Adipose Tissue Insulin Resistance Is Longitudinally Associated With Adipose Tissue Dysfunction, Circulating Lipids, and Dysglycemia: The PROMISE Cohort

Adipose tissue is an important endocrine organ that plays a critical role in both glucose and lipid metabolism (1). Under normal conditions, insulin action at the level of the adipose tissue stimulates glucose uptake and triglyceride (TG) synthesis while suppressing lipolysis, which therefore inhibits TG hydrolysis and nonesterified fatty acid (NEFA) and glycerol release (2). Adipose tissue insulin resistance is defined as a diminished antilipolytic effect of insulin resulting in elevated circulating plasma levels of NEFA (3). Persistent increases in circulating NEFAs have been associated with a number of downstream complications, including impaired muscle signaling, hepatic gluconeogenesis, and impaired glucose-stimulated insulin response (2,4).

The current gold-standard method for the measurement adipose tissue insulin resistance in humans requires tracer-dilution techniques during continuous insulin infusion protocols, such as the hyperinsulinemic-euglycemic clamp combined with [²H₅]glycerol tracer (5). These methods, however, pose challenges for large clinical and epidemiological studies because of cost, invasiveness, and labor intensity. To overcome these limitations, a number of indirect methods have been developed. The adipose tissue insulin resistance index (Adipo-IR) has been proposed as a simple surrogate measure derived from the product of fasting insulin and NEFA concentrations (6). This index has been validated among normal-weight and obese adults (5,7), demonstrating a strong positive correlation against the gold-standard tracer-based clamp technique (r = 0.86, P < 0.001).

To date, a limited number of studies have reported that Adipo-IR is associated with obesity (8), prediabetes (9,10), type 2 diabetes (11), and adipose tissue dysregulation marked by elevated levels of circulating adipokines (12). Although these studies have suggested an important role for adipose tissue insulin resistance in the development of cardiometabolic abnormalities, the majority of published studies were limited because of their cross-sectional designs and small sample sizes.

To our knowledge, there are no existing studies that have assessed baseline adipose tissue insulin resistance status in relation to longitudinal changes in inflammation, circulating lipids, or incident dysglycemia. Therefore, our objective in the current study was to assess the association of the adipose tissue insulin resistance index (Adipo-IR) with longitudinal changes in biomarkers of adipose tissue function (adiponectin and soluble CD163 [sCD163], which is a novel biomarker of adipose tissue macrophage activation [13]), clinically relevant circulating lipids (total cholesterol, HDL, LDL, and TG), systematic inflammation (interleukin-6 [IL-6] and tumor necrosis factor-α [TNF-α]), and incident dysglycemia among participants at high risk for type 2 diabetes in the Prospective Metabolism and Islet Cell Evaluation (PROMISE) cohort.

The current analysis included participants from the PROMISE cohort study, which is an ongoing longitudinal observational study based in Toronto and London, Ontario, Canada. Recruitment and baseline examinations took place from 2004 to 2006 and included individuals with one or more risk factors for type 2 diabetes: obesity, hypertension, family history of diabetes, birth of a macrosomic infant, and/or a history of gestational diabetes mellitus (n = 736) (14). Research ethics approval was obtained from Mount Sinai Hospital and Western University. Written informed consent was obtained from all participants. At baseline and at each 3-year follow-up visit thereafter, participants completed standard health and lifestyle questionnaires and underwent detailed anthropometric measurements and metabolic characterization. Serum circulating lipids, glucose measurements, and glycemic status were assessed at baseline and every follow-up visit up to 9 years. Serum adiponectin, sCD163, IL-6, and TNF-α were measured at baseline and 3 and 6 years. NEFA was measured at the baseline visit. For the current study, we analyzed data on 468 participants after excluding those with baseline type 2 diabetes, no follow-up visits, or missing data on key variables required for the analysis (see Supplementary Fig. 1 for detailed participant selection).

Anthropometric and blood pressure measurements were collected at each clinic visit while following standard procedures. Measurements were conducted twice and averaged. Blood pressure was evaluated using an automated sphygmomanometer on the right arm with the participant seated after resting for 5 min. Height, weight, and waist circumference were assessed through standardized procedures. Height and weight were used to calculate BMI.

Sociodemographic and lifestyle risk factors were assessed using structured, standardized questionnaires at each clinic visit. The Modifiable Activity Questionnaire was administered to determine physical activity levels; this instrument collected information on leisure and occupational activity over the past year (15). Each reported activity from the Modifiable Activity Questionnaire was weighted by its metabolic intensity to estimate MET-h per week.

Blood samples were collected after an 8- to 12-h overnight fast at baseline and at each follow-up visit. At each examination, a 75-g oral glucose tolerance test was administered after the collection of fasting blood samples, with additional blood samples drawn at 30 and 120 min. Blood samples were aliquoted and frozen at −80°C at the Banting and Best Diabetes Centre Core Laboratory at Mount Sinai Hospital. Glucose and insulin concentrations were measured in fasting, 30-min, and 120-min blood samples at baseline and at every follow-up visit. Insulin was assessed using the Elecsys 1010 Immunoassay Analyzer (Roche Diagnostics, Basel, Switzerland) and electrochemiluminescence immunoassay. Glucose was determined using an enzymatic hexokinase assay (Roche Modular, Roche Diagnostics, Mississauga, Ontario, Canada).

The total serum NEFA fraction was quantified using fasting samples collected at baseline. Details on the methodology have been previously described (16). Measurements were conducted on samples that had been stored at −80°C for 4–7 years and had not undergone any freeze-thaw cycles. Heptadecanoic acid (17:0; Nu-Chek Prep, Inc., Elysian, MN) was used as an internal standard before total lipid extraction using the methodology of Folch et al. (17). Thin-layer chromatography isolated serum lipid fractions, including the NEFA pool that was collected and converted to fatty acid methyl esters using boron trifluoride in methanol. Fatty acid methyl esters were separated and quantified using a Varian 430-GC gas chromatograph (Varian, Lake Forest, CA) equipped with a Varian FactorFour capillary column and a flame ionization detector. Fatty acid concentrations (nmol/mL) were calculated by proportional comparison of gas chromatography peak areas to that of the internal standards (18).

Adiponectin, IL-6, sCD163, and TNF-α concentrations were measured using fasting serum samples collected at baseline and 3- and 6-year clinic visits. All measurements were performed at the Keenan Research Centre for Biomedical Sciences, St. Michael’s Hospital. Adiponectin was measured using Meso Scale Discovery singleplex assay kits. IL-6 and TNF-α were measured using Meso Scale Discovery multiplex kits. sCD163 was measured using Quantikine ELISA (R&D Systems, Emeryville, CA), a solid phase sandwich ELISA that has a sensitivity of 0.613 ng/mL (13). Samples were run in duplicate following the manufacturer’s protocol.

Total cholesterol, HDL, and TG were quantified at all clinic visits using Roche Modular enzymatic colorimetric tests. LDL was determined using the Friedewald equation (19).

Adipo-IR was calculated as the product of fasting insulin and NEFA concentrations at baseline (20). Dysglycemia was defined as type 2 diabetes, impaired fasting glucose (IFG), and/or impaired glucose tolerance (IGT). World Health Organization 2006 guidelines were used to classify IFG, IGT, and type 2 diabetes status (21). IFG was defined by fasting blood glucose measures between 6.1 and 6.9 mmol/L and IGT as fasting glucose <7.0 mmol/L and 2-h oral glucose tolerance test blood glucose ≥7.8 and <11.1 mmol/L. Type 2 diabetes was classified on the basis of physician diagnosis, use of diabetes medication, or having a fasting plasma glucose level of ≥7.0 mmol/L or a 2-h plasma glucose level ≥11.1 mmol/L (21).

Whole-body insulin resistance was determined using the reciprocal Matsuda Insulin Sensitivity Index (1/ISI) (22). Hepatic insulin resistance was evaluated using the updated HOMA for insulin resistance calculator on the basis of fasting serum concentrations of both glucose and insulin (HOMA2-IR) (23). Additional surrogate indices of insulin resistance were determined using the TG:HDL ratio (24), calculated by dividing TG by HDL, and the waist circumference-TG index (WTI) (25), calculated as the product of waist circumference and TG.

Descriptive characteristics were assessed by tertiles of Adipo-IR at baseline. Continuous variables were described as mean ± SD or median with interquartile range for normally and nonnormally distributed variables, respectively. Categorical variables were presented as a number and percent. P values for continuous variable differences by category were determined by one-way ANOVA and Kruskal-Wallis tests for normally and nonnormally distributed variables. P values for categorical variable differences were determined using χ² and Fisher exact tests. Baseline univariate associations of Adipo-IR and metabolic markers were assessed through Spearman correlations.

Longitudinal associations were assessed using data collected from baseline and up to 9 years of follow-up visits. For our primary analyses, generalized estimating equation (GEE) models (26) were used to evaluate the association of baseline Adipo-IR with longitudinal changes in the outcome measures. The GEE estimates were interpreted as an expected percent difference in the outcome variable for every SD increase in the predictor variable. Our main outcomes of interest included adiponectin, sCD163, total cholesterol, HDL, LDL, TG, IL-6, TNF-α, fasting glucose, 2-h glucose, glucose area under the curve (AUC), and incident dysglycemia.

GEE is an extension of the generalized linear model and provides a longitudinal population estimate using a semiparametric approach. An important strength in the use of GEE is its ability to accommodate missing values, thus maximizing statistical power. Furthermore, it works under the assumption that within-subject measurements are correlated. For each GEE model, the autoregressive of order 1 correlation matrix was selected. GEE models were adjusted for years since baseline (time) age, sex, ethnicity, family history of type 2 diabetes, physical activity, and smoking. The variables baseline age, sex, and ethnicity were classified as time independent because they were only either measured at baseline or did not change with each follow-up data inclusion.

We assessed the relationship between baseline Adipo-IR and incident dysglycemia over the 9-year follow-up period. For this analysis, participants with baseline dysglycemia (identified as those with IGT or IFG, as baseline type 2 diabetes had already been excluded) were excluded in the modeling (n = 26). Odds ratios (ORs) per SD increase in Adipo-IR were calculated by using the “logit” link function within the GEE tests and adjusted for selected covariates as described above.

All analyses were performed using R 3.6.1 statistical software. GEE models were conducted using the R geepack package (https://cran.r-project.org/web/packages/geepack/index.html). Statistical significance was set at P < 0.05.

Table 1 shows baseline demographic characteristics of PROMISE participants by increasing Adipo-IR tertiles. Both waist circumference and BMI differed across tertiles (P < 0.001), with the top tertile having the highest mean for waist circumference and BMI (108.21 cm and 35.45 kg/m², respectively). Additionally, measures of glucose (fasting glucose, 120-min glucose, and glucose AUC), fasting insulin, and total NEFA were significantly different across increasing tertiles (all P < 0.001), with the highest values in the top tertile. Total cholesterol, LDL, and TG concentrations differed significantly across Adipo-IR tertiles, with the highest concentrations at the top Adipo-IR tertile for each biomarker. Similarly, there were significant differences in sCD163 and IL-6 concentrations across Adipo-IR tertiles, with the highest tertiles presenting elevated concentrations of each biomarker. Conversely, greater physical activity, measured through MET scores, and adiponectin and HDL concentrations were observed in the lowest tertile of Adipo-IR. There were no significant differences across Adipo-IR tertiles for age, sex, ethnicity, smoking status, and TNF-α.

Spearman correlations between baseline Adipo-IR and baseline metabolic parameters are outlined in Table 2. There were significant positive associations of Adipo-IR with BMI and waist circumference (r = 0.56 and 0.52, respectively, both P < 0.001). Fasting glucose, 2-h glucose, and glucose AUC were all positively correlated with Adipo-IR (r = 0.33, 0.38, and 0.38, respectively, all P < 0.001). sCD163 presented a significant positive correlation with Adipo-IR (r = 0.21, P < 0.001), while adiponectin was inversely associated with Adipo-IR (r = −0.29, P < 0.05). IL-6 showed a significant positive association with Adipo-IR (r = 0.23). Total cholesterol, LDL, and TG were all positively correlated with Adipo-IR (r = 0.12, 0.10, and 0.46, respectively, all P < 0.05). Adipo-IR showed a significant inverse association with physical activity (r = −0.15) and HDL (r = −0.34). Age and TNF-α were not correlated with Adipo-IR at baseline.

The main fully adjusted GEE models are presented in Figs. 1–3. Figure 1 displays the association of Adipo-IR with longitudinal changes in fasting glucose, 2-h glucose, and glucose AUC. In all analyses, baseline Adipo-IR showed a positive association with all glucose measures (percent difference per SD increase of Adipo-IR: fasting glucose 6.26% [95% CI 3.26–9.35], 2-h glucose 20.01% [14.28–26.15], glucose AUC 36.71% [22.11–53.06]). Additional adjustment for waist circumference, TG:HDL, and WTI separately slightly attenuated the associations, but they remained significant.

Figure 2 displays the association of baseline Adipo-IR with incident dysglycemia over the 9-year follow-up period. Following adjustment for time, baseline age, sex, family history of diabetes, ethnicity, physical activity, and smoking status, participants with higher levels of Adipo-IR at baseline had an increased risk of incident dysglycemia (OR 1.59 [95% CI 1.09–2.31] per SD increase in Adipo-IR). The addition of waist circumference, TG:HDL, and WTI separately to the model did not change the overall direction of the association or statistical significance.

Baseline Adipo-IR was associated with adipose tissue dysfunction and dyslipidemia but not systemic inflammation (Fig. 3). Markers specific to adipose tissue function, namely adiponectin and sCD163, demonstrated significant longitudinal associations with baseline Adipo-IR (adiponectin −4.20% [95% CI −6.40 to –1.95], sCD163 4.36% [1.73–7.06]) (Fig. 3A). Additional adjustment for waist circumference attenuated the association between Adipo-IR and adiponectin but not sCD163. Similar associations were shown with the addition of TG:HDL and WTI separately. Furthermore, GEE results showed a significant inverse association between baseline Adipo-IR and HDL (−3.87% [95% CI −5.15 to −2.57]) and a significant positive association with TG (9.26% [5.01–13.69]) (Fig. 3B). While the association of baseline Adipo-IR with both total cholesterol and LDL was not significant, the direction of the association was as hypothesized. Biomarkers of systemic inflammation, specifically IL-6 and TNF-α, showed a positive trend with baseline Adipo-IR, although the estimates did not reach significance (Fig. 3C).

We additionally evaluated the impact of BMI on the association of baseline Adipo-IR with longitudinal changes of our outcomes (see Supplementary Figs. 2–4). Similar to the results noted by the adjustment for waist circumference, adjustment for BMI attenuated the association between Adipo-IR and adiponectin; however, it did not change the significance of any of the other outcome variables. This was consistent with the addition of TG:HDL to the BMI models.

Furthermore, we compared the associations of various insulin resistance indices with incident dysglycemia (Supplementary Fig. 5). Fasting insulin, 1/ISI, HOMA2-IR, and Adipo-IR were all significantly associated with incident dysglycemia over time. The CIs of each of these estimates overlapped, indicating no significant difference between them.

To our knowledge, this observational study is the first to evaluate the longitudinal relationship of adipose tissue insulin resistance with measures of adipose tissue function, circulating lipids, and dysglycemia. We showed that in a Canadian population of adults at risk for type 2 diabetes, Adipo-IR (a validated surrogate measure of adipose tissue insulin resistance) was significantly associated with adipose tissue dysfunction as reflected by a decrease in adiponectin and an increase in sCD163 over time. Similarly, elevated baseline levels of Adipo-IR were related to longitudinal changes in HDL and TG. Furthermore, our prospective analysis showed a significant association between Adipo-IR and incident dysglycemia over time. Our findings did not, however, show a significant longitudinal relationship between Adipo-IR and two markers of systemic inflammation, IL-6 and TNF-α.

Adipose tissue is a complex endocrine organ comprising primarily adipocytes, which are responsible for energy storage in the form of TGs (27). In lean, healthy individuals, adipocytes are small in size and function to promote metabolic homeostasis (28). In obese individuals, however, adipocytes undergo hypertrophy and hyperplasia, triggering the recruitment and activation of M1 (classically activated) macrophages, which release proinflammatory cytokines into the circulation, resulting in chronic subclinical inflammation and compromised insulin action at the level of the adipose tissue (28,29). Persistent adipose tissue insulin resistance has been associated with downstream complications, including impaired muscle insulin signaling and promotion of liver gluconeogenesis (30).

In our study, we assessed the association between Adipo-IR and adiponectin, an adipokine known for it its anti-inflammatory and antidiabetic properties (31). Adiponectin mediates the insulin sensitizing effect in tissues through the respective tissues’ receptors (32). A number of cross-sectional studies have investigated the relationship of Adipo-IR with adiponectin in adolescent (8,9) and adult populations (12). These studies reported that Adipo-IR was inversely associated with adiponectin. Our study extends the current literature by documenting a longitudinal decline in adiponectin with elevated baseline Adipo-IR.

We also evaluated the relationship between Adipo-IR and sCD163, a novel biomarker of adipose tissue macrophage activation (33). sCD163 is the ectodomain of the hemoglobin-haptoglobin receptor CD163, which is exclusively found on macrophages and monocytes. CD163⁺ macrophages are highly expressed in obese adipose tissue, and increased circulating sCD163 has been significantly associated with a higher risk of incident type 2 diabetes (34) and related underlying disorders, including whole-body and hepatic insulin resistance (13,35) and β-cell dysfunction (13,36). Only one previous study has evaluated the relationship of sCD163 with Adipo-IR (37). In an analysis of 40 patients without diabetes with nonalcoholic fatty liver disease, Rosso et al. (37) showed that sCD163 was significantly correlated with Adipo-IR (r = 0.32, P = 0.042). These results are consistent with our cross-sectional findings, where we saw a similar moderate correlation of r = 0.21 (P < 0.001). We are the first, however, to show the longitudinal association between baseline Adipo-IR and sCD163. We found that elevated baseline Adipo-IR was associated with a significant longitudinal increase in sCD163, suggesting that insulin resistance at the level of adipose tissue may drive worsening adipose tissue inflammation.

Several studies to date have demonstrated a significant association between Adipo-IR and clinically relevant circulating lipids (11,38–40). Among adolescents (11) and women with polycystic ovarian syndrome (39), Adipo-IR was positively correlated with total cholesterol, LDL, and TG and inversely correlated with HDL, observations that are consistent with our cross-sectional findings. A recent study showed that adults with dysfunctional white adipose tissue, defined by high Adipo-IR and low adiponectin concentrations, had significantly higher levels of TG and lower levels of HDL compared with subjects classified as having functional white adipose tissue; however, there was no significant association with total cholesterol or LDL (38). Furthermore, Jiang et al. (40) showed that there was no significant difference in LDL levels among women classified as presenting Adipo-IR compared with those without Adipo-IR. Our longitudinal analyses extend these cross-sectional studies by showing that Adipo-IR was significantly associated with HDL and TG but not with total cholesterol or LDL over time.

Although we identified significant associations between Adipo-IR and biomarkers of adipose tissue dysfunction, HDL, and TG, we did not see a longitudinal relationship with biomarkers of systemic inflammation, specifically IL-6 and TNF-α. We noted a significant relationship in our cross-sectional analysis between Adipo-IR and IL-6, which was similar to previous studies (8,9). However, the lack of a significant association in the longitudinal analysis may be due to the low circulating concentration of IL-6. Furthermore, to our knowledge, we are the first to investigate the association between Adipo-IR and TNF-α. The lack of a significant association of may be due to the short half-life of TNF-α (33). More studies are needed to better understand the relationship of these biomarkers with Adipo-IR.

In insulin-sensitive adipose tissue, insulin promotes glucose uptake, a response that is impaired in the context of adipose tissue insulin resistance. The findings of our study, which documented an association of Adipo-IR with longitudinal declines in glucose tolerance, are consistent with this mechanism. Although for some of our assessments the percent difference was small, particularly for fasting glucose, the absolute values of these changes may be clinically important. Several cross-sectional studies have shown a positive association of Adipo-IR with fasting and 2-h glucose concentrations (9,11,41) Furthermore, previous research has shown a positive correlation between Adipo-IR and dysglycemia status, including studies in Chinese adults (42), women with polycystic ovary syndrome (39), and adolescents (11). In a cross-sectional study by Gastaldelli et al. (41), Adipo-IR was increased with worsening of glucose tolerance status from normal glucose tolerance to IGT to type 2 diabetes in adults. In an experimental study by Defronzo et al. (43), subjects with IGT after treatment with pioglitazone who reverted to normal glucose tolerance demonstrated an improvement in Adipo-IR, while those who transitioned from IGT to type 2 diabetes showed an increase Adipo-IR. Additionally, weight loss caused by exenatide treatment among morbidly obese adults resulted in improved glucose response and improved hepatic and adipose tissue insulin resistance (44). The current study adds to the current literature by demonstrating a positive association of baseline Adipo-IR with higher concentrations of glucose, and an increased risk of dysglycemia incidence, over time.

This study has several notable strengths. We document the relationship of Adipo-IR with longitudinal changes in adipose tissue and hepatic function as well as incident dysglycemia. We used data from PROMISE, a well-characterized ongoing longitudinal cohort study of subjects at risk for type 2 diabetes development. Detailed clinical assessments from baseline and multiple follow-up visits up to 9 years allowed for longitudinal analyses of outcomes, including adipose tissue function, hepatic function, and dysglycemia. In addition, our longitudinal analyses were conducted using GEE statistical models, which allowed for the maximum sample size to be retained. As well, we were able to adjust for a number of important covariates in our analyses.

Conversely, there are a number of limitations to consider. Our assessment of adipose tissue insulin resistance did not use the gold-standard approach, although Adipo-IR, a validated proxy, was used. NEFA concentrations were only quantified at the baseline visit; therefore, we were not able to assess the impact of changes over time in Adipo-IR. Because of the inclusion of fasting insulin in the calculation of both Adipo-IR and widely used insulin sensitivity indices, we were limited to proxy measures of insulin resistance that did not include insulin measures in their calculation, specifically TG:HDL and WTI, to assess the degree to which our results were influenced by ambient whole-body insulin resistance. The addition of these insulin resistance measures did not change the overall direction or significance of our findings. Our analysis of Adipo-IR and other indices of insulin resistance (fasting insulin, 1/ISI, and HOMA2-IR) with incident dysglycemia showed similar point estimates, with overlapping CIs. However, it is important to recognize that all these indices are surrogate (proxy) measures and are thus misclassified to a degree compared with their gold-standard measures. The question of whether adipose tissue insulin resistance adds additional predictive information beyond whole-body insulin resistance will need to be addressed in future studies that use definitive measures of these phenotypes, such as tracer protocols (adipose tissue insulin resistance) and euglycemic-hyperinsulinemic clamp studies (insulin sensitivity). Given the observational design, we are not able to confirm causality because of the potential for residual confounding. As well, the population assessed was restricted to those at risk for type 2 diabetes; therefore, the generalizability of our findings may be limited to individuals with similar characteristics as participants in PROMISE.

In conclusion, overall, our findings suggest that adipose tissue insulin resistance may be an important early contributor to downstream metabolic complications, which are known to increase the risk of cardiometabolic diseases, including type 2 diabetes. Our results extend the existing literature by documenting that Adipo-IR was longitudinally associated with adipose tissue dysfunction, HDL, and TG, important early pathophysiological phenotypes of type 2 diabetes and related disorders. In addition, we demonstrated that Adipo-IR was associated with an increased risk of incident dysglycemia. Additional research is needed to further clarify the role of adipose tissue insulin resistance, including studies with longer follow-up periods and repeated measures of this trait.

Acknowledgments. The authors thank the research nurses and staff for expert technical assistance and dedication in their work for PROMISE. From the Leadership Sinai Centre for Diabetes, Mount Sinai Hospital, the authors thank Jan Neuman, Paula Van Nostrand, Stella Kink, Nicole Rubio, and Annette Barnie. Additionally, the authors thank Sheila Porter, Mauricio Marin, Marnie Orcutt, and Sue Miller of the Centre for Studies in Family Medicine, Western University, for expert technical assistance and dedication in their work for PROMISE. As well, the authors greatly appreciate the dedication of all of the participants involved in the PROMISE study.

Funding. PROMISE was supported by operating grants from Diabetes Canada and the Canadian Institutes of Health Research. Z.S.-A. was funded by the Canadian Institutes of Health Research Frederick Banting and Charles Best Canada Graduate Master’s and Doctoral Scholarships, Ontario Graduate Scholarships, and the University of Toronto Banting and Best Diabetes Centre Novo Nordisk Studentships. R.R. holds the Boehringer Ingelheim Chair in Beta-Cell Preservation, Function and Regeneration at Mount Sinai Hospital.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. Z.S.-A. analyzed and interpreted the data and prepared the manuscript. Z.S.-A., S.B.H., B.Z., and A.J.H. designed the study and reviewed and edited the manuscript. P.W.C., R.P.B., R.R., and D.J.A.J. reviewed and edited the manuscript. A.J.H. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Prior Presentation. Parts of this study were presented at Obesity Week 2019, Las Vegas, NV, 3–7 November 2019.