Fracture Risk After Gastric Bypass Surgery: A Retrospective Cohort Study

ABSTRACT

 

Gastric bypass surgery constitutes the most common and effective bariatric surgery to treat obesity. Gastric bypass leads to bone loss, but fracture risk following surgery has been insufficiently studied. Furthermore, the association between gastric bypass and fracture risk has not been studied in patients with diabetes, which is a risk factor for fracture and affected by surgery. In this retrospective cohort study using Swedish national databases, 38,971 obese patients undergoing gastric bypass were identified, 7758 with diabetes and 31,213 without. An equal amount of well‐balanced controls were identified through multivariable 1:1 propensity score matching. The risk of fracture and fall injury was investigated using Cox proportional hazards and flexible parameter models. Fracture risk according to weight loss and degree of calcium and vitamin D supplementation 1‐year postsurgery was investigated. During a median follow‐up time of 3.1 (interquartile range [IQR], 1.7 to 4.6) years, gastric bypass was associated with increased risk of any fracture, in patients with and without diabetes using a multivariable Cox model (hazard ratio [HR] 1.26; 95% CI, 1.05 to 1.53; and HR 1.32; 95% CI, 1.18 to 1.47; respectively). Using flexible parameter models, the fracture risk appeared to increase with time. The risk of fall injury without fracture was also increased after gastric bypass. Larger weight loss or poor calcium and vitamin D supplementation after surgery were not associated with increased fracture risk. In conclusion, gastric bypass surgery is associated with an increased fracture risk, which appears to be increasing with time and not associated with degree of weight loss or calcium and vitamin D supplementation following surgery. An increased risk of fall injury was seen after surgery, which could contribute to the increased fracture risk. © 2018 American Society for Bone and Mineral Research.

OSTEOPOROSIS, FRACTURE RISK ASSESSMENT, GENERAL POPULATION STUDIES

Introduction

Obesity, defined by a body mass index (BMI) ≥30, affects 38% of US adults,⁽¹⁾ and is associated with diabetes mellitus, cardiovascular disease, and increased mortality.⁽²⁾ To treat this growing epidemic, bariatric surgery has been increasingly used.⁽³⁾ Although sleeve gastrectomy recently has risen to become the procedure of choice,⁽⁴⁾ Roux‐en‐Y gastric bypass (gastric bypass) is still the historically most utilized method and the most effective and well‐documented bariatric procedure, providing long‐term weight loss, metabolic benefits, and improved survival.⁽⁵, ⁶, ⁷, ⁸⁾ However, gastric bypass surgery has adverse effects on the skeleton, as reflected by increased bone resorption markers and decreased bone mineral density (BMD) postsurgery.⁽⁹⁾ Although several studies have reported that bariatric surgery in general is associated with increased fracture risk, no study has investigated fracture risk in gastric bypass patients in comparison to obese weight‐matched controls.⁽¹⁰, ¹¹, ¹², ¹³, ¹⁴⁾ Studying gastric bypass specifically is important because it is the bariatric procedure most associated with the most profound changes in bone metabolism.⁽⁹⁾

Bone loss due to decreased weight‐mediated mechanical loading and secondary hyperparathyroidism due to vitamin D deficiency have commonly been proposed as the main mechanisms behind the adverse skeletal effects of gastric bypass. Whether weight loss or lack of calcium and vitamin D supplementation after surgery affects fracture risk is unknown.⁽⁹⁾

Diabetes prevalence is a known risk factor for fracture,⁽¹⁵⁾ and in patients with diabetes, BMD is reduced after gastric bypass.⁽¹⁶⁾ Since gastric bypass often leads to diabetes remission,⁽¹⁷⁾ diabetes status should be considered when investigating fracture risk after gastric bypass. The role of diabetes status on fracture risk after gastric bypass has not been studied.

More than 2 million bone fragility–related fractures occur annually in the United States, costing nearly 17 billion USD.⁽¹⁸⁾ Fractures cause suffering, decreased mobility, increased healthcare costs, and increased mortality.⁽¹⁹, ²⁰⁾ Thus, the extensive use of gastric bypass and a potentially increased fracture risk with this surgery would aggravate the already substantial fracture burden.

In this nationwide study with 77,942 patients, we used large databases to investigate if gastric bypass surgery in obese patients, with and without diabetes, was associated with fracture risk, and if the fracture risk was associated with postsurgery weight loss or insufficient calcium and vitamin D supplementation.

Patients and Methods

Study design

Adult patients with BMI ≥30 undergoing gastric bypass surgery were divided into patients with and without diabetes. Gastric bypass patients with diabetes were compared to obese controls with diabetes, matched using multivariable propensity score to achieve well‐balanced groups. Similarly, the gastric bypass patients without diabetes were compared to obese controls without diabetes, also matched using multivariable propensity score (Fig. 1). Baseline characteristics were controlled for and patients were followed over time using multivariable Cox models to assess fracture risk. Because this study is in fact two studies with different cases and controls and with different matching variables, comparing the fracture outcomes between patients with and without diabetes was not possible.

Study population. A gastric bypass patient was considered to have diabetes if one of the following criteria were fulfilled: (i) included in the NDR; (ii) a diabetes diagnosis (ICD‐10, E10–E14) in the NPR; (iii) a dispensed diabetes medication (ATC code A10) in the Swedish Prescribed Drug Register (2005–2014); (iv) a fasting plasma glucose ≥7.0 mmol/L; or (v) HbA_1c ≥48 mmol/mol. Control patients with diabetes were retrieved from the NDR if they met the following criteria: (i) first entry in 2007–2014; (ii) an HbA_1c value ≥20 mmol/mol; (iii) BMI ≥30 kg/m²; and (iv) age, weight, and height values that did not exceed the top and bottom values of the surgery group. GBP = gastric bypass; NDR = National Diabetes Register; NPR = National Patient Register; SOREG = Scandinavian Obesity Surgery Register.

Registers

The study population was created by linking several Swedish national registers. With a registration rate of 99%,⁽²¹⁾ the Scandinavian Obesity Surgery Register (SOReg) was used to identify case patients who had undergone gastric bypass surgery in Sweden. The register includes data on presurgery weight and height as well as follow‐up measurements at 6 weeks, 12 months, and 24 months, as well as glycated hemoglobin (HbA_1c). Obese controls not undergoing gastric bypass surgery were retrieved from two different registers. Patients with diabetes were retrieved from the National Diabetes Register (NDR), whereas patients without diabetes were identified using the obesity diagnosis (International Classification of Diseases and Related Health Problems, 10th Revision [ICD‐10] code E66) in the National Patient Register (NPR) (Fig. 1). The NDR started in 1996 as a tool for quality control of care for persons with diabetes and currently includes the vast majority of Swedes with diabetes.⁽²²⁾ The NPR includes all diagnoses from both inpatient (1987–2014) and outpatient (2001–2014) visits and has high validity.⁽²³⁾

Study population

Depending on diabetes status, the gastric bypass surgery patients were divided into two groups (Fig. 1). Each case group of gastric bypass patients were then matched to its respective controls with a 1 to 1 multivariable propensity score matching, using variables related to obesity, fracture risk, and contraindications for surgery (Table 1). For the patients with diabetes, weight, height, and HbA₁c were also included as matching variables (Table 1). Baseline for the study was the date of surgery for the case patients and the date of first entry in NDR/NPR for the control patients. Exclusion criteria encompassed patients with dates of death prior to baseline (probable register error), patients immigrating later than 2004 (to allow the same length of medication history), and potential control patients not included in the SOReg, but with a code for bariatric surgery in the NPR. Patents who underwent gastric bypass surgery were defined as having diabetes if they had a preoperative blood tests diagnostic for diabetes (fasting glucose ≥7 mmol/l and/or HbA1c ≥48 mmol/mol), diagnosis of diabetes (in NDR or NPR) or a known ongoing diabetes medication. In total, 15.4% of these patients were defined to have diabetes solely due to elevated preoperative blood glucose.

Baseline Characteristics Before and After 1 to 1 Propensity Score Matching

		Patients with diabetes					Patients without diabetes
		Controls					Controls
Baseline variable description	Cases (gastric bypass) (n = 7758)	Prematch (n = 134,872)	St. Diff. (%)	Matched (n = 7758)	St. Diff. (%)	Cases (gastric bypass) (n = 31,213)	Prematch (n = 71,020)	St. Diff. (%)	Matched (n = 31,213)	St. Diff. (%)
Female sex (%)	66.1	42.3	49.2	65.6	1.2	78.9	69.7	21.2	77.3	3.8
Age (years), mean ± SD	47.0 ± 10.4	59.1 ± 11.1	113.1	47.5 ± 10.0	5.4	39.5 ± 10.7	44.7 ± 15.7	38.8	39.2 ± 10.9	2.7
18–30 (%)	7.3	2.1	24.9	6.3	4.2	22.6	23.4	2.0	23.8	2.9
31–40 (%)	18.7	4.3	46.2	17.0	4.4	30.4	22.6	17.6	32.0	3.4
41–50 (%)	33.1	14.0	46.4	35.2	4.3	30.8	16.4	34.4	27.6	7.2
51–60 (%)	32.7	26.9	12.6	33.6	1.9	14.1	16.3	5.9	14.6	1.3
61–75 (%)	8.2	52.7	110.5	8.0	0.5	2.1	21.3	62.7	2.1	0.2
Weight (kg), mean ± SD	122.3 ± 21.5	100.4 ± 16.2	115.1	120.8 ± 25.2	6.6	121.4 ± 20.6	N/A	N/A	N/A	N/A
Height (cm), mean ± SD	169.8 ± 9.5	171.3 ± 9.7	14.7	169.7 ± 10.0	1.0	169 ± 8.8	N/A	N/A	N/A	N/A
BMI (kg/m2), mean ± SD	42.3 ± 5.7	34.2 ± 4.4	160.1	41.8 ± 7.1	8.0	42.4 ± 5.4	N/A	N/A	N/A	N/A
Inclusion 2007–2009 (%)3	18.4	38.0	44.6	19.6	3.1	18.8	27.8	21.4	18.5	0.7
Inclusion 2010–2012 (%)3	54.0	38.6	31.1	51.0	6.1	52.0	39.5	25.3	52.1	0.0
Inclusion 2013–2014 (%)3	27.7	23.4	9.8	29.5	4.0	29.2	32.7	7.6	29.4	0.5
Known fracture‐free time (years), mean ± SD	23.3 ± 5.3	22.8 ± 5.1	11.1	23.4 ± 5.3	0.9	23.6 ± 4.8	23.3 ± 5.2	7.0	23.7 ± 4.8	1.0
Any previous fracture (%)	7.2	6.4	3.2	7.2	0.0	5.8	6.5	2.7	5.8	0.2
Number of previous fractures ≥2 (%)	1.6	1.7	0.7	2.0	2.6	1.4	1.8	2.6	1.4	0.6
Previous hip fracture (%)	0.1	0.4	6.8	0.1	0.5	0.1	0.3	5.1	0.1	0.2
Previous vertebral fracture (%)	0.5	0.5	0.9	0.6	1.8	0.4	0.6	3.2	0.4	0.6
Previous fall injury without fracture (%)	2.2	2.3	0.8	2.3	0.5	2.2	2.5	1.4	2.3	0.3
Rheumatoid arthritis (%)	1.5	1.5	0.4	1.5	0.3	1.2	1.5	2.8	1.1	0.9
Alcohol related diseases (%)	2.6	2.9	1.7	2.6	0.3	2.1	2.2	1.0	2.0	0.6
Intense glucocorticoid treatment (%)4	5.7	6.5	3.5	6.3	2.7	4.3	7.0	11.7	4.5	1.1
Prevalent calcium and vitamin D (%)5	3.8	3.4	2.4	3.8	0.4	2.3	3.1	5.1	2.1	1.4
Osteoporosis (%)6	0.6	1.0	4.3	0.7	1.7	0.3	0.9	7.8	0.3	0.0
Secondary osteoporosis (%)7	22.0	16.8	13.1	19.6	6.0	4.8	4.6	0.9	4.7	0.6
Charlson Comorbidity Index, mean ± SD	1.7 ± 1.2	2.1 ± 1.9	28.6	1.7 ± 1.3	1.3	0.3 ± 0.8	0.6 ± 1.4	28.5	0.3 ± 0.8	1.5
0 (%)	0	0	N/A	0	N/A	83.5	74.0	23.4	84.8	3.5
1 (%)	63.8	56.5	15.1	63.7	0.3	10.1	11.8	5.7	9.4	2.1
2 (%)	20.3	16.8	8.8	20.1	0.4	4.7	7.2	10.5	3.9	4.0
≥3 (%)	15.9	26.7	26.6	16.2	0.8	1.7	7.0	26.0	1.9	1.4
HbA1c (mmol/mol), mean ± SD	56.4 ± 17.8	56.1 ± 16.1	1.8	57.1 ± 16.8	3.8	36.7 ± 4.1	N/A	N/A	N/A	N/A
Any diabetes complication (%)	13.0	15.5	7.0	13.3	0.7	0	0	N/A	0	N/A
Insulin (%)	24.4	22.0	5.7	23.9	1.3	0	0	N/A	0	N/A
Other diabetes medications (%)	31.8	32.4	1.3	32.6	1.6	0	0	N/A	0	N/A

Propensity score matching was performed using these parameters, as well as 50 other diagnoses and medications (Supporting Table 7). To measure prematching and postmatching imbalances, standardized difference (St. Diff.) was calculated = ABS(mean1 − mean2)/√[(variance1 + variance2)/2]. A postmatch standardized difference of less than 10% indicates a relatively small imbalance.⁽²⁹⁾ All variables, including the 50 variables in Supporting Table 7, have a postmatch difference of less than 10%.

N/A = not available; SD = standardized difference.

a

Inclusion time was matched using month starting January 2007 as a variable, separated in three different intervals.

b

Five milligrams prednisolone equivalents or more per day for more than 3 consecutive months, not necessarily recently.

c

Dispensations up to 2 years prior to baseline was accounted for.

d

Osteoporosis, M80‐M81

e

Secondary osteoporosis is defined in Supporting Table 6.

Compliance with ethical standards

The study was approved by the regional ethical review board in Gothenburg.

Definition of outcomes

Using the NPR, fracture was defined as all non‐malignant fracture diagnoses in ICD‐10 (or ICD‐9 translated to ICD‐10), regardless of type of trauma, apart from head fractures (Supporting Table 1). If a fracture diagnosis on the same skeletal site was repeated within five months, the later diagnosis was discarded since it most likely reflected a revisit rather than a new fracture. The risk of fracturing specific skeletal sites was also investigated as well as the risk of fall without a fracture (Supporting Table 2). All patients with fractures in Sweden, including those not requiring surgical treatment, are treated in an orthopedic clinic and thereby included in the NPR. To further ascertain the occurrence of hip fracture a simultaneous code for surgical procedure was required. Identification of hip fracture in registers using this combination has high accuracy.⁽²⁴⁾

Definition of covariates

Data regarding prior illnesses was collected from the NPR, but information from primary care clinics was not available. The Charlson comorbidity index was calculated to summarize and quantify comorbidity.⁽²⁵⁾ All medication data were collected from the Swedish Prescribed Drug Register for the years 2005 to 2014.⁽²⁶⁾ Previous medication (prior to baseline) was accounted for if repeated and recent; ie, the prescribed drug (identified using Anatomical Therapeutic Chemical [ATC] code) was dispensed two or more times prior to baseline and was dispensed not more than 90 days prior to baseline. The definitions of the covariates are presented in Supporting Table  3–6.

Statistical analysis

To control for potential confounders, gastric bypass patients without diabetes were matched to obese patients, also without diabetes, through a multivariable one to one propensity score matching.⁽²⁷, ²⁸⁾ The 63 matching variables included age, sex, inclusion month, as well as previous illnesses and medications associated with fracture risk, obesity, and contraindications for surgery. Due to large prematch imbalance, age was divided in five intervals, enabling the beta value in the logistic regression to vary linearly differently in each age interval, to achieve conforming age distributions. Similarly, inclusion month was divided in three intervals. When matching the gastric bypass patients with diabetes to the controls with diabetes, matching variables also included weight, height, BMI, HbA_1c, diabetes complications, and diabetes medications. Matching variables are presented in Table 1. To assess prematch imbalance and postmatch balance, standardized differences were estimated for all baseline covariates before and after matching.⁽²⁹⁾ Crude event rates were calculated as number of events divided by number of person years with 95% confidence intervals (CIs) estimated using exact Poisson limits. Associated p values were obtained from Poisson regression.

To investigate the association between fracture risk and gastric bypass surgery, we used a Cox proportional hazards model starting at baseline. The Cox regression model used the length of each individual's follow‐up period and was censored for death (from the Causes of Death register), emigration (from Statistics Sweden) or end of follow‐up (December 31, 2014). The analyses were performed for any fracture as well as for different skeletal sites and fall injury without fracture. We also analyzed the association between gastric bypass and death. The assumption of proportional hazard was tested by introducing an interaction term between treatment group and log(time). A flexible parametric survival model was also applied, allowing the continuous hazard ratio (HR) with 95% CI to vary with time.

To assess if the risk of any fracture was associated by weight loss after surgery, we started the Cox model analysis for the patients undergoing surgery at the time when the follow‐up weight was measured; ie, 6 weeks, 12 months, and 24 months. Also, weight loss at 1 year was divided in tertiles, where the medium and high tertiles were compared to the low tertile of weight loss. The difference between patients who did not return for the 1‐year postsurgery follow‐up and those who did was measured using standardized difference.

To assess if fracture risk was associated with postsurgery supplements, we repeated the multivariable Cox model, but started 1 year postsurgery, and adjusted for the first year of calcium and vitamin D dispensation (zero if none). We also started the multivariable Cox model at baseline, and instead adjusted for a time‐dependent covariate of incident calcium and vitamin D (updated every 6 months as average daily doses based on dispensed medication during 12 historic months and then expressed as percentage of recommended daily dose, 1000 mg/800 International Units).

All tests were 2‐tailed and conducted at 0.05 significance level. Propensity score matching was performed using R 3.3.2 (R Foundation for Statistical Computing, Vienna, Austria; https://www.r-project.org/). Crude event rates, Cox proportional assumption test, and the flexible parameter analyses were performed using SAS software version 9.4 (SAS Institute, Inc., Cary, NC, USA). All other statistical analyses were performed using IBM SPSS software, version 22 (IBM Corp., Armonk, NY, USA).

Results

The present study included 38,971 gastric bypass patients, of which 7758 had diabetes and 31,213 did not, and equally many matched controls (Fig. 1). The total of 77,942 patients had a median and total follow‐up time of 3.1 (IQR, 1.7 to 4.6) years and 251,310 person‐years, respectively. Patients immigrating to Sweden in 2004 or earlier accounted for 14.2% of all patients. Propensity score matching generated well‐balanced control groups for the obese patients both with and without diabetes (Table 1; Supporting Table  7).

After 6 weeks, 12 months, and 24 months, the mean ± SD weight loss among the gastric bypass patients with diabetes was 17.4 ± 6.9 kg, 35.5 ± 12.3 kg, and 35.1 ± 14.1 kg with dropout rates of 2.4%, 10.4%, and 36.1%, respectively. Among the gastric bypass patients without diabetes, the equivalent mean weight loss was 17.6 ± 6.6 kg, 39.3 ± 11.9 kg, and 39.9 ± 13.7 kg with dropout rates of 3.8%, 11.9%, and 34.6%, respectively (Supporting Table 8). Patients who did not return for the 1 year for follow‐up were younger and had a higher prevalence of alcohol related diseases than patients who returned for the 1‐year visit, but there was no difference in weight loss at 6 weeks (Supporting Table 9). This dropout only affected the weight follow‐up. All included patients were followed for the whole study duration in NPR. Among the controls with diabetes, 4513 (58%) returned around a year after baseline with slightly less weight (117.7 ± 23.8 kg) versus baseline (119.9 ± 24.0 kg, p < 0.001).

Gastric bypass surgery was associated with increased risk of any fracture, in obese patients both with diabetes and without diabetes (unadjusted Cox model, HR 1.26; 95% CI, 1.04 to 1.52; p = 0.01 and HR 1.31; 95% CI, 1.17 to 1.46; p < 0.001; respectively), and the association was maintained after multivariable adjustment (Table 2). The risk of major osteoporotic fracture was also increased, regardless of diabetes status, whereas the risk of upper limb fracture was increased only for patients without diabetes, and the risk of lower leg fracture was significantly lower in patients without diabetes (Table 2).

Incidences and Hazard Ratios in Gastric Bypass Patients Versus Controls in Patients With and Without Diabetes

	Patients with diabetes			Patients without diabetes
Description	Matched controls	Gastric bypass	p	Matched controls	Gastric bypass	p
Patients, n	7758	7758		31 213	31 213
Time at risk (years), median (IQR)	3.09 (1.70–4.60)	3.21 (1.84–4.58)	0.005	3.11 (1.72–4.59)	3.18 (1.77–4.64)	<0.001
Any fracture
Events, n (%)	195 (2.5)	251 (3.2)	0.007	579 (1.9)	768 (2.5)	<0.001
Per 1000 person years (95% CI)	8.0 (7.0–9.2)	10.1 (8.9–11.4)	0.02	5.9 (5.4–6.4)	7.7 (7.1–8.2)	<0.001
Crude Cox, HR (95% CI)	1 [Ref.]	1.26 (1.04–1.52)9	0.02	1 [Ref.]	1.31 (1.17–1.46)9	<0.001
Multivariable1	1 [Ref.]	1.28 (1.06–1.54)9	0.01	1 [Ref.]	1.33 (1.19–1.48)9	<0.001
Multivariable2	1 [Ref.]	1.26 (1.05–1.53)9	0.01	1 [Ref.]	1.32 (1.18–1.47)9	<0.001
Major osteoporotic fracture10
Events, n (%)	54 (0.7)	81 (1.0)	0.02	139 (0.4)	252 (0.8)	<0.001
Per 1000 person years (95% CI)	2.2 (1.7–2.9)	3.2 (2.6–4.0)	0.03	1.4 (1.2–1.6)	2.5 (2.2–2.8)	<0.001
Crude Cox, HR (95% CI)	1 [Ref.]	1.46 (1.04–2.07)9	0.03	1 [Ref.]	1.78 (1.45–2.19)9	<0.001
Multivariable1	1 [Ref.]	1.55 (1.09–2.18)9	0.01	1 [Ref.]	1.79 (1.45–2.20)9	<0.001
Multivariable2	1 [Ref.]	1.51 (1.07–2.15)9	0.02	1 [Ref.]	1.79 (1.46–2.21)9	<0.001
Hip fracture
Events, n (%)	4 (0.1)	22 (0.3)	0.001	11 (0.0)	27 (0.1)	0.01
Per 1000 person years (95% CI)	0.2 (0.1–0.4)	0.9 (0.6–1.3)	0.002	0.1 (0.1–0.2)	0.3 (0.2–0.4)	0.01
Crude Cox, HR (95% CI)	1 [Ref.]	5.32 (1.83–15.44)	0.002	1 [Ref.]	2.41 (1.19–4.85)	0.01
Multivariable1	1 [Ref.]	5.75 (1.97–16.77)	0.001	1 [Ref.]	2.50 (1.24–5.06)	0.01
Multivariable2	1 [Ref.]	5.03 (1.70–14.88)	0.004	1 [Ref.]	2.62 (1.28–5.36)	0.008
Upper limb fracture
Events, n (%)	60 (0.8)	73 (0.9)	0.30	148 (0.5)	254 (0.8)	<0.001
Per 1000 person years (95% CI)	24.4 (18.9–31.4)	29.0 (23.0–36.4)	0.32	14.9 (12.7–17.5)	25.1 (22.2–28.4)	<0.001
Crude Cox, HR (95% CI)	1 [Ref.]	1.19 (0.85–1.67)	0.32	1 [Ref.]	1.70 (1.38–2.08)9	<0.001
Multivariable1	1 [Ref.]	1.24 (0.88–1.74)	0.22	1 [Ref.]	1.69 (1.38–2.07)9	<0.001
Multivariable2	1 [Ref.]	1.24 (0.88–1.75)	0.22	1 [Ref.]	1.69 (1.38–2.07)9	<0.001
Lower leg fracture
Events, n (%)	58 (0.7)	57 (0.7)	0.93	181 (0.6)	147 (0.5)	0.07
Per 1000 person years (95% CI)	2.4 (1.8–3.0)	2.3 (1.7–2.9)	0.82	1.8 (1.6–2.1)	1.4 (1.2–1.7)	0.04
Crude Cox, HR (95% CI)	1 [Ref.]	0.95 (0.66–1.38)	0.80	1 [Ref.]	0.79 (0.64–0.99)9	0.04
Multivariable1	1 [Ref.]	0.96 (0.66–1.38)	0.82	1 [Ref.]	0.80 (0.64–0.99)9	0.04
Multivariable2	1 [Ref.]	0.95 (0.65–1.37)	0.77	1 [Ref.]	0.79 (0.63–0.98)9	0.03
Death
Events, n (%)	206 (2.7)	111 (1.4)	<0.001	366 (1.2)	192 (0.6)	<0.001
Per 1000 person years (95% CI)	8.3 (7.3–9.5)	4.4 (3.6–5.3)	<0.001	3.7 (3.3–4.1)	1.9 (1.6–2.2)	<0.001
Crude Cox, HR (95% CI)	1 [Ref.]	0.52 (0.42–0.66)	<0.001	1 [Ref.]	0.52 (0.43–0.61)9	<0.001
Multivariable1	1 [Ref.]	0.54 (0.43–0.68)	<0.001	1 [Ref.]	0.55 (0.46–0.65)9	<0.001
Multivariable3	1 [Ref.]	0.53 (0.42–0.66)	<0.001	1 [Ref.]	0.55 (0.46–0.65)9	<0.001
Fall injury with no fracture
Events, n (%)	191 (2.5)	240 (3.1)	0.02	633 (2.0)	797 (2.6)	<0.001
Per 1000 person years (95% CI)	7.8 (6.8–9.0)	9.6 (8.5–10.9)	0.03	6.4 (5.9–6.9)	8.0 (7.4–8.5)	<0.001
Crude Cox, HR (95% CI)	1 [Ref.]	1.23 (1.01–1.48)	0.04	1 [Ref.]	1.24 (1.12–1.38)9	<0.001
Multivariable1	1 [Ref.]	1.25 (1.03–1.51)	0.02	1 [Ref.]	1.24 (1.12–1.38)9	<0.001
Multivariable4	1 [Ref.]	1.26 (1.04–1.52)	0.02	1 [Ref.]	1.24 (1.12–1.38)9	<0.001

Multivariable1 = Cox model adjusted for propensity score, age, sex, weight (only for patients with diabetes), and height (only for patients with diabetes); Multivariable2 = Cox model adjusted for propensity score, age, sex, weight (only for patients with diabetes), height (only for patients with diabetes), rheumatoid arthritis, alcohol related diseases, fracture free time, any previous fracture, previous hip fracture, previous vertebral fracture, previous number of fractures, previous fall injury without fracture, previous osteoporosis, previous secondary osteoporosis, previous glucocorticoids (≥5 mg of prednisolone equivalents per day more than 3 months), previous calcium and vitamin D, and Charlson comorbidity index; Multivariable3 = Cox model adjusted for propensity score, age, sex, weight (only for patients with diabetes), height (only for patients with diabetes), and Charlson comorbidity index; Multivariable4 = Cox model adjusted for propensity score, age, sex, weight (only for patients with diabetes), height (only for patients with diabetes), previous fall injury without fracture, and Charlson comorbidity index.

a

Proportional hazards assumption not fulfilled.

b

Major osteoporotic fracture = fractured hip, vertebra, wrist, or collum chirurgicum.

Using a flexible parametric survival model (Fig. 2A, B), a significantly reduced risk of any fracture among patients without diabetes was observed for gastric bypass surgery patients the first year, but after the first year, the fracture risk gradually increased regardless of diabetic status.

Risk of any fracture after gastric bypass surgery. Risk of any fracture after gastric bypass in patients (A) with and (B) without diabetes shown using flexible parametric survival models because the Cox proportional hazards assumption was not fulfilled. The model was censored for death, emigration, or end of study (December 31, 2014). Horizontal lines represent 1.00 and the hazard ratio from the overall Cox model. One, two, three, and four different spline variables were tested and equally many time‐varying covariates. Here we show the best model fit for the event any fracture, which was for one spline variable (equivalent to Weibull model),⁽³⁸, ³⁹⁾ determined by using AIC. AIC = Akaike's information criterion.

The mortality risk was lower in gastric bypass surgery patients with and without diabetes, 47% and 45%, respectively, than in controls, using multivariable Cox models. In contrast, the risk of fall injury without a fracture was higher in gastric bypass patients with and without diabetes, 26% (HR 1.26; 95% CI, 1.04 to 1.52) and 24% (HR 1.24; 95% CI, 1.12 to 1.38), respectively, than in controls, using multivariable Cox models (Table 2).

During the first year after surgery, 75% of the gastric bypass patients without diabetes collected a prescription for calcium and vitamin D supplementation, compared to 3.1% among the controls (for patients with diabetes 74% and 4.3%, respectively). The proportion of gastric bypass surgery patients with supplementation dropped rapidly after the first year (Supporting Figs. 1 and 2). Starting the Cox model after 1 year and adjusting for age, sex, and updated fracture covariates (fracture‐free time, previous fracture, previous hip fracture, previous vertebral fracture, previous fall injury), the mean dose of calcium and vitamin D supplementation during the first postsurgery year was not significantly associated with any fracture, among patients with (HR 1.15; 95% CI, 0.89 to 1.5) and without (HR 1.09; 95% CI, 0.91 to 1.29) diabetes. When adding a time‐varying covariate of mean yearly calcium and vitamin D doses to the multivariable Cox model starting at baseline, no association between calcium and vitamin D supplementation and any fracture was found (patients with [HR 0.99; 95% CI, 0.76 to 1.29] and without [HR 1.03; 95% CI, 0.88 to 1.22] diabetes).

The risk of any fracture per 10‐kg weight loss was analyzed using a Cox model starting 1 year postsurgery. Among the gastric bypass patients with diabetes, there was no significant association with any fracture (Supporting Table 8). Gastric bypass patients without diabetes, had a lower risk of any fracture, HR 0.91 (95% CI, 0.85 to 0.98) per 10‐kg weight loss after 1 year, also in multivariable models (Supporting Table 8). Analyses of tertiles of weight loss after 1 year, revealed no association with risk of fracture for gastric bypass patients with diabetes. However, in gastric bypass patients without diabetes, the tertiles with medium and the greatest weight loss, each had 22% reduced risk of any fracture compared to the tertile with the least weight loss, after adjustment for covariates (Table 3). The risk of lower leg fracture was reduced by one‐half in patients in tertiles with medium and high weight loss compared to the tertile with the least weight loss. The risk of major osteoporotic fracture, upper limb fracture, and fall injury without fracture was not associated with tertile of weight loss (Table 3).

Risk of Fracture per Tertile of Weight Loss One Year Postsurgery

(A) Gastric bypass patients with diabetes	Low (≤29 kg)	Medium (30–39 kg)	High (≥40 kg)
Weight loss (kg), mean ± SD	23.0 ± 5.8	34.4 ± 2.8	49 ± 9.2
Any fracture
At risk after first year's censoring and dropouts, n12	1916	2124	1968
Any fracture, n (%)	53 (2.8)	65 (3.1)	59 (3.0)
Cox model starting 1 year postsurgery
Unadjusted, HR (95% CI)	1 [Ref.]	1.12 (0.78–1.62)	1.14 (0.78–1.65)
Multivariable1a, HR (95% CI)	1 [Ref.]	1.17 (0.81–1.70)	1.26 (0.85–1.85)
Multivariable5a, HR (95% CI)	1 [Ref.]	1.22 (0.84–1.76)	1.29 (0.87–1.90)
Covariates 1 year postsurgery
Female sex, n (%)	1366 (71.3)	1478 (69.6)	1187 (60.3)
Age (years), mean ± SD	50.8 ± 9.3	48.3 ± 10.4	46 ± 10.6
Weight (kg), mean ± SD	87.8 ± 17.1	85.5 ± 16.5	87.9 ± 18.6
Height (cm), mean ± SD	167.4 ± 9.5	169.4 ± 9.0	172.5 ± 9.3
Fracture‐free time (years), mean ± SD	23.9 ± 5.2	24.1 ± 4.8	24 ± 5.2
Any previous fracture, n (%)	139 (7.3)	139 (6.5)	149 (7.6)
Previous hip fracture, n (%)	4 (0.2)	0 (0.0)	0 (0.0)
Previous vertebral fracture, n (%)	7 (0.4)	11 (0.5)	11 (0.6)
Previous fall injury without fracture, n (%)	55 (2.9)	45 (2.1)	44 (2.2)
(B) Gastric bypass patients without diabetes	Low (≤33 kg)	Medium (34–43 kg)	High (≥44 kg)
Weight loss (kg), mean ± SD	26.8 ± 5.8	38.4 ± 2.8	52.3 ± 8.4
Any fracture
At risk after first year's censoring and dropouts, n12	7531	8454	7866
Any fracture, n (%)	198 (2.6)	163 (1.9)	151 (1.9)
Cox model starting 1 year postsurgery
Unadjusted, HR (95% CI)	1 [Ref.]	0.82 (0.66–1.01)	0.80 (0.64–1.01)
Multivariable1b, HR (95% CI)	1 [Ref.]	0.81 (0.65–1.00)	0.79 (0.63–1.00)
Multivariable5b, HR (95% CI)	1 [Ref.]	0.78 (0.64–0.96)	0.78 (0.63–0.96)
Covariates 1 year postsurgery
Female sex, n (%)	6464 (85.8)	7049 (83.4)	5514 (70.1)
Age (years), mean ± SD	44.1 ± 10.4	40.9 ± 10.2	37.7 ± 10.1
Weight (kg), mean ± SD	83 ± 15.8	80.3 ± 15.5	83.5 ± 17.4
Height (cm), mean ± SD	166 ± 8.1	168.4 ± 8.1	172.2 ± 9.0
Fracture‐free time (years), mean ± SD	24.3 ± 4.5	24.4 ± 4.5	24.4 ± 4.5
Any previous fracture, n (%)	400 (5.3)	455 (5.4)	432 (5.5)
Previous hip fracture, n (%)	5 (0.1)	1 (0.0)	7 (0.1)
Previous vertebral fracture, n (%)	31 (0.4)	25 (0.3)	17 (0.2)
Previous fall injury without fracture, n (%)	174 (2.3)	213 (2.5)	203 (2.6)
Lower leg
At risk after first year's censoring and dropouts, n12	7556	8494	7904
Lower leg fracture, n (%)	52 (0.7)	32 (0.4)	23 (0.3)
Cox model starting 1 year postsurgery
Unadjusted, HR (95% CI)	1 [Ref.]	0.59 (0.38–0.92)	0.45 (0.28–0.74)
Multivariable1b, HR (95% CI)	1 [Ref.]	0.57 (0.36–0.90)	0.45 (0.27–0.77)
Multivariable5b, HR (95% CI)	1 [Ref.]	0.57 (0.36–0.90)	0.46 (0.27–0.78)
Major osteoporotic fracture13
At risk after first year's censoring and dropouts, n12	7556	8482	7898
Major osteoporotic fracture, n (%)	67 (0.9)	64 (0.8)	52 (0.7)
Cox model starting 1 year postsurgery
Unadjusted, HR (95% CI)	1 [Ref.]	0.91 (0.64–1.28)	0.79 (0.55–1.13)
Multivariable1b, HR (95% CI)	1 [Ref.]	1.01 (0.71–1.44)	0.96 (0.65–1.42)
Multivariable5b, HR (95% CI)	1 [Ref.]	1.03 (0.72–1.47)	0.97 (0.65–1.44)
Upper limb
At risk after first year's censoring and dropouts, n12	7551	8479	7903
Upper limb fracture, n (%)	62 (0.8)	59 (0.7)	53 (0.7)
Cox model starting 1 year postsurgery
Unadjusted, HR (95% CI)	1 [Ref.]	0.91 (0.64–1.31)	0.88 (0.61–1.26)
Multivariable1b, HR (95% CI)	1 [Ref.]	1.09 (0.75–1.57)	1.19 (0.80–1.77)
Multivariable5b, HR (95% CI)	1 [Ref.]	1.09 (0.76–1.58)	1.18 (0.80–1.76)
Fall injury without fracture
At risk after first year's censoring and dropouts, n12	7529	8454	7861
Fall injury without fracture, n (%)	186 (2.5)	183 (2.2)	177 (2.3)
Cox model starting 1 year postsurgery
Unadjusted, HR (95% CI)	1 [Ref.]	0.93 (0.76–1.14)	0.96 (0.78–1.18)
Multivariable1b, HR (95% CI)	1 [Ref.]	0.89 (0.72–1.09)	0.85 (0.68–1.07)
Multivariable5b, HR (95% CI)	1 [Ref.]	0.89 (0.72–1.09)	0.85 (0.68–1.07)

Multivariable1a = adjustment for age, sex, weight, and height; Multivariable1b = adjustment for age and sex; Multivariable5a = adjustment for age, sex, weight, height, fracture‐free time, any previous fracture, previous hip fracture, previous vertebral fracture, previous vertebral fracture, and previous fall injury without fracture; Multivariable5b = adjustment for age, sex, fracture free time, any previous fracture, previous hip fracture, previous vertebral fracture, previous vertebral fracture, and previous fall injury without fracture.

a

Preanalysis censoring means that patients with events, death, emigration, or end of study within a year after surgery were not included in the analyses.

b

Major osteoporotic fracture = fractured hip, vertebra, wrist, or collum chirurgicum.

Discussion

In this retrospective cohort study, obese patients treated with gastric bypass surgery had an overall increased risk of any, major osteoporotic, and hip fracture during the follow‐up period of 3.1 years irrespective of diabetes status. Fracture risk appeared to increase with time.

The mechanisms behind the gastric bypass‐induced fracture risk are still unknown,⁽⁹⁾ but mechanical unloading due to weight loss and secondary hyperparathyroidism due to vitamin D deficiency have been the most commonly suggested explanations. The present study is the first to investigate the relationship between gastric bypass surgery–induced weight loss, degree of calcium and vitamin supplementation following surgery, and fracture risk. It was an unexpected finding that greater weight loss was not associated with greater increase in risk of any fracture. This suggests that bone loss and increased fracture risk after surgery are not due to reduced mechanical loading. This is supported by reports of continued bone loss up to 3 and 5 years after gastric bypass surgery,⁽³⁰, ³¹⁾ long after weight loss has ceased. No association between fracture risk and calcium and vitamin D supplementation was found, suggesting that inadequate calcium and vitamin D intake and absorption may not explain the increased fracture risk. This finding is supported by previous studies reporting bone loss and increased bone turnover after gastric bypass without signs of secondary hyperparathyroidism.⁽⁹⁾

The present study is the first to demonstrate an increased risk of fall injury following gastric bypass surgery. After gastric bypass, the level of physical activity is increased,⁽³²⁾ which is likely to result in increased exposure and therefore a higher risk of falls. Thus, our results indicate that the increased fracture risk following gastric bypass surgery could be, at least to some extent, due to increased susceptibility to falls or is dependent on other mechanism (than weight loss and secondary hyperparathyroidism) affecting skeletal metabolism, perhaps including alterations in adipokines, gonadal steroids, or gut‐derived hormones, leading to bone loss.⁽³³⁾

We found an increased risk of any fracture and a reduced risk of lower leg fracture, which is consistent with the findings of Rousseau and colleagues,⁽¹⁰⁾ who compared a group of 12,676 patients undergoing bariatric surgery to obese controls; however, they lacked statistical power to analyze gastric bypass specifically. By comparing gastric bypass patients with gastric banding patients, Yu and colleagues⁽¹¹⁾ minimized the impact of indication bias, though introducing a reference group with another bariatric procedure, with unknown fracture risk. Their inclusion covered only the morbidly obese (≥40 kg/m²), but with that limitation in mind, their findings of an increased fracture risk, gradually increasing in time, are consistent with our findings. Shanbhogue and colleagues⁽³⁴⁾ showed that despite weight stabilization and maintenance of metabolic parameters, bone loss and deterioration in bone strength continued and were substantial in the second year following gastric bypass surgery. This supports our finding that fracture risk does not appear to be dependent on weight loss, and that the fracture risk increases with time. Other smaller studies including all bariatric procedures, suggest an increasing fracture risk over time, similar to our results.⁽¹², ¹³, ¹⁴⁾ Gastric bypass was associated with reduced mortality, which was in line with previous findings.⁽⁸, ³⁵⁾

The present study has several strengths. It is the only sufficiently powered study investigating fracture risk in gastric surgery patients and obese controls and the first to adequately consider diabetes status, which affects fracture risk and is affected by gastric bypass. It is the first study to assess risk of fall injury and whether weight loss or degree of calcium and vitamin D supplementation is associated with fracture risk. Extensive propensity score matching was performed to obtain well‐balanced control groups and a wide range of comorbidity data to control for confounders was used.

This study has several limitations. First, the identification of control patients without diabetes using an obesity diagnosis, which likely only identifies a small minority of obese patients and with unknown weight. Because obesity and fracture risk are associated in a site‐specific manner,⁽³⁶⁾ this could have an impact on the obtained result. However, the obese patients without diabetes were also matched according to a number of other metabolic diagnoses and medication, which minimizes the risk of potential bias. Second, accounting only for prescribed and not over‐the‐counter sales of calcium and vitamin D supplementation. However, prescribed calcium and vitamin D supplementation is discounted in Sweden, which probably explains why it accounted for 97.9% of the 78 million defined daily doses (DDDs) sold in 2017 (ATC code A12AX).⁽³⁷⁾ Third, densitometry, smoking, and heredity data were not available. Fourth, for the gastric bypass patients both with and without diabetes, the mortality was almost one‐half compared to their respective controls. If gastric bypass patients survive longer, they would be at risk for a longer time period, possibly resulting in falsely increased fracture risk. However, the mortality was reduced from low levels (2.7% and 1.2%, respectively) and there was not a large difference in time at risk between the groups. Thus, the competing risk of death is not likely to have a major impact on the results. Fifth, even though fracture diagnoses are captured in the NPR with high accuracy, some fractures (eg, rib fractures) may not be verified by X‐rays, which could have resulted in some false‐positive fractures included in the analyses.

In conclusion, gastric bypass surgery is associated with an increased fracture risk, which appears to be increasing with time and not associated with degree of weight loss or calcium and vitamin D supplementation following surgery. An increased risk of fall injury was seen after surgery, which could contribute to the increased fracture risk.

Disclosures

All authors have completed the ICMJE uniform disclosure form at www.icmje.org/coi_disclosure.pdf and declare: no support from any organization for the submitted work. KFA has received lecture fees from Lilly, Meda/Mylan, and Amgen, all outside the submitted work. BE reports personal fees for lectures, board memberships, or consultancies from Amgen, AstraZeneca, Boehringer Ingelheim, Eli Lilly, Merck Sharp & Dohme, Mundipharma, Navamedic, NovoNordisk, RLS Global, and Sanofi, all outside the submitted work. IN reports personal fees from Baricol Bariatrics AB, Sweden, for consulting and lectures, outside the submitted work. MW, ES, HW, and DL state that they have no conflicts of interest. ML has received lecture or consulting fees from Amgen, Lilly, Meda, UCB Pharma, Renapharma, Radius Health, and Consilient Health, all outside the submitted work.

Acknowledgments

This study was funded by the Research fund at Skaraborg Hospital Skövde, Sweden, the ALF/LUA grant from the Sahlgrenska University Hospital, the Swedish Research Council (VR) and Gustaf V:s och Drottning Victorias Frimurarstiftelse. The authors affirm that the manuscript is an honest, accurate, and transparent account of the study being reported, that no important aspects of the study have been omitted, and that any discrepancies from the study as planned (and, if relevant, registered) have been explained.

Authors’ roles: All authors contributed to the study design, the interpretation of the results, and reviewed the manuscript. DL assembled the database. KFA and ML performed the statistical analyses and wrote the first draft of the manuscript and are guarantors.