Dysfunctional Adiposity and the Risk of Prediabetes and Type 2 Diabetes in Obese Adults James A de Lemos, MD University of Texas Southwestern Medical Center
Mar 26, 2015
Dysfunctional Adiposity and the Risk of Prediabetes and Type 2
Diabetes in Obese Adults
James A de Lemos, MD
University of Texas Southwestern Medical Center
Study Rationale• Increasing rates of diabetes and obesity have
contributed to a slowed decline in CVD.1
• Diabetes development is heterogeneous and BMI does not adequately discriminate risk.2
• Previous studies – Cross sectional with little longitudinal data– Not focused on obese– Ethnically homogeneous– Limited application of advanced imaging
• Factors that differentiate obese persons who will develop prediabetes and diabetes from those who will remain metabolically healthy have not been well characterized.
1. Wijeysundera et al. JAMA. 2010;303:1841-472. Despres JP. Circulation. 2012;126:1301-13
Obesity is Heterogeneous
Obesity is Heterogeneous
Diabetes
Diabetes
Diabetes
DiabetesPrediabetes
PrediabetesPrediabetes
Obesity is Heterogeneous
Study Aim
Investigate associations between markers of general and dysfunctional adiposity and risk of incident prediabetes and diabetes in multiethnic cohort of obese adults.
Imaging
Cohort F/U
GeneticMarkers
The Dallas Heart Study
Biomarkers
Representative Population Sample
EBCTCardiac MRIAortic MRIMRI AbdomenDEXA
n=6101
n3500
n3000
2002
1
2000
Year 2DHS-1 Exam
2007
6 83 4 5 7 9
2009
DHS-2 Exam
Methods
N=732BMI ≥ 30
No DMNo CVD
• Body Composition and Abdominal Fat Distribution
MRI and DEXA
• Blood Biomarkers
• Cardiac Structure and Function CT and MRI
Incident Diabetes
• FBG ≥ 126
• non-FBG ≥ 200
• Hgb A1C ≥ 6.5
Weight Gain
Subgroup with FBG<100 (n=512) Incident Prediabetes
Mean Age 4365% Women71% Nonwhite
Baseline Measurements: Body Composition
Abdominal MRIPatient #1: 21 AA Female
BMI = 36Patient #2: 59 W Male
BMI = 31
Results – Overall CohortMedian (IQR) or % No Diabetes (n=648) Incident Diabetes (n=84) P value
Family History of Diabetes 42% 63% <0.001
Waist/Hip ratio 0.91 (0.85, 0.97) 0.95 (0.90, 1.00) <0.001
Systolic Blood Pressure (mmHg)
123 (115, 134) 131 (122, 144) <0.001
Glucose (mg/dL) 93 (87, 100) 101 (92, 114) <0.001
Fructosamine (µmol/L) 199 (188, 210) 211 (196, 224) <0.001
Triglycerides (mg/dL) 99 (70, 146) 124 (90, 187) 0.001
Results – Overall CohortMedian (IQR) or % No Diabetes (n=648) Incident Diabetes (n=84) P value
Family History of Diabetes 42% 63% <0.001
Waist/Hip ratio 0.91 (0.85, 0.97) 0.95 (0.90, 1.00) <0.001
Systolic Blood Pressure (mmHg)
123 (115, 134) 131 (122, 144) <0.001
Glucose (mg/dL) 93 (87, 100) 101 (92, 114) <0.001
Fructosamine (µmol/L) 199 (188, 210) 211 (196, 224) <0.001
Triglycerides (mg/dL) 99 (70, 146) 124 (90, 187) 0.001
Lower Body Fat (kg) 12.6 (9.6, 16.3) 11.2 (9.0, 15.1) 0.02
Adiponectin (ng/mL) 5.9 (4.3, 8.4) 5.0 (3.4, 7.8) 0.04
Results – Overall CohortMedian (IQR) or % No Diabetes (n=648) Incident Diabetes (n=84) P value
Family History of Diabetes 42% 63% <0.001
Waist/Hip ratio 0.91 (0.85, 0.97) 0.95 (0.90, 1.00) <0.001
Systolic Blood Pressure (mmHg)
123 (115, 134) 131 (122, 144) <0.001
Glucose (mg/dL) 93 (87, 100) 101 (92, 114) <0.001
Fructosamine (µmol/L) 199 (188, 210) 211 (196, 224) <0.001
Triglycerides (mg/dL) 99 (70, 146) 124 (90, 187) 0.001
Lower Body Fat (kg) 12.6 (9.6, 16.3) 11.2 (9.0, 15.1) 0.02
Adiponectin (ng/mL) 5.9 (4.3, 8.4) 5.0 (3.4, 7.8) 0.04
Body Mass Index (kg/m2) 34.9 (31.9, 38.9) 35.4 (33.0, 39.3) 0.35
Total Body Fat (kg) 35.5 (29.3, 43.4) 35.3 (28.8, 42.7) 0.51
HDL Cholesterol (mg/dL) 46 (39, 54) 45 (38, 54) 0.48
C-reactive protein (mg/L) 4.4 (2.2, 9.4) 3.6 (1.9, 9.3) 0.40
Results – Overall CohortDiabetes Incidence by Sex-Specific Tertiles
of Abdominal Fat Distribution
Results – Overall CohortDiabetes Incidence by Sex-Specific Tertiles
of Abdominal Fat Distribution
Results – Overall Cohort – Incident Diabetes
Multivariable analysis:
Variable Odds Ratio (95% CI) Χ2 value
Fructosamine (per 1 SD)* 2.0 (1.4-2.7) 17.7
Visceral fat mass (per 1 SD)* 2.4 (1.6-3.7) 17.0
Fasting glucose (per 1 SD)* 1.9 (1.4-2.6) 16.1
Weight gain (per 5 kg) 1.3 (1.1-1.2) 9.8
Systolic blood pressure (per 10 mm Hg)
1.3 (1.1-1.5) 7.6
Family history of diabetes 2.3 (1.3-4.3) 7.1
*Log-transformed
Results – Overall Cohort – Incident Diabetes
Multivariable analysis:
Variable Odds Ratio (95% CI) Χ2 value
Fructosamine (per 1 SD)* 2.0 (1.4-2.7) 17.7
Visceral fat mass (per 1 SD)* 2.4 (1.6-3.7) 17.0
Fasting glucose (per 1 SD)* 1.9 (1.4-2.6) 16.1
Weight gain (per 5 kg) 1.3 (1.1-1.2) 9.8
Systolic blood pressure (per 10 mm Hg)
1.3 (1.1-1.5) 7.6
Family history of diabetes 2.3 (1.3-4.3) 7.1
*Log-transformed
Results – Overall Cohort – Incident Diabetes
Multivariable analysis:
Variable Odds Ratio (95% CI) Χ2 value
Fructosamine (per 1 SD)* 2.0 (1.4-2.7) 17.7
Visceral fat mass (per 1 SD)* 2.4 (1.6-3.7) 17.0
Fasting glucose (per 1 SD)* 1.9 (1.4-2.6) 16.1
Weight gain (per 5 kg) 1.3 (1.1-1.2) 9.8
Systolic blood pressure (per 10 mm Hg)
1.3 (1.1-1.5) 7.6
Family history of diabetes 2.3 (1.3-4.3) 7.1
*Log-transformed
Results – Overall Cohort – Incident Diabetes
Multivariable analysis:
Variable Odds Ratio (95% CI) Χ2 value
Fructosamine (per 1 SD)* 2.0 (1.4-2.7) 17.7
Visceral fat mass (per 1 SD)* 2.4 (1.6-3.7) 17.0
Fasting glucose (per 1 SD)* 1.9 (1.4-2.6) 16.1
Weight gain (per 5 kg) 1.3 (1.1-1.2) 9.8
Systolic blood pressure (per 10 mm Hg)
1.3 (1.1-1.5) 7.6
Family history of diabetes 2.3 (1.3-4.3) 7.1
*Log-transformed
Results – Overall Cohort – Incident Diabetes
Multivariable analysis:
Variable Odds Ratio (95% CI) Χ2 value
Fructosamine (per 1 SD)* 2.0 (1.4-2.7) 17.7
Visceral fat mass (per 1 SD)* 2.4 (1.6-3.7) 17.0
Fasting glucose (per 1 SD)* 1.9 (1.4-2.6) 16.1
Weight gain (per 5 kg) 1.3 (1.1-1.2) 9.8
Systolic blood pressure (per 10 mm Hg)
1.3 (1.1-1.5) 7.6
Family history of diabetes 2.3 (1.3-4.3) 7.1
*Log-transformed
Results – Overall Cohort – Incident Diabetes
Multivariable analysis:
Variable Odds Ratio (95% CI) Χ2 value
Fructosamine (per 1 SD)* 2.0 (1.4-2.7) 17.7
Visceral fat mass (per 1 SD)* 2.4 (1.6-3.7) 17.0
Fasting glucose (per 1 SD)* 1.9 (1.4-2.6) 16.1
Weight gain (per 5 kg) 1.3 (1.1-1.2) 9.8
Systolic blood pressure (per 10 mm Hg)
1.3 (1.1-1.5) 7.6
Family history of diabetes 2.3 (1.3-4.3) 7.1
*Log-transformed
Multivariable analysis:
*Log-transformed
Variable Odds Ratio (95% CI) Χ2 value
Weight gain (per 5 kg) 1.5 (1.3-1.6) 40.9
Fasting blood glucose (per 1 SD)* 1.7 (1.3-2.1) 16.0
Age (per 10 years) 1.5 (1.2-1.9) 10.9
Visceral fat mass (per 1 SD)* 1.5 (1.2-1.9) 10.8
Fructosamine (per 1 SD)* 1.4 (1.1-1.8) 10.2
Insulin (per 1 SD)* 1.3 (1.1-1.7) 6.1
Nonwhite race 1.8 (1.1-2.9) 5.2
Family history of diabetes 1.6 (1.1-2.4) 4.8
Results – Subgroup with FBG<100 – Incident Prediabetes or Diabetes
ResultsPrevalence of Subclinical CVD at Baseline
Stratified by Diabetes Status
Conclusions
• Dysfunctional adiposity phenotype associated with incident prediabetes and diabetes in obese population.– Excess visceral fat mass
– Insulin resistance
• No association between general adiposity and incident prediabetes or diabetes.
• Obesity is a heterogeneous disorder with distinct adiposity sub-phenotypes.
Clinical Implications
Risk Stratification
?
Intensive Lifestyle
Modification
Pharmacologic Therapy
Bariatric Surgery
jamanetwork.comCopyright restrictions apply.
Available at www.jama.com
IJ Neeland and coauthors
Dysfunctional Adiposity and the Risk of Prediabetes and Type 2 Diabetes in Obese Adults
Visceral Fat stratified by Subgroups
Study Population and Follow-Up
VariableParticipated in DHS-2
(n=732)
Did not participate in DHS-2(n=345)
P-value
Weight (kg) 98.4 (87.5, 109.8) 98.0 (87.1, 109.3) 0.69
Body Mass Index (kg/m2) 35.0 (32.0, 38.9) 34.4 (31.8, 38.6) 0.21
Waist Circumference (cm) 109.0 (101.0, 117.5) 108.7 (101.5, 116.5) 0.68
Waist/Hip ratio 0.91 (0.85, 0.98) 0.92 (0.87, 0.98) 0.08
Impaired Fasting Glucose, No. (%)
211 (28.8) 96 (27.8) 0.50
Family History of Diabetes, No. (%)
290 (44.1) 129 (42.6) 0.66
Hypertension, No. (%) 258 (35.8) 132 (38.7) 0.36
Metabolic Syndrome, No. (%)
348 (47.5) 164 (47.5) 1.00
Total Fat Mass (kg) 35.5 (29.2, 43.4) 34.1 (28.0, 42.7) 0.08
Abdominal Visceral Fat (kg)
2.5 (1.9, 3.1) 2.5 (2.0, 3.1) 0.84
Non-Participants
Single slicemeasurementat L2-L3 levelprovidesexcellentaccuracyfor abdominalfat mass measured at all inter-vertebral levels(R2=85-96%)
Abdominal MRI Measurements Correlation coefficient (r)
Study and reference
Method Location of single
slice Tot-SAT : SS-SAT Tot-VAT : SS-VAT
Borkan et al 1 (n=8 M)
CT 6, 4, 2 cm above, at, and 2, 4, 6 cm below umbilicus
0.99 0.95-0.99
Tokunaga et al 2 (n= 8 M)
CT Umbilicus 0.99 ---
Kvist et al 3
(n=17 M, 10 F) CT L3-L4 --- 0.94-0.98
L4-L5 0.98-0.99 Ross et al 4 (n=27 M)
MRI 15 cm above L4-
L5 --- 0.96
10 cm above L4-L5 (corresponding approximately to
L2-L3 level)
--- 0.96
5 cm above L4-L5 --- 0.97 L4-L5 0.97 0.95 5 cm below L4-L5 --- 0.89
Armellini et al 5 (n=18 M, 72 F)
CT T12-L1 --- 0.95
L2-L3 --- 0.98 L3-L4 --- 0.95 L4-L5 --- 0.92
Abate et al 6 (n=49 M)
MRI T12-L1 0.93 0.83
L1-L2 0.94 0.89 L2-L3 0.92 0.92 L3-L4 0.89 0.93 L4-L5 0.87 0.87 L5-S1 0.96 0.86
• Criteria for entry = 0.1
• Criteria for backward selection = 0.05
• Assessment for Overfitting: Shrinkage coefficient calculated as: [Likelihood model chi-square-p]/Likelihood model chi-square, where p=# of covariates in the model– Incidence diabetes = 0.94– Incident prediabetes or diabetes = 0.95
• Evaluation for Collinearity: Variance inflation factors (VIFs) calculated using the dependent variable from logistic regression analysis as a dependent variable in a linear regression. No evidence of collinearity found (VIFs all <1.7).
Multivariable Models
For model validation, we used 500 bootstrap samples to see how the beta coefficients, and hence, odds ratios, changed across variations within our dataset. We used bootstrapping as opposed to cross-validation, as the entire dataset is used for model development in the samples and we wanted to preserve the inclusion of as many events as possible. Bootstrapping also provides fairly unbiased estimates. The odds ratios changed minimally.
T2DM Model Full sample 500 Bootstrap Replications
Fasting Glucose 1.88 (1.38-2.56) 1.90 (1.31-2.77) Family History 2.32 (1.25-4.29) 2.42 (1.24-4.73) Systolic BP 1.26 (1.07-1.48) 1.27 (1.06-1.52) Visceral fat 2.42 (1.59-3.68) 2.49 (1.62-3.84) Fructosamine 1.95 (1.43-2.67) 1.97 (1.38-2.82) Weight gain 1.06 (1.02-1.10) 1.06 (1.02-1.11) The c-statistic based on the original dataset was 0.845, while the bootstrap c-statistics derived from fitting the bootstrapped equation to the original dataset on average was 0.835. Thus, the diminution in c-statistics is small. Furthermore, we were only interested in determining associations with incident diabetes and not constructing a risk prediction model.
Model Validation
• Diabetes: 12/84 = 14%
• Prediabetes: 67/161 = 42%
• Findings insensitive to excluding these participants from the multivariable models.
Diagnoses Exclusively by Hgb A1C
Visceral fat and Insulin Resistance are Additive
Anthropometric Measures of Abdominal Obesity are Insufficient
Variable Odds Ratio (95% CI) X2
Waist Circumference (per 1 cm)
0.99 (0.97-1.0) 0.01
Log WHR (per 1-SD) 1.4 (0.96-2.0) 3.0
Added to the Incident Diabetes Model without Visceral Fat
Weight Gain over the Study Interval
Potential Mechanisms
• ↓ Subcutaneous fat storage = ↑ Visceral and ectopic fat
• Resistance to diabetes may be due to shunting excess fat away from ectopic sites and preferentially depositing it in the lower body subcutaneous compartment.
• Visceral fat and insulin resistance may contribute to subclinical CVD prior to the clinical manifestations of metabolic disease.
Subcutaneous Fat Expandability and Metabolic Health
Tran et al. Cell Metab. 2008;7:410-420
Strengths and Limitations• Strengths :
– diverse sample of adults applicable to the general obese population
– extensive and detailed phenotyping using advanced imaging and laboratory techniques
– longitudinal follow-up in a prospective cohort
• Limitations: – absence of glucose tolerance testing in the DHS and of Hgb A1C
measurements in DHS-1 – modest number of diabetes events– time of pre-diabetes or diabetes onset not available. – findings not necessarily generalizable to individuals older than
age 65 or of Asian descent/ethnicity.
Prior Studies
Colditz et al. Ann Intern Med. 1995;122:481-86Stern et al. Ann Intern Med. 2002;136:575-81Schmidt et al. Diabetes Care. 2005;28:2013-18Wilson et al. Arch Intern Med. 2007;167:1068-74
Author, Year Study Population Mean Weight or BMI Summary of Findings
Colditz et al, 1995 Nurses Health Study 57 kg BMI, Weight gain
Stern et al, 2002 San Antonio Heart Study 24-28 kg/m2 BMI, Blood pressure, TGs, HDL-C
Schmidt et al, 2005Atherosclerosis Risk in
Communities Study26 kg/m2 Waist circumference, TGs, HDL-C
Wilson et al, 2007Framingham Offspring Cohort
Study27 kg/m2 BMI, Blood pressure, TGs, HDL-C