ORIGINAL CONTRIBUTION

Association of Weight Status With Mortality in Adults With Incident Diabetes Mercedes R. Carnethon, PhD Peter John D. De Chavez, MS Mary L. Biggs, PhD Cora E. Lewis, MD James S. Pankow, PhD Alain G. Bertoni, MD, MS Sherita H. Golden, MD, MS Kiang Liu, PhD Kenneth J. Mukamal, MD, MPH Brenda Campbell-Jenkins, PhD Alan R. Dyer, PhD

T

YPE 2 DIABETES IN NORMAL weight adults is an understudied representation of the metabolically obese normalweight phenotype1 that has become increasingly common over time.2 It is not known whether the “obesity paradox” that has been observed in chronic diseases such as heart failure, chronic kidney disease, and hypertension extends to adults who are normal weight at the time of incident diabetes.3-5 In 2 contemporary studies, the Translating Research Into Action for Diabetes (TRIAD) study6 and the PROactive trial,7 participants with diabetes who were normal weight at the baseline examination or who lost weight during the trial (PROactive) experienced higher mortality than participants who were overweight or obese. Limitations of these prevalent disease studies are that participants had diabetes of unknown duration and participants from the PROactive trial had preexisting cardiovascular disease at baseline.

For editorial comment see p 619.

Context Type 2 diabetes in normal-weight adults (body mass index [BMI] ⬍25) is a representation of the metabolically obese normal-weight phenotype with unknown mortality consequences. Objective To test the association of weight status with mortality in adults with newonset diabetes in order to minimize the influence of diabetes duration and voluntary weight loss on mortality. Design, Setting, and Participants Pooled analysis of 5 longitudinal cohort studies: Atherosclerosis Risk in Communities study, 1990-2006; Cardiovascular Health Study, 1992-2008; Coronary Artery Risk Development in Young Adults, 1987-2011; Framingham Offspring Study, 1979-2007; and Multi-Ethnic Study of Atherosclerosis, 20022011. A total of 2625 participants with incident diabetes contributed 27 125 personyears of follow-up. Included were men and women (age ⬎40 years) who developed incident diabetes based on fasting glucose 126 mg/dL or greater or newly initiated diabetes medication and who had concurrent measurements of BMI. Participants were classified as normal weight if their BMI was 18.5 to 24.99 or overweight/obese if BMI was 25 or greater. Main Outcome Measures Total, cardiovascular, and noncardiovascular mortality. Results The proportion of adults who were normal weight at the time of incident diabetes ranged from 9% to 21% (overall 12%). During follow-up, 449 participants died: 178 from cardiovascular causes and 253 from noncardiovascular causes (18 were not classified). The rates of total, cardiovascular, and noncardiovascular mortality were higher in normal-weight participants (284.8, 99.8, and 198.1 per 10 000 personyears, respectively) than in overweight/obese participants (152.1, 67.8, and 87.9 per 10 000 person-years, respectively). After adjustment for demographic characteristics and blood pressure, lipid levels, waist circumference, and smoking status, hazard ratios comparing normal-weight participants with overweight/obese participants for total, cardiovascular, and noncardiovascular mortality were 2.08 (95% CI, 1.52-2.85), 1.52 (95% CI, 0.89-2.58), and 2.32 (95% CI, 1.55-3.48), respectively. Conclusion Adults who were normal weight at the time of incident diabetes had higher mortality than adults who are overweight or obese. www.jama.com

JAMA. 2012;308(6):581-590

To minimize the influence of diabetes duration and unintentional or intentional weight loss secondary to dia-

betes development and diagnosis,8 we compared mortality between participants who were normal weight and

Author Affiliations: Department of Preventive Medicine, Feinberg School of Medicine, Northwestern University, Chicago, Illinois (Drs Carnethon, Liu, and Dyer and Mr De Chavez); Department of Biostatistics, University of Washington, Seattle (Dr Biggs); Department of Preventive Medicine, University of Alabama School of Medicine, Birmingham (Dr Lewis); Division of Epidemiology and Community Health, University of Minnesota, Minneapolis (Dr Pankow); Division of Epidemiology, Department of Health Studies, Wake Forest University, Winston-Salem, North Carolina (Dr Bertoni); Division of Endocrinology, Department of Medi

cine and Welch Center for Prevention, Johns Hopkins University, Baltimore, Maryland (Dr Golden); Division of Internal Medicine, Department of Medicine, Harvard University and Beth Israel Deaconess Medical Center, Brookline, Massachusetts (Dr Mukamal); and Jackson Heart Study Coordinating Center, Jackson State University, Jackson, Mississippi (Dr Campbell-Jenkins). Corresponding Author: Mercedes R. Carnethon, PhD, Department of Preventive Medicine, Feinberg School of Medicine, Northwestern University, 680 N Lake Shore Dr, Ste 1400, Chicago, IL 60611 (carnethon @northwestern.edu).

©2012 American Medical Association. All rights reserved.

JAMA, August 8, 2012—Vol 308, No. 6

Corrected on October 10, 2012

Downloaded From: http://jamanetwork.com/pdfaccess.ashx?url=/data/journals/jama/24745/ on 01/27/2017

581

WEIGHT STATUS AND MORTALITY IN ADULTS WITH INCIDENT DIABETES

overweight/obese at the time of incident adult-onset diabetes. We hypothesized that participants who were normal weight at the time of incident diabetes would experience higher mortality than participants who were overweight or obese. METHODS Our study included 2625 participants from the Atherosclerosis Risk in Communities (ARIC) study, Cardiovascular Health Study (CHS), Coronary Artery Risk Development in Young Adults (CARDIA) study, Framingham Offspring Study (FOS), and Multi-Ethnic Study of Atherosclerosis (MESA) who developed incident diabetes. We selected these studies because they had repeated measures of body weight, fasting glucose level, and medication use; a comprehensive set of commonly measured covariates; and longitudinal follow-up for events and mortality.9-13 The eTable, available at http://www.jama .com, summarizes each study’s size, follow-up duration, number of examinations, and examination dates. Institutional review boards at each of the institutions reviewed the protocols and procedures and approved the research. All participants provided written informed consent at each examination. Data were deidentified for our analysis, and the Northwestern University institutional review board approved the research. Diabetes and Weight Status

Diabetes was determined as fasting (ⱖ8 hours) glucose of 126 mg/dL or greater9,11-15 or reported use of oral hypoglycemic medications or insulin. (To convert glucose to mmol/L, multiply by 0.0555.) Incident diabetes was determined among participants who were free from diabetes at baseline and who met one of the these 2 criteria at a subsequent follow-up examination. Body mass index (BMI) was calculated as weight in kilograms divided by height in meters squared. Normal weight, overweight, and obese were defined as a BMI of 18.5 to 24.9, 25 to 29.9, and 30 or greater, respectively.16 582

JAMA, August 8, 2012—Vol 308, No. 6 Corrected on October 10, 2012

Participants’ weight status was assigned at the examination when diabetes was identified (ie, baseline of this analysis sample). Follow-up Time and Mortality

Participants were followed up from the examination at which diabetes was identified until they died, reached the end of their cohort surveillance, or were lost to follow-up. Mortality was determined annually using cohort-specific surveillance protocols, and investigators adjudicated cause of death after review of all available medical records. Cardiovascular death (ie, myocardial infarction, stroke) was adjudicated using a combination of review of death certificates for codes indicating cardiovascular disease as an underlying cause of death and proxy interviews.10-13,17 Causes of noncardiovascular death were not uniformly adjudicated across studies. Covariates

Demographic characteristics, health behaviors, and clinical factors available in each of the cohort studies were measured using standard protocols.9-13 We selected covariatesthatwerecommonlymeasured across studies. Race/ethnicity was determined according to self-report and was assessedbyeachcomponentcohortstudy because of the known relevance of race/ ethnicity to cardiovascular disease. Covariates were determined at the time of incident diabetes (ie, baseline); however, if the measures were not available from that examination, the most recent value fromapriorcohortexaminationwasused instead. Statistical Analysis

We compared means and standard deviations or proportions of study characteristics between normal-weight and overweight/obese participants who had incident diabetes within each cohort using t tests and ␹2 tests, respectively. Kaplan-Meier survival curves with logrank ␹2 are presented to compare mortality by weight status. Because the number of participants remaining after 15 years was small, we truncated the presentation to 15 years of follow-up.

Following confirmation of proportional hazards using log-log survival plots, we modeled the mortality hazards comparing normal-weight with overweight/obese participants with diabetes (referent). We used 2 strategies to generate pooled estimates. First, we performed cohort-specific analyses to generate effect estimates that were pooled together using fixed- and randomeffects meta-analysis. Because effect estimates were relatively homogenous across cohorts, there were no differences between fixed and random effects and so we present fixed effects. Second, we performed a pooled cohort analysis using Cox modeling with a stratification term for cohort. Because waist circumference and lipids were measured using different protocols and assays, we transformed them to z scores in the pooled analysis. Model 1 was adjusted for age, race (nonwhite vs white), sex, and education (less than high school vs high school graduate or more). Model 2 was adjusted for model 1 and waist circumference, total cholesterol level, highdensity lipoprotein cholesterol level, systolic blood pressure, and smoking status (current or former vs never). Variance inflation factor and tolerance statistics indicated that the covariates in the model were not collinear.18 We tested whether sex, race, age at diabetes incidence (⬍65 vs ⱖ65 years), or smoking status modified the association of weight status with mortality using multivariable Cox models with a multiplicative interaction term between each characteristic of interest and normal-weight status. We determined statistical significance for the interaction based on the maximum likelihood ␹2 from a nested model with and without the interaction term. Analyses were repeated for each cause of mortality. We carried out a series of sensitivity analyses for our primary outcome of total mortality to explore alternative explanations for our findings. We analyzed the association between BMI per standard deviation higher and total

©2012 American Medical Association. All rights reserved.

Downloaded From: http://jamanetwork.com/pdfaccess.ashx?url=/data/journals/jama/24745/ on 01/27/2017

WEIGHT STATUS AND MORTALITY IN ADULTS WITH INCIDENT DIABETES

mortality and the association between waist circumference per standard deviation higher and total mortality. In an attempt to reduce variability in the duration of new-onset diabetes, we restricted our analysis to participants who had elevated fasting glucose but who were not taking medications to control diabetes. To test whether defining diabetes using a single glucose measurement contributed to misclassification, we restricted the definition of diabetes to participants taking medications only. Because Asian adults are more likely to develop diabetes at a lower BMI, we performed an analysis excluding Asian participants.

To reduce the possibility that unmeasured illness at the time of diabetes identification resulted in weight loss prior to imminent death, we excluded participants who were followed up for less than 2 years after diabetes identification. We excluded 162 participants whose BMI decreased by more than 2 units from the baseline examination, which may have reflected other illnesses that might predispose to death. Given prior reports that overweight adults have the lowest mortality risk (particularly among older adults), we calculated mortality hazard ratios (HRs) comparing normal-weight and obese participants with overweight participants.

All analyses were carried out using SAS version 10 (SAS Institute). Statistical significance was determined at P⬍ .05 (2-sided). RESULTS Demographic, clinical, and behavioral characteristics at the time of incident diabetes are stratified by weight status in TABLE 1. Across cohorts, 293 participants (11.2%) had normal-weight diabetes; normal-weight diabetes was most common in CHS (21%) and lowest in ARIC (9%). Half (50%) of the participants were women, 36% were nonwhite, and the mean (SD) age of participants ranged from 41 (6) years

Table 1. Distribution of Covariates Stratified by Weight Status a ARIC Normal Weight

O/O

Participants, No. (%)

108 (8.7)

1132 (91.3)

Age, mean (SD), y

59.8 (6.1)

59.1 (5.9)

Nonwhite race, No. (%)

34 (31.5)

Female sex, No. (%)

CARDIA P Value b

Normal Weight

O/O

28 (10.2)

246 (89.8)

.23

39.4 (6.4)

41.0 (5.8)

384 (33.9)

.61

14 (50.0)

51 (47.2)

584 (51.6)

.38

Less education, No. (%) c

21 (19.4)

316 (28.0)

Systolic BP, mean (SD), mm Hg

126.1 (19.3)

Diastolic BP, mean (SD), mm Hg

CHS P Value

Normal Weight

O/O

37 (20.6)

143 (79.4)

.16

78.6 (5.6)

75.1 (4.7)

164 (66.7)

.08

5 (13.5)

14 (50.0)

139 (56.5)

.51

.06

3 (10.7)

26 (10.6)

⬎.99

128.9 (18.5)

.14

111.6 (13.3)

121.4 (16.5)

72.0 (11.1)

73.9 (10.4)

.07

71.8 (10.8)

Hypertension, No. (%) d

51 (47.2)

671 (59.3)

.02

Ever smoked, No. (%)

69 (63.9)

704 (62.5)

BMI, mean (SD)

23.2 (1.6)

Waist circumference, mean (SD), cm

FOS P Value

Normal Weight

MESA

P O/O Value b

48 (10.4)

413 (89.6)

⬍.001

59.5 (10.7)

58.5 (9.0)

25 (17.5)

.56

0

0

20 (54.1)

79 (55.2)

.90

22 (45.8)

174 (42.1)

16 (43.2)

43 (30.1)

.13

5 (12.8)

.003

133.7 (20.3)

134.4 (20.0)

.85

79.2 (12.0)

.002

67.6 (9.9)

70.1 (11.6)

6 (21.4)

112 (45.5)

.01

19 (51.4)

.77

15 (55.6)

100 (41.2)

.15

32.8 (5.7)

⬍.001

22.4 (1.8)

37.0 (7.7)

90.4 (9.4)

111.1 (13.3)

⬍.001

77.6 (7.0)

Total cholesterol, mean (SD), mg/dL

205.5 (49.8)

208.8 (42.3)

.51

HDL cholesterol, mean (SD), mg/dL

45.5 (13.2)

42.7 (13.0)

.03

Triglycerides, median (IQR), mg/dL e

126.9 (69.9)

LDL cholesterol, mean (SD), mg/dL

131.0 (45.2)

Normal Weight

P O/O Value b

72 (15.3)

398 (84.7)

68.9 (10.0)

63.6 (9.5)

⬍.001

54 (75.0)

274 (68.8)

.30

.62

44 (61.1)

202 (50.8)

.11

36 (10.5)

.59

22 (30.6)

76 (19.1)

.03

129.8 (21.9)

136.3 (18.8)

.03

123.9 (21.3)

126.8 (19.8)

.26

.24

76.0 (10.2)

79.4 (12.1)

.06

68.1 (10.4)

71.5 (10.6)

.01

92 (64.3)

.15

25 (52.1)

285 (69.0)

.02

43 (59.7)

272 (68.3)

.15

24 (68.6)

82 (58.6)

.28

38 (79.2)

290 (70.2)

.20

31 (43.1)

234 (59.5)

.009

⬍.001

22.8 (1.5)

31.1 (4.7)

⬍.001

23.1 (1.7)

32.6 (5.5)

⬍.001

23.2 (1.5)

33.0 (5.8)

⬍.001

109.0 (16.7)

⬍.001

90.3 (9.1)

108.5 (16.9)

⬍.001

88.3 (10.5)

109.7 (12.8)

⬍.001

87.5 (6.7)

109.7 (14.2)

⬍.001

180.4 (50.6)

188.7 (38.9)

.42

200.0 (51.6)

204.2 (41.8)

.61

194.2 (46.0)

209.8 (39.5)

.01

182.9 (43.3)

182.6 (35.7)

.96

58.3 (26.8)

42.7 (11.4)

.006

51.7 (11.2)

47.9 (11.6)

.08

45.6 (16.8)

40.6 (11.9)

.05

52.7 (15.7)

46.5 (12.4)

.002

152.4 ⬍.001 (104.9)

90.0 (83.0)

121.5 (109.0)

.005

118.0 (122.0)

159.0 (122.0)

.02

136.5 (112.0)

181.0 (135.5)

.001

108.0 (83.0)

131.0 (103.0)

.002

130.8 (37.0)

97.2 (40.3)

114.7 (34.5)

.02

122.1 (38.0)

122.9 (32.1)

.89

117.1 (37.8)

142.1 (36.3)

.01

106.2 (33.8)

104.8 (31.2)

.73

.96

.47

Abbreviations: ARIC, Atherosclerosis Risk in Communities study; BP, blood pressure; BMI, body mass index; CARDIA, Coronary Artery Risk Development in Young Adults study; CHS, Cardiovascular Health Study; HDL, high-density lipoprotein; IQR, interquartile range; FOS, Framingham Offspring Study; LDL, low-density lipoprotein; MESA, Multi-Ethnic Study of Atherosclerosis; O/O, overweight/obese. SI conversion factors: To convert total, HDL, and LDL cholesterol to mmol/L, multiply by 0.0259; to convert triglycerides to mmol/L, multiply by 0.0113. a BMI is calculated as weight in kilograms divided by height in meters squared. Normal weight was defined as a BMI of 18.5 to 24.9; overweight/obese defined as a BMI of ⱖ25. b P value based on t test for continuous variables and ␹2 test for categorical variables. c Did not graduate from high school. d Systolic blood pressure ⱖ140 mm Hg or diastolic blood pressure ⱖ90 mm Hg or reported use of antihypertensive medications. e Statistical significance determined using a Wilcoxon rank sum test.

©2012 American Medical Association. All rights reserved.

JAMA, August 8, 2012—Vol 308, No. 6

Corrected on October 10, 2012

Downloaded From: http://jamanetwork.com/pdfaccess.ashx?url=/data/journals/jama/24745/ on 01/27/2017

583

WEIGHT STATUS AND MORTALITY IN ADULTS WITH INCIDENT DIABETES

Figure 1. Kaplan-Meier Survival Estimates Comparing Mortality in Participants Stratified by Weight Status at the Time of Incident Diabetes A Total mortality 1.0 Overweight/obese

Survival Proportion

0.8 Normal weight 0.6

0.4

0.2

Log-rank χ2 = 27.6 P