American Journal of Epidemiology
Copyright O 1997 by The Johns Hopkins University School of Hygiene and Public Health
All rights reserved
Vol. 146, No. 3
Printed In U.S.A
ORIGINAL CONTRIBUTIONS
Weight Change and Diabetes Incidence: Findings from a National Cohort of
US Adults
Ear) S. Ford, 1 David F. Williamson,2 and Simin Liu 3
To examine how long-term patterns of weight change affect the risk for diabetes, especially non-insulindependent diabetes mellitus, the authors examined the relation of weight change over a period of about 10
years (from the baseline examination in 1971-1975 until the first follow-up examination in 1982-1984) to the
9-year incidence of diabetes mellitus (1984-1992) in a national cohort of 8,545 US adults from the National
Health and Nutrition Examination Survey Epidemiologic Followup Study. Diabetes incidence was identified
from death certificates, hospitalization and nursing home records, and self-report. In this cohort, 487 participants developed diabetes. The hazard ratios were 2.11 (95% confidence interval (Cl) 1.40-3.18) for participants who gained 5-<8 kg , 1.19 (95% Cl 0.75-1.89) for participants who gained 8-<11 kg, 2.57 (95% Cl
1.84-3.85) for participants who gained 11-<20 kg, and 3.85 (95% Cl 2.04-7.22) for participants who gained
20 kg or more compared with participants whose weights remained relatively stable. The authors found no
evidence that the results differed by age, sex, or race. They estimated that the population attributable risk was
27% for weight increases of 5 kg or more. Results from this study and other recent studies suggest that the
increase in body mass index in the United States that occurred during the 1980s may portend an increase in
the incidence of non-insulin-dependent diabetes mellitus with important public health consequences in future
years. Am J Epidemiol 1997;146:214-22.
body weight; cohort studies; diabetes mellitus; obesity; risk factors; weight gain
The role of obesity in the pathogenesis of diabetes
mellitus, particularly non-insulin-dependent diabetes
mellitus (NIDDM), has long been recognized (1, 2).
Although the exact mechanisms through which obesity
increases the risk of diabetes are not fully understood,
obesity results in insulin resistance, a state that is also
characteristic of impaired glucose tolerance and
NIDDM (3). Insulin resistance is characterized by a
reduction in the number and function of insulin receptors and by a disruption of the postreceptor cascade of
events. In addition, obesity is often associated with
increased insulin production by pancreatic cells. Perhaps as a function of the duration and magnitude of
obesity, further deterioration in glucose homeostatic
mechanisms occurs when /3-cells become glucose incompetent and clinical diabetes develops (4, 5). Also,
elevated levels of free fatty acids in obesity may
increase hepatic glucose production (3) and play a role
in the insulin resistance of target tissues and decompensation of /3-cells (5), thus promoting NIDDM.
Various measures of being overweight, such as relative weight, body mass index, and body fat distribution, have repeatedly been shown to be risk factors for
NIDDM (6-16). However, the relation of weight
change and risk for diabetes has been less well studied.
Although an earlier study showed that weight loss was
associated with increased risk for diabetes (17), more
recently, prospective studies in select populations have
showed that weight gain was associated with increased
diabetes incidence in retired adults living in Rancho
Received for publication June 17, 1996, and accepted for publication March 5, 1997.
Abbreviations: Cl, confidence interval; ICD-9-CM, International
Classification of Diseases, Ninth Revision, Clinical Modification;
NHANES, National Health and Nutrition Examination Survey;
NHEFS, National Health and Nutrition Examination Survey Epidemiologic Followup Study; NIDDM, non-insulin-dependent diabetes
mellitus.
1
Division of Nutrition and Physical Activity, National Center for
Chronic Disease Prevention and Health Promotion, Centers for Disease Control and Prevention, Atlanta, GA
2
Division of Diabetes Translation, National Center for Chronic
Disease Prevention and Health Promotion, Centers for Disease
Control and Prevention, Atlanta, GA.
3
Department of Epidemiology, Harvard School of Public Health,
Boston, MA.
Reprint requests to Dr. Earl Ford, Division of Nutrition, National
Center for Chronic Disease Prevention and Health Promotion, Centers for Disease Control and Prevention, 4770 Birford Hwy., N.E.,
Mailstop K-26, Atlanta, GA 30341.
This paper was prepared under the auspices of the US Government and is therefore not subject to copyright.
214
Weight Change and Diabetes
Bernardo, California (18), male college alumni (13),
male health professionals (15), Pima Indians (19), and
female nurses (16). Furthermore, the last study also
showed that weight loss was associated with a reduced
risk for diabetes.
To our knowledge, however, the relation between
weight change and diabetes incidence has not been
reported from a nationally representative sample of US
adults. We used such a nationally representative sample to estimate the association between weight change
and diabetes risk as well as the population attributable
risk in the United States population. In addition, we
examined evidence that the association between
weight change and diabetes varies by age, sex, and
race.
MATERIALS AND METHODS
A cohort of 14,407 persons who were 25 years of
age or older at the First National Health and Nutrition
Examination Survey (NHANES I) baseline examination between 1971 and 1975 have been followed
through 1992. The methods of the original NHANES
I survey and the subsequent NHANES I Epidemiologic Followup Study (NHEFS) have been presented
in detail elsewhere (20-25).
Within this cohort, we examined the relation between weight change that occurred between the baseline examination conducted from 1971 through 1975
and the first follow-up examination in 1982-1984 and
the incidence of diabetes mellitus that occurred during
the ensuing years through 1992. Using three data
sources (questionnaires, institutional records, and
death certificates), we identified study participants
who developed diabetes from 1982-1984 through
1992. The date of onset was considered to be the
earliest occurrence of diabetes determined from these
three sources.
From questionnaires, diabetes was identified from
affirmative self- or surrogate responses to the following question: "Did a doctor ever tell (you/him/her) that
(you/he/she) had diabetes or sugar diabetes?" This
question was asked in 1986 (for participants who were
55 years of age or older), 1987, and 1992. Respondents were then asked to report the year when participants had first been told that they had diabetes. We
assigned the midpoint of the reported year of onset of
diabetes as the date of onset.
Participants were asked about hospitalizations during the study period. If any were reported, permission
was obtained to abstract certain data, such as admission and discharge dates and diagnoses. Medical
records had been recorded by trained medical coders
using International Classification of Diseases, Ninth
Revision, Clinical Modification (ICD-9-CM), codes.
Am J Epidemiol
Vol. 146, No. 3, 1997
215
We used the ICD-9-CM code 250, diabetes mellitus, to
identify deaths from any of the causes of death or
hospitalizations listed among any of the 10 possible
diagnoses that were abstracted from the hospital discharge summary or medical records facesheet.
Deaths were identified through searches of the National Death Index and the Health Care Financing
Administration enrollee files and through other tracing
mechanisms. A participant was considered deceased
only if a death certificate had been received or a proxy
interview had been completed to verify the death.
Death certificates have been obtained for 97 percent of
deceased participants through 1992.
Participants who were identified as having diabetes
from the baseline or first follow-up interview in 1982—
1984, who from a later interview were identified as
having diabetes with a date of diagnosis prior to the
first follow-up interview, or who had a hospitalization
listing the ICD-9-CM code 250 were excluded from
the analyses.
Weight was measured at baseline (1971-1975) and
during the first follow-up in 1982-1984. At baseline,
participants were weighed on a self-balancing scale
that recorded the weight to within 0.25 lb (0.114 kg)
under controlled conditions in mobile examination
trailers. Examination clothing consisted of disposable
paper uniforms and foam-rubber slippers. The weight
of the examination clothing was not subtracted from
the participants' measured weight, but it varied between 0.2 and 0.6 lb (0.09 and 0.27 kg, respectively).
At follow-up, participants were weighed at home. After removing their shoes and any extra articles of
clothing, such as heavy sweaters or jackets, participants were weighed on a portable spring scale. Each
scale was used for 40-50 participants before being
returned to the survey center for routine maintenance
and calibration. If the first weight was considered
faulty, a second measurement was obtained. Because
of limitations of the scales, weights were recorded
only up to 300 lb (136.2 kg). After excluding participants with weights over 300 lb from the analysis (n =
9), we calculated the weight change that occurred from
baseline (1971-1975) until the first follow-up interview and examination (1982-1984).
All covariates included in the analyses were derived
from the baseline examination and interview and included age, race (black, white), education (years),
cigarette smoking (never, former, current), systolic
blood pressure (mmHg), antihypertensive medication
(yes, no), serum cholesterol (mg/dl), baseline body
mass index (weight (kg)/height (m)2), and alcohol
consumption (0, 1-2, S 3 drinks per day). Because of
the small number of survey participants with a race
designation other than white or black (n = 172), we
216
Ford eta).
excluded them from the analysis. For smoking, we
used a variable that was constructed in part from
responses obtained during the baseline interview and
in part from the first follow-up interview. The validity
of this approach has been previously demonstrated
(26, 27). One blood pressure measurement had been
obtained, in the sitting position, for some participants
and two had been obtained for a subgroup of participants. We used only the first blood pressure. One
cholesterol measurement had been obtained using a
modification of the Abell-Kendall method (25). For
some analyses, overweight was defined as a body
mass index >27.3 kg/m2 for men and >27.8 kg/m2 for
women.
Using the direct method, we adjusted rates of diabetes incidence by age using proportions of five age
groups derived from the 1980 US population. To examine the independent association of weight change
with diabetes incidence, we used proportional hazards
regressions that included age, race, education, smoking status, systolic blood pressure, antihypertensive
medication, cholesterol, baseline body mass index,
and alcohol consumption. To examine the possibility
of residual confounding, we examined additional models with squared terms added for continuous covariates; results for these models appear below. Follow-up
time started with the 1982-1984 examination and
ended when one of the following conditions occurred:
1) the participant became diabetic; 2) the participant
left the study; or 3) the study ended. We performed all
analyses presented here with the software SUDAAN,
which incorporates the complex sampling design of
the study (28). To test the validity of the proportional
hazards assumptions, we examined interactions between survival time and independent variables using
the SAS procedure Proc PHREG (SAS Institute, Inc.,
Cary, North Carolina). For weight change categories
not meeting the proportionality assumption, we calculated hazard ratios for various follow-up periods by
stratifying the analyses on follow-up time. Population
attributable risk (PAR) for variables with multiple
categories of exposure were calculated using the
formula
ered incident cases of diabetes, were included in the
analysis (table 1). Of these incident cases, 313 were
identified from interviews (276 subject interviews and
37 proxy interviews), 144 from institutional records
(138 from hospital records and six from nursing home
records), and the remaining 30 from death certificates.
To examine the possible impact of excluding participants who were lost to follow-up and who had
incomplete data records, we compared the distributions of categorical variables and means of continuous
variables with those of participants included in the
analyses. Participants who were included in the analysis were more likely to be older, to be white, and to
have graduated from high school and less likely to
have used antihypertensive medication than those in
the other two groups. The distributions of body mass
index and alcohol consumption were similar for the
three groups. In addition, participants who were included in the analysis were more likely to be male, to
have a cholesterol level of <200 mg/dl, or to have a
systolic blood pressure reading of <140 mmHg than
those who had missing data and were less likely to be
current smokers or to have a systolic blood pressure
reading of < 140 mmHg than those who were lost to
follow-up.
Participants who developed diabetes during the follow-up period were older and were more likely to be
black, have lower educational attainment, have higher
blood pressure, and be overweight at baseline than
those who did not develop diabetes (table 2). In addition, the mean weight gain for participants who developed diabetes was greater than that for participants
who did not develop diabetes.
As shown in a previous analysis of these data (14),
the baseline body mass index was positively related to
diabetes incidence (table 3). A small, nonsignificant
increase in risk was present for participants with a
TABLE 1. Algorithm for determining analytical sample size,
National Health and Nutrition Examination Survey
EpIdemloJogic Followup Study, 1971-1992
Sample
Action
StZB
(no.)
PAR = 1 -
-i
where pt is the proportion of the population in the ith
exposure category and where /?, is the relative risk
associated with exposure at level i (29).
RESULTS
After we excluded participants for the reasons stated
above, 8,545 participants, of whom 487 were consid-
Raw numbers
Exclude participants lost to follow-up
Exclude participants with missing
weight change data
Exclude participants with prevalent
disease at baseline
Exclude missing data for diabetes
incidence
Exclude pregnant women
Exclude participants with missing
data for covariates
Am J Epidemiol
No. erf
hckJert
cases of
diabetes
14,407
13,861
9,995
8,927
559
8,871
8,790
508
506
8,545
487
Vol. 146, No. 3, 1997
Weight Change and Diabetes
body mass index of 22-22.9 kg/m2 compared with
participants whose body mass index was <22 kg/m2.
With increasing body mass index, the risk for developing diabetes increased rapidly. For each unit of
increase in the body mass index (about 2.7-3.6 kg for
the average participant), the risk for developing diabetes increased 12.1 percent (95 percent confidence
interval (CD 9.9-14.4).
Using the hazard ratios adjusted for multiple confounders from table 3 and body mass index category
proportions derived from the Second National Health
and Nutrition Examination Survey (NHANES H) for
the 25- to 74-year-old US population, we calculated
the population attributable risk of diabetes for body
mass index to be 68 percent during the period of this
study. Performing a similar calculation with body
mass index data from phase I of the Third National
Health and Nutrition Examination Survey (NHANES
HI), we calculated that the population attributable risk
was 72 percent.
Among the entire cohort, weight gain was associated with increased risk for diabetes, whereas weight
loss was associated with a statistically nonsignificant,
decreased risk of diabetes (table 4). In separate analyses, we found a 4.5 percent (95 percent CI 3.3-5.9)
change in risk for every kilogram of change in weight.
Because the analyses suggested that the proportionality assumption was not met for the most extreme
category of weight gain, 20 kg or more, we calculated
the hazard ratios as a function of follow-up time by
dichotomizing follow-up time. For participants who
gained 2:20 kg, the hazard ratio was 0.63 (95 percent
CI 0.17-2.30) during the first 4 years of the follow-up
period and 5.98 (95 percent CI 3.08-11.62) during the
remaining years.
We found no statistically significant interactions for
weight change with age, sex, or race, suggesting that
the risks for diabetes as a function of weight change
were similar among various age groups, among men
and women, and among whites and blacks. For example, hazard ratios for the 30- to 55-year-old group
differed only slightly from those of the full sample,
and the shape of the hazard ratio versus weight change
curve is similar for the two samples. Using the hazard
ratios from table 4, which are adjusted for multiple
confounders including baseline body mass index and
estimates of weight change prevalence from this study,
we estimate that the population attributable risk was
27 percent for weight change during the study period.
The estimate was calculated using the stable weight
gain as the reference category. Thus, we estimate that
about 27 percent of new cases of diabetes could have
been avoided if, by avoiding weight gain, the incidence rates of diabetes among people who gained
Am J Epidemiol
Vol. 146, No. 3, 1997
217
weight could have been reduced to those of people
who maintained relatively stable weights.
Because the Spearman correlation coefficient between baseline body mass index and weight change
was about —0.23, we explored the relation among
baseline body mass index, weight change, and subsequent risk for developing diabetes mellitus further by
running separate proportional hazards models for participants stratified by baseline body mass index (<25
kg/m2, 25-<29 kg/m2, and >29 kg/m2) (table 5). In
general, hazard ratios were similar for the weight gain
categories, whereas some variation in the hazard ratios
for the weight loss category appeared to exist despite
overlapping confidence intervals.
DISCUSSION
Consistent with other studies conducted in special
populations, our results, which are based on a national
sample of adults, show an increased risk for developing diabetes with weight gain (13, 15, 16, 18, 19). The
magnitude of the risks associated with weight gain
from our analysis is more consistent with those reported from the Rancho Bernardo Study (18) and the
University of Pennsylvania Alumni Study (13) than
with the very large risk estimates reported in the
Nurses' Health Study (16) and the Health Professionals' Study (15). Differences in study methodology,
such as sample size, ascertainment of endpoints,
weight measurement, demographic composition of
samples, data analysis, and the period of weight
change, can perhaps account for our lower risk estimates. A potentially important difference is that
weight was measured directly in our study, whereas it
was self-reported in the Nurses' Health Study and the
Health Professionals' Study. In addition, we controlled
for body mass at entry into the study in contrast to the
most recent analysis of the Nurses' Health Study that
controlled for body mass index at 18 years of age (16).
The association between baseline body mass index
and skinfolds and diabetes incidence through 1987 in
the NHEFS was previously reported by Lipton et al.
(14). Although our results are very similar, our analysis differs in several aspects from the earlier analysis
by extending the follow-up time through 1992, by
using body mass index cutpoints similar to those from
the Nurses' Health Study (16), and by using survival
analysis rather than logistic regression. More importantly, we focused primarily on the association of
weight change and diabetes incidence. The shape of
the baseline diabetes risk curve as a function of baseline body mass index, using the same body mass index
categories, was very similar to those reported by
Colditz et al. (16) except that the magnitude of our risk
estimates was consistently lower.
218
Ford et al.
Our study is subject to several limitations. The definition of diabetes was based on data from questionnaires, institutional records, and death certificates. Although death certificates may capture only about one
third to one half of all deaths among persons with
diabetes (30), and hospital records capture about 60
percent of their hospitalizations (31), good agreement
has been reported between self-reported diabetes from
interviews and other sources of data (13, 32). Thus, it
is probable that we underestimated the number of
incident cases of diabetes and the incidence rates. The
hazard ratios may be accurately estimated if similar
proportions of participants with diabetes were misclassified across all body mass index and weight change
categories. Because the association between obesity
and diabetes was well established, those participants
who were leaner or gained smaller amounts of weight
may have been less likely to have had diabetes detected, resulting in overestimates of the hazard ratios.
However, results from NHANES II among persons
without a diabetic parent showed that diabetes had
been diagnosed among a larger percentage of nonobese participants (47.3 percent) than among obese
participants (41.9 percent) (33). In addition, the use of
body mass index to determine obesity has often been
criticized because of the inability of this measure to
discriminate between weight from body fat and weight
from other body structures such as muscle and bone
mass. The likely effect would have been to underestimate the association between weight change and diabetes incidence. Nevertheless, body mass index is a
commonly used measure that has been strongly related
TABLE 2. Baseline characteristic* by diabetes status, Rrst National Health and Nutrition Examination
Survey, 1971-1975
Characteristic
Age (years)t
18-39
40-49
50-59
60-69
£70
Sex
Men
Women
Race
Black
White
Education (years)
<12
£12
Smoking status
Current
Former
Never
Cholesterol (mg/d)
<200
200-239
£240
Systolic Wood pressure (mmUg)
<140
£140
Antihypertenave medication
\BS
No
Body mass index
Overweight
Normal
Drinks per day
<1
1-2
£3
Participants who
developed diabetes
Sample
Age-adjusted
size
percentage ±
(no.)
standard trror*
Participants remaining
free of diabetes
Sample
Age-adjusted
size
percentage ±
(no.)
standard error*
126
111
111
111
28
23.9
24.2
30.3
19.1
2.5
± 2.5
±2.1
± 2.4
±2.2
± 0.5
3,202
1,691
1,256
1,373
536
39.5 ± 0.8
23.9 ± 0.7
21.9 ± 0 . 6
12.7 ± 0 . 5
2.9 ± 0 . 2
190
297
46.0 ± 3.3
54.0 ± 3.3
3,030
5,028
46.1 ± 0 . 8
53.9 ± 0.8
103
384
17.0 ± 2.8
83.0 ± 2.8
901
7,157
7.2 ± 0.6
92.8 ± 0.6
233
44.0 ± 3.0
56.0 ± 3.0
2,874
5,184
30.5 ± 0.9
69.5 ± 0.9
170
82
235
43.4 ± 2.9
17.1 ±2.1
39.5 ±3.1
2,960
1,330
3,768
39.2 ± 0.7
17.9 ± 0 . 6
42.9 ± 0.7
125
183
179
27.6 ± 2.7
38.2 ±3.1
34.2 ±3.1
3,011
2,695
2,352
36.8 ± 0.9
33.8 ± 0.8
29.4 ± 0.7
238
249
58.3 ± 2.9
41.7 ±2.9
2,349
5,709
72.9 ± 0.7
27.1 ± 0 . 7
106
381
19.0 ± 2.8
81.0 ±2.8
1,000
7,058
10.8 ± 0.4
89.2 ± 0.4
102
385
26.5 ± 2.8
73.5 ± 2.8
841
7,217
13.1 ± 0 . 5
86.9 ± 0.5
422
36
29
80.3 ± 2.8
9.0 ± 1.6
10.7 ± 2.4
6,739
80.3 ± 0.9
851
468
ias±o.7
254
7.1 ±0.5
Table 2 continues
Am J Epidemiol
Vol. 146, No. 3, 1997
Weight Change and Diabetes
TABLE 2.
Continued
Participants who
developed diabetes
Age-adjusted
Sample
size
percentage ±
standard error*
(no.)
Characteristic
Weight change*
Loss £11 kg
Loss 5-<11 kg
Loss <5 to gain <5 kg
Gain 5—e8 kg
Gain 8—e11 kg
Gain 11-<20 kg
Gain £20 kg
24
49
218
81
31
60
24
£8
8.6
42.4
17.3
6.2
16.5
6.1
Sample
size
(no.)
Age
Education (years)
Systolic blood pressure (mmHg)
Cholesterol (mg/dT)
Body mass index (kg/m2)
Weight change (kg)
Alcohol consumption (drinks per
day)
± 0.7
± 1.8
± 3.0
± ^6
± 1.3
± ^5
± 1.6
Age-adjusted
mean±
standard error*
487
487
487
487
487
487
281
688
4,480
1,159
670
656
124
2.9
8.3
55.6
14.8
8.9
7.9
1.6
± 0.2
± 0.4
± 0.7
±0.5
± 0.4
± 0.4
±0.2
Sample
size
(no.)
Age-adjusted
mean ±
standard error*
± 0.7
±0.2
± 1.3
± 2.5
± 0.3
± 0.5
8,058
8,058
8,058
8,058
8,058
8,058
44.7
11.9
129.0
218.3
25.0
2.6
0.7 ±0.1
8,058
49.2
11.0
136.2
224.4
28.9
4.8
487
Participants remaining
free o< dbbetes
Sample
Age-adjusted
size
percentage ±
(no.)
standard error*
± 0.2
±0.1
±0.4
±0.9
±0.1
±0.1
0.6 ± <0.1
* Weighted estimate.
t Age-specific results.
i Weight changes determined from baseline (1971-1975) and first follow-up examinations (1982-1984).
TABLE 3. Age-adjusted inoldenca rates of diabetes and risk of diabetes as a function of baseline body mass index, National
Health and Nutrition Examination Survey Eptdemiologic FoDowup Study, 1971-1992
Basetne
body mass
Index
Oqj/mi)
No. of
Incident
cases of
diabetes
Person-years
of
Crude
Incidence
rate (per
100,000
person-years)*
Age-ad)usted
Incidence
rate (per
100,000
person-years)*
Hazard
ratJot
Hazard
raflot
Entim sample
<22
22-22.9
23-23.9
24-24.9
25-26.9
27-28.9
29-30.9
31-32.9
33-34.9
£35
35
16
33
29
73
88
72
43
40
58
19,116
7,090
6,699
6,376
11,600
8,265
4,996
2,958
1,718
2,104
159
196
428
592
555
951
1,095
1,478
1,982
2,636
<22
22-22.9
23-23.9
24-24.9
25-26.9
27-28.9
29-30.9
31-32.9
33-34.9
£35
19
8
20
18
43
53
45
20
26
44
11,443
4,246
3,992
3,776
6,756
4,829
2,832
1,623
940
1,298
118
101
468
571
610
965
1,165
1,183
2,618
3,155
224
255
428
539
553
899
1,074
1,480
1,892
2,461
1.00
1.18(0.48-2.88)§
£ 4 4 (1.32-4.51)
2.97 (1.54-5.73)
3.04(1.75-5.28)
5.07 (3.02-8.52)
5.70 (3.30-9.87)
8.21 (4.44-15.19)
10.89(5.46-21.70)
14.64 (8.68-24.69)
1.00
1.16(0.48-2.82)
2.39 (1.30-4.40)
2.82(1.45-5.50)
2.75(1.55-4.91)
4.63 (2.69-7.96)
4.88 (2.77-8.59)
6.96 (3.79-12.81)
9.28 (4.60-18.72)
11.24(6.66-18.96)
30- to 55-year-old participants
42
75
343
432
597
876
1,084
1,156
2,200
3,293
1.00
0.85 (0.32-2.27)
3.89(1.73-8.75)
4.59(2.07-10.18)
4.94 (2.55-9.57)
7.74 (4.26-14.05)
9.33(4.78-18.23)
9.90 (4.25-23.07)
21.24 (9.61-46.94)
25.08 (14.05-^*4.78)
1.00
0.84 (0.31-2.25)
3.87(1.74-8.58)
4.77(2.15-10.57)
4.91 (2.46-9.81)
7.86(4.23-14.61)
8.88 (4.32-18.26)
8.41 (3.53-19.99)
18.81 (8.61-41.09)
19.07(10.29-35.34)
* Rates were calculated using sampBng weights,
f Adjusted for age and age2.
i Adjusted forage, age2, sex, race, education, education2, smoking status, cholesterol, cholesterol2, systolic Uood pressure, systolic
Hood pressure2, antihypertensive medication, baseline body mass index, and alcohol consumption.
§ Numbers in parentheses, 95% confidence interval.
Am J Epidemiol
Vol. 146, No. 3, 1997
219
220
Ford et al.
TABLE 4. Age-adjusted 1Incidence rates of diabetes and risk of diabetes as a function of weight change, National Health and
Nutrition Examination Survey Epldemlologlc Followup Study, 1971-1992
Weight
change
nu.
oi
Incident
cases of
dtabetes
Parsooysaxs
of
follow-up
Age-trusted
Crude
Incidence
Incidence
rale (per
rale (per
100,000
100,000
person-years)* peison-years)*
Hazard
rattot
Hazard
rattot
Entire sample
Loss £11 kg
Loss5-<11 kg
Loss <5 to gain <5 kg
Gain 5-<8kg
Gain 8—c11 kg
Gain 11-<20kg
Gain 220 kg
24
49
218
81
31
60
24
1,909
5,475
39,420
10,607
6,133
6,139
1,239
1,004
843
481
777
402
1,010
1,554
922
747
477
882
424
1,172
1,432
1.99(1.14-3.46)§
1.66(1.07-2.59)
1.00
1.90(1.23-2.95)
1.09(0.70-1.69)
£ 8 5 (2.02-4.02)
4.60 ( £ 5 1 - 8 . 4 4 )
0.80(0.46-1.40)
1.13(0.72-1.80)
1.00
2.11 (1.40-3.18)
1.19(0.75-1.89)
2.66(1.84-3.85)
3.84 (2.04-7.22)
2.05 (0.93-^.52)
1.52(0.81-2.83)
1.00
2.14(1.42-3.23)
0.82(0.36-1.88)
0.91 (0.46-1.83)
1.00
2.39(1.61-3.54)
1.42(0.83-2.44)
2.70(1.79-*».07)
4.07 (2.05-8.07)
30- to 55-yoar-old participants
Loss 211 kg
Loss5-<11 kg
Loss <5 to gain <5 kg
Gain 5-<8 kg
Gain8-<11 kg
Gain 11-<20 kg
Gain 220 kg
11
21
110
60
24
49
21
769
2,486
22,504
6,786
4,101
4,251
838
913
668
443
888
•
486
1,190
2,099
603
383
360
607
396
1,138
2,419
1.23(0.75-£04)
3.13 (2.09-4.67)
5.28(2.76-10.13)
* Rates were calculated using sampling weights,
t Adjusted for age and age2.
i Adjusted forage, age2, sex, race, education, education2, smoking status, cholesterol, cholesterol2, systoGc blood pressure, systolic
blood pressure2, antihypertensh/e medication, baseline body mass index, and alcohol consumption.
§ Numbers in parentheses, 95% confidence interval.
to adiposity and strongly predictive of risk of future
disease (34).
Furthermore, we were unable to control for other
confounders such as family history of diabetes or
genetic susceptibility. Even though a crude measure of
physical activity had been obtained at the baseline and
first follow-up interviews, we did not consider these
variables to be sufficiently sensitive to allow examination of any interaction between changes in physical
activity and body mass index on the incidence of
diabetes. Because participants had not been systematically tested for glucose intolerance, it is possible that
weight change could have occurred after the onset of
subclinical diabetes. If weight gain occurred after the
onset of diabetes, hazard ratios were likely overestimated. If weight loss occurred after the onset of diabetes, hazard ratios were likely underestimated. Finally,
we were unable to separate persons with insulindependent diabetes mellitus from those with NIDDM.
However, because insulin-dependent diabetes mellitus
usually has an early onset (35), we believe that the vast
majority of incident cases of diabetes were NIDDM.
With the more recent reports, a solid body of evidence has confirmed the increased risk in diabetes
incidence associated with weight gain. More data are
needed to determine the effect of lifetime weight patterns on the risk for NIDDM. For example, results
from the Rancho Bernardo Study suggested that being
underweight as a child or teenager was associated with
increased risk for diabetes (18), and lifetime maximum
weight may be a better predictor of diabetes incidence
than current body weight (1).
In summary, we found that weight gain over a
10-year period was strongly associated with increased
risk for diabetes. In light of the last decade's increase
in the proportion of the population that is considered
overweight and the possibility that the rate of increase
in overweight may be accelerating (36, 37), future
increases in the incidence and prevalence of diabetes
are likely. The induction time for obesity-related diabetes is not well known, but our study shows that
within 10 years of weight gain, increases in diabetes
incidence can occur.
Our data suggest that, for every kilogram of increase
in weight, the risk for diabetes increases by 4.5 percent. Consequently, the average weight gain of 3.6 kg
recorded from NHANES U to NHANES m (36) could
theoretically result in an approximately 16 percent
increase in the incidence of diabetes by the year 2000
compared with that in 1990. This estimate is consistent
with an 18 percent increase in risk for diabetes for
each unit increment in body mass index reported by
Helmrich et al. (13).
The US population is losing the battle of the bulge
despite current public health programs and a multibillion dollar per year weight loss industry. The difficulty
Am J Epidemiol
Vol. 146, No. 3, 1997
Weight Change and Diabetes
221
TABLE 5. Age-adjusted incidence ratM of diabetes «nd risk of diabetes as a function of weight change stratified by baseline
body maw index, National Health and Nutrition Examination Survey Epidemic-logic Followup Study, 1971-1992
No. erf
Incident
cases of
Weigh!
change
Person-years
of
ID Dow-up
Crude
Age-adjusted
Incidence
Incidence
rale (per
rate (per
100,000
100,000
person-years)* person-years)*
Hazanl
raliot
Hazard
rattoj
2.76(1.30-5.85)§
1.00
1.84(1.04-3.26)
1.02(0.42-Z45)
3.33(1.49-7.42)
2.83 (0.99-8.08)
2.27(1.04-^.95)
1.00
1.79(1.01-3.19)
1.10(0.46-2.65)
3.36(1.50-7.51)
3.73(1.13-12.36)
0.67(0.32-1.40)
1.00
2.37(1.33-4.21)
1.34(0.66-2.69)
1.99(1.06-3.76)
4.32(1.80-10.34)
1.06(0.64-1.76)
1.00
2.06 (0.73-5.83)
1.23(0.58-2.58)
2.75(1.55-4.88)
2.92(1-17-7.26)
<25kg/m*
Loss £5 kg
Loss <5 to gain <5 kg
Gain 5-<8 kg
Gain 8—c11 kg
Gain 11—<20 kg
Gam ^20 kg
13
53
23
6
14
4
1,835
22,550
7,073
3,812
3,501
Loss £5 kg
Loss <5 to gain <5 kg
Gain 5-<8 kg
Gain8-<11 kg
Gain 11-<20kg
Gain £20 kg
12
76
34
13
17
2,729
11,199
2,361
1,539
1,657
9
381
Loss £5 kg
Loss <5 to gain <5 kg
Gain 5-<8 kg
Gain 8—<11 kg
Gain 11-c20kg
Gain £20 kg
48
89
24
2,820
5,675
1,174
12
29
11
782
981
348
510
567
793
219
312
155
458
406
242
384
192
440
491
25-<29 kg/mi
627
549
1,002
2,592
1,025
3,548
0.70(0.34-1.46)
1.00
2.38(1.37-4.14)
1.34(0.67-£71)
2.29(1.28-4.09)
6.11 (2.45-15.19)
1,513
1,244
2,456
1,111
2,901
2,049
1,320
1,256
2,928
1,156
3,963
1,449
1.25(0.76-2.06)
1.00
2.11 (0.70-6.39)
1.06(0.50^24)
2.64(1.53-4.56)
2.26(0.94-5.41)
393
580
377
536
1,280
1,164
* Rates were calculated using sampling weights,
f Adjusted for age and age2.
X Adjusted forage, age2, sex, race, education, education2, smoking status, cholesterol, cholesterol2, systolic blood pressure, systolic
blood pressure2, antihypertensive medication, baseline body mass index, and alcohol consumption.
§ Numbers in parentheses, 95% confidence interval.
in losing weight and maintaining a healthy body
weight is well known (38). Clearly, new public health
actions are urgently needed to prevent weight gain in
the US population. Along with other researchers (16),
we urge health care providers to more consistently
counsel their patients to prevent weight gain. In light
of the accumulating evidence linking weight gain to an
increased risk of diabetes, the appropriateness of
screening for diabetes among persons experiencing
substantial weight gain deserves review.
7.
8.
9.
10.
11.
REFERENCES
1. West KM. Epidemiology of diabetes and its vascular lesions.
New York: Elsevier North-Holland, Inc, 1978:234-48.
2. Barrett-Connor E. Epidemiology, obesity, and non-insulindependent diabetes mellitus. Epidemiol Rev 1989;11:172—81.
3. Reaven GM. Role of insulin resistance in human disease.
Diabetes 1988;37:1595-607.
4. Felber JP, Acheson KJ, Tappy L. From obesity to diabetes.
New York: John Wiley & Sons, Inc, 1993.
5. Unger RH. Lipotoxicity in the pathogenesis of obesity-dependent NIDDM. Diabetes 1995;44:863-70.
6. Westlund K, Nicolaysen R. Ten-year mortality and morbidity
Am J Epidemiol
Vol. 146, No. 3, 1997
12.
13.
14.
related to serum cholesterol. Scand J Clin Lab Invest 1972;
30(suppl 127): 1-24.
Pyflrala K, Lehtovirta E, Ilomaki L. The relationship of obesity to prevalence and incidence of diabetes in Helsinki policemen. Acta Endocrinol 1974;181:23-4.
Medalie JH, Papier CM, Goldbourt U, et al. Major factors in
the development of diabetes mellitus in 10,000 men. Arch
Intern Med 1975;135:811-18.
Knowler WC, Pettitt DJ, Savage PJ, et al. Diabetes incidence
in Pima Indians: contributions of obesity and parental diabetes. Am J Epidemiol 1981;113:144-56.
Wilson PW, McGee DL, Kannel WB. Obesity, very low
density lipoproteins, and glucose intolerance over fourteen
years. Am J Epidemiol 1981; 114:697-704.
Modan M, Karasik A, Halkin H, et al. Effect of past and
concurrent body mass index on prevalence of glucose intolerance and type 2 (non-insulin-dependent) diabetes and on
insulin response. Diabetologia 1986;29:82-9.
Ohlson LO, Larsson B, BjOrntorp P, et al. Risk factors for type
2 (non-insulin-dependent) diabetes mellitus: thirteen and onehalf years of follow-up of the participants in the study of men
born in 1913. Diabetologia 1988;31:789-805.
Helmrich SP, Ragland DR, Leung RW, et al. Physical activity
and reduced occurrence of non-insulin-dependent diabetes
mellitus. N Engl J Med 1991 ;325:147-52.
Lipton RB, Liao Y, Cao G, et al. Determinants of incident
non-insulin-dependent diabetes mellitus among blacks and
whites in a national sample. Am J Epidemiol 1993; 138:
826-39.
222
Ford et al.
15. Chan JM, Stampfer MJ, Rimm EB, et al. Obesity, fat distribution, and weight gain as risk factors for clinical diabetes in
men. Diabetes Care 1994;17:961-9.
16. Colditz GA, Willett WC, Rotnitzky A, et al. Weight gain as a
risk factor for clinical diabetes mellitus in women. Ann Intern
Med 1995;122:481-6.
17. Noppa H. Body weight change in relation to incidence of
ischemic heart disease and change in risk factors for ischemic
heart disease. Am J Epidemiol 1980; 111:693-704.
18. Holbrook TL, Barrett-Connor E, Wingard DL. The association
of lifetime weight and weight control patterns with diabetes
among men and women in an adult community. Int J Obes
1989;13:723-9.
19. Hanson RL, Narayan KMV, McCance DR, et al. Rate of
weight gain, weight fluctuation, and incidence of NIDDM.
Diabetes 1995;43:261-6.
20. National Center for Health Statistics. Plan and operation of the
Health and Nutrition Examination Survey, United
States—1971-1973. Washington, DC: US GPO, 1973. (Vital
and health statistics, Series 1: no. 10a) (DHHS publication no.
(PHS) 79-1310).
21. National Center for Health Statistics. Plan and operation of the
Health and Nutrition Examination Survey, United
States—1971-1973. Washington, DC: US GPO, 1977. (Vital
and health statistics, Series 1: no. 10b) (DHHS publication no.
(PHS) 79-1310).
22. National Center for Health Statistics. Plan and operation of the
HANES I Augmentation Survey of Adults 25-74 Years,
United States, 1974-1975. Washington, DC: US GPO, 1978.
(Vital and health statistics, Series 1: no. 14) (DHHS publication no. (PHS) 78-1314).
23. National Center for Health Statistics. Plan and operation of the
NHANES I Epidemiologic Followup Study, 1982-1984.
Washington, DC: US GPO, 1987. (Vital and health statistics,
Series 1: no. 22) (DHHS publication no. (PHS) 87-1324).
24. National Center for Health Statistics. Plan and operation of the
NHANES I Epidemiologic Followup Study, 1987. Washington, DC: US GPO, 1992. (Vital and health statistics, Series 1:
no. 27) (DHHS publication no. (PHS) 92-1303).
25. National Center for Health Statistics. HANES I hematology
and clinical chemistry procedures developed or utilized by the
Center for Disease Control, Bureau of Laboratories,
1971-1975. Washington, DC: US GPO, 1979.
26. McLaughlin JK, Dietz MS, Mehl ES, et al. Reliability of
surrogate information on cigarette smoking by type of informant. Am J Epidemiol 1987;126:144-6.
27. Machlin SR, Kleinman JC, Madans JH. Validity of mortality
analysis based on retrospective smoking information. Stat
Med 1989;8:997-1009.
28. Shah BV, Barnwell BG, Hunt PN, et al. SUDAAN user's
manual. Release 5.50. Research Triangle Park, NC: Research
Triangle Institute, 1991.
29. Kleinbaum DG, Kupper LL, Morgenstem H. Epidemiologic
research: principles and quantitative methods. New York: Van
Nostrand Reinhold Company, 1982:163.
30. Bild DE, Stevenson JM. Frequency of recording of diabetes on
US death certificates: analysis of the 1986 National Mortality
Followback Survey. J Clin Epidemiol 1991;45:275-81.
31. Ford ES, Wetterhall S. The validity of diabetes on hospital
discharge diagnoses. (Abstract). Diabetes 1991;40(suppl 1):
449A.
32. Harlow SD, Linet MS. Agreement between questionnaire data
and medical records: the evidence for accuracy of recall. Am J
Epidemiol 1989;129:233-48.
33. Harris MI, Hadden WC, Knowler WC, et al. Prevalence of
diabetes and impaired glucose tolerance and plasma glucose
levels in U.S. population aged 20-74 yr. Diabetes 1987;36:
523-34.
34. Spiegelman D, Israel RG, Bouchard C, et al. Absolute fat
mass, percent body fat, and body-fat distribution: Which is the
real determinant of blood pressure and serum glucose? Am J
Clin Nutr 1992;55:1033-44.
35. LaPorte RE, Matsushima M, Chang Y-F. Prevalence and
incidence of insulin-dependent diabetes. In: National Diabetes
Data Group. Diabetes in America. Bethesda, MD: National
Institutes of Health, 1995:37-46. (NIH publication no.
95-1468).
36. Kuczmarski RJ, Flegal KM, Campbell SM, et al. Increasing
prevalence of overweight among US adults. JAMA 1994;272:
205-11.
37. Galuska DA, Serdula M, Pamuk E, et al. Trends in overweight
among U.S. adults from 1987-1993: results from a multistate
telephone survey. Am J Public Health 1996;86:1729-35.
38. Goodrick GK, Foreyt JP. Why treatments for obesity don't
last. J Am Diet Assoc 1991;91:1243-7.
Am J Epidemiol
Vol. 146, No. 3, 1997