.150
Age (years)*	39.8 (11.5)	51.8 (11.1)	1.056	45.6 (11.5)	48.0 (13.2)	0.193
Age (years)			1.068			0.384
18–29	71 733 (24.9)	55 320 (6.5)		114 369.9 (12.7)	169 552.6 (14.5)
30–39	80 712 (28.0)	86 238 (10.1)		174 606.3 (19.5)	163 793.9 (14.0)
40–49	76 502 (26.5)	169 699 (19.9)		258 289.5 (28.8)	228 470.4 (19.5)
50–59	49 077 (17.0)	322 799 (37.9)		269 483.1 (30.0)	371 619.8 (31.8)
60–65	10 580 (3.7)	217 687 (25.6)		80 925.2 (9.0)	236 274.3 (20.2)
BMI (kg/m2)			0.685			0.229
< 40	119 326 (41.3)	624 852 (73.4)		461 111.8 (51.4)	732 794.9 (62.6)
40–50	145 426 (50.4)	198 481 (23.3)		377 368.8 (42.0)	377 114.9 (32.2)
> 50	23 852 (8.3)	28 410 (3.3)		59 193.5 (6.6)	59 801.2 (5.1)
Tobacco use	32 746 (11.3)	144 309 (16.9)	0.161	122 146.9 (13.6)	179 260.8 (15.3)	0.049
Sleep apnoea syndrome	84 775 (29.4)	233 643 (27.4)	0.043	295 974.0 (33.0)	327 577.7 (28.0)	0.108
Hypertension	75 829 (26.3)	560 604 (65.8)	0.864	425 851.7 (47.4)	633 230.5 (54.1)	0.134
Hypercholesterolaemia	37 832 (13.1)	321 642 (37.8)	0.59	227 958.9 (25.4)	359 146.2 (30.7)	0.118
Myocardial infarction	2688 (0.9)	65 101 (7.6)	0.336	28 502.8 (3.2)	67 841.6 (5.8)	0.127
Congestive heart failure	5087 (1.8)	118 034 (13.9)	0.463	53 761.9 (6.0)	123 334.8 (10.5)	0.166
Peripheral vascular disease	3042 (1.1)	79 779 (9.4)	0.381	31 915.3 (3.6)	82 878.9 (7.1)	0.158
Cerebrovascular disease	4615 (1.6)	72 963 (8.6)	0.321	38 022.1 (4.2)	77 758.5 (6.6)	0.106
Dementia	179 (0.1)	5540 (0.7)	0.099	1864.1 (0.2)	5722.8 (0.5)	0.048
Chronic pulmonary disease	25 269 (8.8)	110 736 (13.0)	0.137	99 823.6 (11.1)	137 716.6 (11.8)	0.021
Rheumatic disease	2006 (0.7)	18 702 (2.2)	0.126	15 209.8 (1.7)	20 794.3 (1.8)	0.006
Peptic ulcer disease	10 313 (3.6)	18 101 (2.1)	0.087	23 924.4 (2.7)	28 143.8 (2.4)	0.016
Mild liver disease	26 956 (9.3)	62 813 (7.4)	0.071	82 096.4 (9.1)	93 817.6 (8.0)	0.040
Moderate or severe liver disease	657 (0.2)	5569 (0.7)	0.064	4006.0 (0.4)	6282.5 (0.5)	0.062
Diabetes without chronic complication	39 282 (13.6)	337 340 (39.6)	0.615	264 861.3 (29.5)	378 451.4 (32.4)	0.133
Diabetes with chronic complication	7578 (2.6)	142 017 (16.7)	0.49	78 250.1 (8.7)	149 995.1 (12.8)	0.052
Hemiplegia or paraplegia	2906 (1.0)	31 678 (3.7)	0.179	19 299.1 (2.1)	34 757.8 (3.0)	0.111
Renal disease	2720 (0.9)	58 233 (6.8)	0.308	27 109.2 (3.0)	61 082.8 (5.2)	0.013
Human immunodeficiency virus	494 (0.2)	2925 (0.3)	0.034	2499.9 (0.3)	3438.9 (0.3)	0.003

Values in parentheses are percentages unless indicated otherwise; *values are mean(s.d.). SMD, standardized mean difference.

In total, 48 411 incident cancers were identified (Table 2). In the bariatric group, 4483 patients (1.6 per cent) developed a cancer. In the control group, a cancer was observed in 43 928 patients (5.2 per cent). The four most common cancers were colorectal (4763, 9.8 per cent), postmenopausal breast (4468, 9.2 per cent), prostate (4376, 9.0 per cent), and lung cancer (3730, 7.7 per cent).

Table 2

Types of cancer diagnosed

	Total (n = 48 411)	Bariatric group (n = 4483)	Control group (n = 43 928)
Colon and rectum	4763 (9.84)	329 (7.34)	4434 (10.09)
Breast (postmenopausal)	4468 (9.23)	331 (7.38)	4137 (9.42)
Prostate	4376 (9.04)	119 (2.65)	4257 (9.69)
Lung	3730 (7.7)	183 (4.08)	3547 (8.07)
Breast (premenopausal)	3605 (7.45)	1029 (22.95)	2576 (5.86)
Skin (except melanoma)	3466 (7.16)	314 (7)	3152 (7.18)
Bladder	2478 (5.12)	104 (2.32)	2374 (5.4)
Kidney	2451 (5.06)	210 (4.68)	2241 (5.1)
Corpus uteri	1895 (3.91)	203 (4.53)	1692 (3.85)
Thyroid gland	1517 (3.13)	296 (6.6)	1221 (2.78)
Pancreas	1426 (2.95)	98 (2.19)	1328 (3.02)
Secondary: respiratory and digestive	1301 (2.69)	86 (1.92)	1215 (2.77)
Non-Hodgkin lymphoma	1287 (2.66)	100 (2.23)	1187 (2.7)
Secondary: other	1037 (2.14)	79 (1.76)	958 (2.18)
Liver and intrahepatic bile ducts	954 (1.97)	40 (0.89)	914 (2.08)
Lip, oral cavity and pharynx	854 (1.76)	50 (1.12)	804 (1.83)
Skin: cutaneous melanoma	683 (1.41)	84 (1.87)	599 (1.36)
Central nervous system (except meningioma)	635 (1.31)	58 (1.29)	577 (1.31)
Ovary	599 (1.24)	87 (1.94)	512 (1.17)
Gastric, non-cardia	575 (1.19)	31 (0.69)	544 (1.24)
Multiple myeloma	555 (1.15)	33 (0.74)	522 (1.19)
Lymphoid leukaemia	512 (1.06)	42 (0.94)	470 (1.07)
Secondary: lymph nodes	483 (1)	49 (1.09)	434 (0.99)
Cervix uteri	478 (0.99)	108 (2.41)	370 (0.84)
Oesophagus	457 (0.94)	22 (0.49)	435 (0.99)
Other cancers	431 (0.89)	35 (0.78)	396 (0.9)
Myeloid leukaemia	361 (0.75)	45 (1)	316 (0.72)
Larynx	305 (0.63)	14 (0.31)	291 (0.66)
Connective and soft tissue	255 (0.53)	40 (0.89)	215 (0.49)
Small intestine	220 (0.45)	19 (0.42)	201 (0.46)
Gastric cardia	218 (0.45)	15 (0.33)	203 (0.46)
Hodgkin lymphoma	214 (0.44)	32 (0.71)	182 (0.41)
Extrahepatic biliary tract	194 (0.4)	10 (0.22)	184 (0.42)
Myeloid leukaemia	190 (0.39)	13 (0.29)	177 (0.4)
Immunoproliferative diseases	134 (0.28)	11 (0.25)	123 (0.28)
Vulva	132 (0.27)	20 (0.45)	112 (0.25)
Bone and articular cartilage	119 (0.25)	26 (0.58)	93 (0.21)
Eye and adnexa	112 (0.23)	18 (0.4)	94 (0.21)
Testis	101 (0.21)	20 (0.45)	81 (0.18)
Adrenal gland	96 (0.2)	13 (0.29)	83 (0.19)
Leukaemia, other	90 (0.19)	11 (0.25)	79 (0.18)
Retroperitoneum and peritoneum	83 (0.17)	11 (0.25)	72 (0.16)
Meningioma	81 (0.17)	7 (0.16)	74 (0.17)
Gallbladder	77 (0.16)	2 (0.04)	75 (0.17)
Other endocrine glands	76 (0.16)	5 (0.11)	71 (0.16)
Uterus, part unspecified	75 (0.15)	10 (0.22)	65 (0.15)
Breast (men)	68 (0.14)	1 (0.02)	67 (0.15)
Lymphoid, haematopoietic and related tissues	67 (0.14)	8 (0.18)	59 (0.13)
Heart, mediastinum and pleura	64 (0.13)	8 (0.18)	56 (0.13)
Mesothelioma	63 (0.13)	4 (0.09)	59 (0.13)

Values in parentheses are percentages.

Outcomes

The pseudo-population emulated by IPTW included 2 067 384 patients, of whom 897 674 (43.4 per cent) underwent bariatric surgery and 1 169 710 (56.6 per cent) were not operated on (Table 1). The average SMD decreased from 0.35 to 0.09 after weighting (Fig. S2). In the pseudo-population, the causal HRs associated with bariatric surgery were 0.92 (95 per cent c.i. 0.88 to 0.97) for all cancers, 0.89 (0.83 to 0.95) for obesity-related cancers, and 0.96 (0.91 to 1.01) for cancers not related to obesity (Fig. 1).

Fig. 1

Forest plot of multivariable Cox proportional-hazards models for cancers on the weighted population

Hazard ratios are shown with 95 per cent confidence intervals.

Other analyses

In the first sensitivity analysis, the study design was modified, and 1 : 2 propensity score matching was undertaken. Two similar populations were obtained with a mean SMD of 0.06 (Table S1). In the matched population, the HR was 0.88 (95 per cent 0.84 to 0.92) for all cancers, 0.81 (0.76 to 0.86) for obesity-related cancers, and 0.94 (0.89 to 0.99) for cancers not related to obesity (Fig. S3). In the subgroup of patients receiving sleeve gastrectomy or gastric bypass, the HR for obesity-related cancers decreased to 0.86 (0.80 to 0.93).

Discussion

In this study, a national database was used as a framework for comparative effectiveness based on emulation of an RCT to estimate the causal effect of bariatric surgery on cancer incidence in patients with obesity. More than 280 000 patients who underwent bariatric surgery were included, who developed over 4400 cancers, which is considerably higher than reported in previous studies^11,24. A reduction of 11 per cent in obesity-related cancers among patients who underwent bariatric surgery was found compared with non-surgical obese patients. These results strengthen previous observations on the benefit of weight loss for cancer prevention. No association was seen between bariatric surgery and non-obesity-related cancers, suggesting that confounders were properly controlled for.

In 2009, Adams and colleagues⁷ published a 12-year follow-up study on more than 6000 patients who had bariatric surgery compared with an obese control group, reporting a 24 per cent reduction in overall cancer incidence and a 38 per cent reduction in obesity-related cancers. That study used a Cox regression model which was adjusted for three co-variables (age, sex, and BMI). In the same year, the Swedish Obese Subjects (SOS) Study⁸ reported on the incidence of cancer in a cohort of 2000 patients matched with obese controls. Matching was done using 18 clinical variables. After a mean follow-up exceeding 10 years, bariatric surgery was associated with a 33 per cent reduction in cancer incidence. More recently, in 2019, Schauer et al.¹¹ confirmed the effect of bariatric surgery on cancer incidence for any cancer (HR 0.67, 95 per cent c.i. 0.60 to 0.74; P < 0.001) and for obesity-related cancers (HR 0.65, 0.51 to 0.69; P < 0.001) in a cohort of more than 22 000 patients who had bariatric surgery and 66 000 obese individuals. Mean follow-up was shorter (3.5 years), and the two matched populations were compared by means of a Cox proportional hazards model. This group also found an effect of bariatric surgery on cancer not associated with obesity, and confirmed this finding in premenopausal breast cancer (not related to obesity) in another publication²⁵.

A study from Bailly and co-workers²⁶ used the same data source (PMSI database) as the present study. A benefit of bariatric surgery on the risk of colorectal cancer was reported (HR 0.68, 0.60 to 0.77). This remains controversial as other studies^9,24,27–29 from Sweden and the UK have reported an increased risk of colorectal cancer after bariatric surgery. It is interesting that all studies which found an increased risk of colorectal cancer did not compare a bariatric surgery population with an equivalent obese population, but used the method of standardized incidence ratio, thereby comparing patients who had bariatric surgery with the general population. This highlights the importance of an adequate control group, which, by definition, can only be provided by randomization. As an RCT is not always feasible (for ethical, logistical or financial reasons), methods have been developed to emulate this type of trial using observational data. If the emulation is successful, the analysis of observational data could yield effect estimates similar to those of a target trial, responding to the initial causal question¹². The inadequacy of controlling for confounders may yield contradictory or counterintuitive results, where bariatric surgery may increase the risk of obesity-related cancers (as in the case of colorectal cancer) or decrease the risk of cancers that are not related to obesity.

There is concern regarding the possible increased risk of oesophagogastric neoplasm following bariatric surgery, in particular for reflux-inducing procedures such as sleeve gastrectomy³⁰. At present, there is no evidence that sleeve gastrectomy is associated with an increased risk³¹. A recent study³² from Canada compared about 4000 patients after sleeve gastrectomy or duodenal switch with 12 000 non-surgical controls with obesity. After adjustment, the risk of oesophageal cancer was similar in both groups.

The main limitations of this study relate to its observational nature, as unmeasured confounding may persist between bariatric and control groups. Nevertheless, the specification of a target trial permitted an analysis to be built in order to minimize bias. Only patients potentially eligible for bariatric surgery were included (exchangeability assumption). With this method, some patients who actually received bariatric surgery did not meet the inclusion criteria and were excluded. This highlights the fact that the national discharge database does not capture all clinical details associated with the indication for bariatric surgery. However, the same criteria were also applied to the control group to improve comparability of the groups. Admission bias, also known as Berkson bias, could be a concern in interpretation of the results. All patients, including those in the control group, were identified by hospital admissions. Hospital cases might have higher-risk exposures or disease than those from the population at large³³. For this reason, patients with a history of cancer in the 2 years after inclusion were excluded. Another limitation is the assessment of history of smoking through ICD-10 codes, which was probably underestimated as the prevalence of smoking in the French population is estimated at 30 per cent³⁴. However, it is unlikely that this represents confusion bias, because the prevalence of smoking was similar in the two groups in the weighted and matched populations. Finally, it has to be considered that the mean age of patients having bariatric surgery is around 40 years, which is considerably lower than the mean age for development of most cancer types assessed in this study. Further studies could focus on the elderly population more at risk of developing neoplasms.

Funding

Data analysis was funded by Centre Hospitalier Intercommunal de Créteil.

Disclosure. The authors declare no conflict of interest.

Supplementary material

Supplementary material is available at BJS online.

Author notes

Presented to the French Obesity Surgery Society (SOFFCO-MM) Congress, Paris, France, September 2020