BMJ
Helping doctors make better decisions
- Isao Muraki, research fellow1,
- Fumiaki Imamura, investigator scientist2,
- JoAnn E Manson, professor of medicine345,
- Frank B Hu, professor of nutrition and epidemiology135,
- Walter C Willett, professor of epidemiology and nutrition135,
- Rob M van Dam, associate professor16,
- Qi Sun, assistant professor15
- Correspondence to: Q Sun qisun@hsph.harvard.edu
- Accepted 10 July 2013
Abstract
Objective To determine whether individual fruits are differentially associated with risk of type 2 diabetes.
Design Prospective longitudinal cohort study.
Setting Health professionals in the United States.
Participants
66 105 women from the Nurses’ Health Study (1984-2008), 85 104 women
from the Nurses’ Health Study II (1991-2009), and 36 173 men from the
Health Professionals Follow-up Study (1986-2008) who were free of major
chronic diseases at baseline in these studies.
Main outcome measure Incident cases of type 2 diabetes, identified through self report and confirmed by supplementary questionnaires.
Results
During 3 464 641 person years of follow-up, 12 198 participants
developed type 2 diabetes. After adjustment for personal, lifestyle, and
dietary risk factors of diabetes, the pooled hazard ratio of type 2
diabetes for every three servings/week of total whole fruit consumption
was 0.98 (95% confidence interval 0.96 to 0.99). With mutual adjustment
of individual fruits, the pooled hazard ratios of type 2 diabetes for
every three servings/week were 0.74 (0.66 to 0.83) for blueberries, 0.88
(0.83 to 0.93) for grapes and raisins, 0.89 (0.79 to 1.01) for prunes,
0.93 (0.90 to 0.96) for apples and pears, 0.95 (0.91 to 0.98) for
bananas, 0.95 (0.91 to 0.99) for grapefruit, 0.97 (0.92 to 1.02) for
peaches, plums, and apricots, 0.99 (0.95 to 1.03) for oranges, 1.03
(0.96 to 1.10) for strawberries, and 1.10 (1.02 to 1.18) for cantaloupe.
The pooled hazard ratio for the same increment in fruit juice
consumption was 1.08 (1.05 to 1.11). The associations with risk of type 2
diabetes differed significantly among individual fruits (P<0.001 in
all cohorts).
Conclusion Our findings
suggest the presence of heterogeneity in the associations between
individual fruit consumption and risk of type 2 diabetes. Greater
consumption of specific whole fruits, particularly blueberries, grapes,
and apples, is significantly associated with a lower risk of type 2
diabetes, whereas greater consumption of fruit juice is associated with a
higher risk.
Introduction
Fruits
are rich in fibre, antioxidants, and phytochemicals that may have
beneficial health effects. Increasing fruit consumption has been
recommended for the primary prevention of many chronic diseases,
including type 2 diabetes,1 although epidemiologic studies have generated somewhat mixed results regarding the link with risk of type 2 diabetes.2 3 4 5 6 7 8 9 10
The inconsistency among these studies may be explained by differences
in types of fruits consumed in different study populations as well as
difference in participants’ characteristics, study design, and
assessment methods, although a meta-analysis did not show that the
associations differed by sex, study design, or location.10
Furthermore, in a recent study, the greater variety, but not quantity,
of fruits consumed was associated with a lower risk of type 2 diabetes.4
This finding suggested that individual fruits might not be equally
associated with risk of type 2 diabetes in that fruits have highly
variable contents of fibre, antioxidants, other nutrients, and
phytochemicals that jointly may influence the risk.11 12
Additionally, the glycemic index, which represents the quality of
carbohydrate, or glycemic load, which represents the quality and
quantity of carbohydrate and their interaction, vary substantially for
individual fruits.13
We
examined the associations of individual fruit consumption in relation
to risk of type 2 diabetes using data from three prospective cohort
studies in US adults. Moreover, we estimated substitution effects of
individual fruits for fruit juice in relation to risk of type 2
diabetes. Secondarily, we examined the associations of fruit groups
based on their glycemic index and glycemic load values with risk of type
2 diabetes.
Methods
Study population
We
used data from the Nurses’ Health Study (established in 1976;
n=121 700), the Nurses’ Health Study II (established in 1989;
n=116 671), and the Health Professionals Follow-up Study (established in
1986; n=51 529). These cohort studies are discussed in detail
elsewhere.14 15 16
Every two years since baseline, follow-up questionnaires have been
mailed to the participants to collect and update information on
lifestyle practices and occurrence of chronic diseases. In all three
cohorts the follow-up rates are approximately 90%.
We
excluded participants who reported a diagnosis of diabetes (including
types 1 and 2 and gestational diabetes), cardiovascular disease, or
cancer at baseline (n=10 134 for the Nurses’ Health Study, 6155 for the
Nurses’ Health Study II, and 6707 for the Health Professionals Follow-up
Study), those who had missing data for individual fruits and fruit
juice or an unusual level of total energy intake (<500 or >3500
kcal/day for the Nurses’ Health Study and the Nurses’ Health Study II
and <800 or >4200 kcal/day for the Health Professionals Follow-up
Study) (n=4765 for the Nurses’ Health Study, 5647 for the Nurses’ Health
Study II, and 5750 for the Health Professionals Follow-up Study), those
whose diagnosis date of type 2 diabetes was unclear (n=200 for Health
Professionals Follow-up Study), and those who completed only the
baseline questionnaire (n=719 for the Nurses’ Health Study, 699 for the
Nurses’ Health Study II, and 1103 for the Health Professionals Follow-up
Study). After excluding these participants, 66 105 women in the Nurses’
Health Study, 85 104 women in the Nurses’ Health Study II, and 36 173
men in the Health Professionals Follow-up Study were available for the
analysis.
Assessment of fruit consumption
In
1984 a 118 item food frequency questionnaire was sent to the
participants of the Nurses’ Health Study to assess their habitual diet
in the past year. In 1986 and every four years thereafter, a similar but
expanded questionnaire was sent to the participants to update their
dietary information. The expanded questionnaire was also administered
every four years to assess diet among the participants in the Health
Professionals Follow-up Study since 1986 and those in the Nurses’ Health
Study II since 1991. In all food frequency questionnaires we asked the
participants how often, on average, they consumed each food in a
standard portion size. Participants could choose from nine possible
responses, ranging from “never, or less than once per month” to “six or
more times per day.” We consistently asked about 10 individual fruits
since baseline: grapes or raisins; peaches, plums, or apricots; prunes;
bananas; cantaloupe; apples or pears; oranges; grapefruit; strawberries;
and blueberries. We calculated total whole fruit consumption by summing
the consumption levels of the 10 individual fruits and watermelon,
which we inquired about sporadically during follow-up. Fruit juice
included apple, orange, grapefruit, and other juices. The food frequency
questionnaires were validated against diet records among 173
participants in the Nurses’ Health Study in 1980 and 127 participants in
the Health Professionals Follow-up Study in 1986.17 18 19
Corrected correlation coefficients between food frequency questionnaire
and diet record assessments of individual fruit consumption were 0.80
for apples, 0.79 for bananas, and 0.74 for oranges in women,17
and 0.67 for total whole fruits, 0.76 for fruit juice, 0.95 for
bananas, 0.84 for grapefruit, 0.76 for oranges, 0.70 for apples and
pears, 0.59 for raisins and grapes, and 0.38 for strawberries in men.18 19
For some individual fruits, the corrected correlation coefficients were
not available owing to large within person variability in the
comparison methods.
Assessment of covariates
In
the follow-up questionnaires administered every two years, we inquired
and updated information on anthropometric and lifestyle factors for
chronic diseases, including body height and weight, cigarette smoking,
physical activity, multivitamin use, and family history of diabetes.
Among participants in the Nurses’ Health Study and Nurses’ Health Study
II, we ascertained menopausal status, post-menopausal hormone use, and
oral contraceptive use (Nurses’ Health Study II only). Estimates of
total physical activity levels were calculated by multiplying the energy
expenditure in metabolic equivalent tasks (METs) measured in hours per
week of each activity by hours spent on the activity and summing the
values of all activities. Each MET hour is the caloric need per kilogram
of body weight per hour of an activity, divided by the caloric need per
kilogram of weight per hour at rest. Based on the food frequency
questionnaire, we derived a score of the alternate healthy eating index,
an indicator of adherence to healthy eating behavior, described in
detail elsewhere.20
In brief, the alternate healthy eating index score summarizes the
consumption of 11 foods or nutrients (including consumption of
vegetables, fruits, whole grains, sugar sweetened beverages and fruit
juice, nuts and legumes, red and processed meat, trans fat,
long chain n-3 fat, polyunsaturated fat, sodium, and alcohol). Each
component was scored on a scale of 0 to 10. In the current analysis, we
excluded fruits and fruit juice when calculating the alternate healthy
eating index score.
Assessment of diabetes and death
In
all three cohorts, to inquire about symptoms, diagnostic tests, and
diabetes drug use we mailed a supplementary questionnaire to
participants who reported physician diagnosed diabetes in the follow-up
questionnaires. A type 2 diabetes diagnosis was confirmed if
participants met at least one of the following National Diabetes Data
Group criteria21:
one or more classic symptoms (excessive thirst, polyuria, weight loss,
and hunger) plus raised blood glucose levels (fasting levels ≥140 mg/dL
(7.8 mmol/L), random blood levels ≥200 mg/dL (11.1 mmol/L), and/or two
hour blood glucose levels ≥200 mg/dL during oral glucose tolerance
testing), raised blood glucose levels on two different occasions in the
absence of symptoms, or treatment with antidiabetic drugs (insulin or
oral antidiabetic agent). The diagnostic criteria changed in June 1998
and a fasting blood glucose level of 126 mg/dL (7.0 mmol/L) instead of
140 mg/dL was considered the threshold for the diagnosis of diabetes.22 The validity of the supplementary questionnaire for the diagnosis of diabetes has been examined in validation studies.23 24
Of 62 self reported cases of type 2 diabetes randomly selected in the
Nurses’ Health Study, 61 (98%) were confirmed after an endocrinologist
reviewed the medical records without the information from the
supplementary questionnaire23;
and in the Health Professionals Follow-up Study, 57 of 59 self reported
cases of type 2 diabetes (97%) were confirmed by a review of medical
records.24
Deaths
were identified by reports from next of kin or postal authorities, or
by searching the national death index. At least 98% of deaths were
identified among the participants.25
Statistical analysis
We
calculated each participant’s person years from the return date of the
baseline food frequency questionnaire to the date of the type 2 diabetes
diagnosis, date of death, last return of a valid follow-up
questionnaire, or end of follow-up (2008 for the Nurses’ Health Study
and Health Professionals Follow-up Study, or 2009 for the Nurses’ Health
Study II), whichever came first. To represent long term dietary intake
and minimize within person variation, we calculated and used the
cumulative average of dietary intake based on valid assessments from
baseline to the end of follow-up.26
To minimize the effects of chronic diseases diagnosed during follow-up
on subsequent diet, we stopped updating dietary information after self
reported diagnosis of hypertension, hypercholesterolemia, gestational
diabetes, cardiovascular disease, or cancer, since these chronic
diseases may lead to changes of fruit consumption levels in the cohorts.27
To reduce the effect of potential outliers and to pool the results from
the three cohorts, we used the same cut-off points to categorize
consumption levels in these studies. The highest two consumption levels
were combined for prunes, cantaloupe, and blueberries owing to the small
number of participants with high consumption levels of these fruits.
To
minimize missing covariates, we replaced missing data on body mass
index and physical activity with the last valid values. For missing data
on body mass index and physical activity at baseline, we created a
dummy variable when making categories for these two continuous
covariates. Similarly, we also used missing indicator variables to
include participants with missing categorical variables, including
cigarette smoking, oral contraceptive use (Nurses’ Health Study II
only), menopausal status, and post-menopausal hormone use. The overall
percentages of missing data for body mass index and physical activity
were, respectively, 6.8% and 9.5% in the Nurses’ Health Study, 6.7% and
8.5% in the Nurses’ Health Study II, and 12.8% and 14.0% in the Health
Professionals Follow-up Study. The overall percentages of missing values
during follow-up ranged from 0.6% (for cigarette smoking) to 5.1% (for
menopausal status and post-menopausal hormone use) in the Nurses’ Health
Study, from 0.3% (for cigarette smoking) to 4.7% (for menopausal status
and post-menopausal hormone use) in the Nurses’ Health Study II, and
from 0.1% (for physical activity) to 12.2% (for cigarette smoking) in
the Health Professionals Follow-up Study.
Using Cox
proportional hazard regression, we estimated the hazard ratios and 95%
confidence intervals of type 2 diabetes for fruit consumption. We tested
the proportional hazard assumption by including interaction terms
between individual fruit consumption and duration of follow-up, and the
assumption was unlikely violated (P>0.05 for all tests). We examined
linear trend by modelling the median values for fruit consumption
categories as a continuous variable. Using a fixed effects model, we
pooled multivariable adjusted hazard ratios from three cohorts, and we
used the Cochrane Q statistic and the I2 statistic to examine the heterogeneity of associations among the cohorts.
To
examine whether the associations with risk of type 2 diabetes were
heterogeneous among individual fruits, we fitted two fully adjusted
models: one with total fruit consumption and the other with total fruit
consumption plus consumption of individual fruits excluding oranges
(which had the most similar association as the total fruit consumption)
to avoid over-fitting. Then we used the likelihood ratio test to examine
whether the model including individual fruits had better fit than that
including total fruit consumption only.
We also
estimated potential effects of substituting specific fruit consumption
for fruit juice consumption by examining the median values for
consumption categories of individual fruits and fruit juice in the same
multivariate model; the hazard ratios and 95% confidence intervals for
substitution effects were calculated based on the differences in point
estimates, and the variance and covariance for the regression
coefficients of specific fruits and fruit juice.28
To examine the robustness of our findings, we also conducted four
sensitivity analyses: evaluating the influence of adjustment for major
dietary variables including polyunsaturated to saturated fat ratio, and
intakes of trans fat, red meat, fish, whole grains, sugar sweetened
beverages, coffee, and nuts (all in fifths) instead of the modified
alternate healthy eating index score; adjusting for baseline body mass
index instead of updated body mass index to estimate the impact of
potential over-adjustment; using baseline consumption levels as an
exposure instead of cumulative average of intake levels; and stopping
updating diet after diagnosis of gestational diabetes, cardiovascular
disease, or cancer only when calculating the cumulative averages.
We
further examined whether the associations of individual fruit
consumption with risk of type 2 diabetes depended on the glycemic
index/glycemic load values of fruits. We calculated the glycemic load
values per serving for individual fruits based on the glycemic index
values from the international glycemic index database13 and the amount of carbohydrate in fruits from the USDA nutritional database11
(see supplementary table 1). We categorized individual fruits into
three groups based on their glycemic load values per serving: prunes,
bananas, grapes, raisins, apples, and pears for high glycemic load
fruits (glycemic load 8.1-19.2); cantaloupe, blueberries, and oranges
for moderate glycemic load fruits (5.7-8.0); and peaches, plums,
apricots, strawberries, and grapefruit for low glycemic load fruits
(1.3-5.6). In terms of the categorization of fruits by the glycemic
index values, high glycemic index fruits included cantaloupe, bananas,
grapes, raisins (glycemic index 60-70); moderate glycemic index fruits
included prunes, blueberries, and grapefruit (47-59); and low glycemic
index fruits included apples, pears, oranges, peaches, plums, apricots,
and strawberries (34-46). Moreover, to estimate the degree to which the
observed associations were explained by flavonoid intake, in a secondary
analysis we further adjusted for intake of flavonoid subtypes
(flavonols, flavones, flavanones, flavan-3-ols, and anthocyanins).
The analysis was stratified jointly by age and calendar year and adjusted for body mass index (kg/m2;
<23, 23.0-24.9, 25.0-26.9, 27.0-28.9, 29.0-30.9, 31.0-32.9,
33.0-34.9, 35.0-36.9, 37.0-38.9, 39.0-40.9, 41.0-42.9, 43.0-44.9, ≥45.0,
or missing), ethnicity (white, African-American, Hispanic, or Asian),
physical activity (MET hours/week; <3, 3.0-8.9, 9.0-17.9, 18.0-26.9,
≥27.0, or missing), cigarette smoking (never, former, currently smoke
1-14 cigarettes/day, currently smoke 15-24 cigarettes/day, or currently
smoke ≥25 cigarettes/day, or missing), multivitamin use (yes or no),
family history of diabetes (yes or no), menopausal status and
post-menopausal hormone use (premenopause, post-menopause (never,
former, or current hormone use), or missing) (for women), oral
contraceptive use (yes, no, or missing) (Nurses’ Health Study II only),
total energy intake (kcal/day), fruit juice consumption (fifths), and
the modified alternate healthy eating index score (fifths).20
When examining the association for total whole fruit, we included total
fruit consumption in the multivariate model without further adjusting
for individual fruits. Likewise, when examining the associations for
individual fruits or fruit groups based on their glycemic index/glycemic
load values, we included consumption levels of all other individual
fruits or fruit groups instead of total fruit consumption in the final
model.
Statistical analyses were performed with SAS 9.2.
All P values were two sided, and statistical significance was defined
as P<0.05.
Results
During
3 464 641 person years of follow-up, 12 198 participants developed type
2 diabetes (Nurses’ Health Study: 6358 cases/1 394 127 person years;
Nurses’ Health Study II: 3153 cases/1 416 111 person years; Health
Professionals Follow-up Study: 2687 cases/654 403 person years). The
rate of loss to follow-up was low and similar between extreme comparison
groups of total fruit consumption levels: the average rate of loss to
follow-up for each two year follow-up cycle was 0.8% for <4
servings/week and 0.7% for ≥3 servings/day of total fruit consumption in
the Nurses’ Health Study. These values were 0.1% and 0.1% in the
Nurses’ Health Study II and 1.0% and 1.1% in the Health Professionals
Follow-up Study, respectively.
In all three cohorts,
total whole fruit consumption was positively correlated with age,
physical activity, multivitamin use, total energy intake, fruit juice
consumption, and the modified alternate health eating index score, and
was inversely associated with body mass index and current smoking (table
1⇓).
Whole fruit consumption was associated with an increased probability of
using post-menopausal hormones in the Nurses’ Health Study and with a
reduced probability of using oral contraceptives in the Nurses’ Health
Study II. Individual fruits were correlated with each other weakly to
moderately; the highest Spearman correlation coefficients were 0.44
between apples and oranges in the Nurses’ Health Study, 0.47 between
strawberries and peaches in the Nurses’ Health Study II, and 0.48
between strawberries and blueberries in the Health Professionals
Follow-up Study (see supplementary table 2). The Spearman correlation
coefficients for total whole fruits in relation to the modified
alternate healthy eating index score were 0.22 for the Nurses’ Health
Study, 0.29 for the Nurses’ Health Study II, and 0.28 for the Health
Professionals Follow-up Study; those for individual fruits ranged from
0.09 (for bananas) to 0.23 (for apples and pears) in the Nurses’ Health
Study, from 0.13 (for bananas) to 0.24 (for apples and pears) in the
Nurses’ Health Study II, and from 0.09 (for peaches, plums, and
apricots) to 0.24 (for apples and pears) in the Health Professionals
Follow-up Study.
View this table:
Table 1
Baseline characteristics of women in Nurses’ Health Study and Nurses’
Health Study II and men in Health Professionals Follow-up Study. Values
are means (standard deviations) or percentages unless stated otherwise
Total
whole fruit consumption was weakly associated with a lower risk of type
2 diabetes: the hazard ratio (95% confidence interval) of type 2
diabetes for every three servings/week of whole fruit consumption was
0.98 (0.96 to 0.99) (table 2⇓).
In the age adjusted model, each individual fruit consumption was
inversely associated with risk of type 2 diabetes in all cohorts (all
P<0.001) (see supplementary table 3). Adjustment for personal
factors, lifestyle, fruit juice consumption, and the modified alternate
health eating index score attenuated these associations. The inverse
association for cantaloupe consumption was no longer statistically
significant after multivariable adjustments of the aforementioned
covariates. Further adjustment for other individual fruit consumption
changed the associations to various degrees (table 3⇓).
The inverse associations for grapes and blueberries were attenuated,
albeit remaining statistically significant. In contrast, associations
for strawberries were attenuated toward the null, and cantaloupe
consumption was associated with an increased risk of type 2 diabetes
after adjustment for other individual fruits. For every three
servings/week, the pooled hazard ratios (95% confidence intervals) of
risk for type 2 diabetes was 0.74 (0.66 to 0.83) for blueberries, 0.88
(0.83 to 0.93) for grapes and raisins, 0.93 (0.90 to 0.96) for apples
and pears, 0.95 (0.91 to 0.98) for bananas, and 0.95 (0.91 to 0.99) for
grapefruit. In contrast, the pooled hazard ratio (95% confidence
interval) of risk for type 2 diabetes for the same increment in
cantaloupe consumption was 1.10 (1.02 to 1.18). A test for heterogeneity
among three cohorts was significant for the associations of bananas and
strawberries (P for heterogeneity <0.001 for bananas and 0.01 for
strawberries). In the Nurses’ Health Study II and Health Professionals
Follow-up Study, banana consumption was associated with a lower risk of
type 2 diabetes, whereas in the Nurses’ Health Study a non-significant
positive association was found. The association for strawberry
consumption was significantly positive in the Health Professionals
Follow-up Study but was non-significant and inverse in the Nurses’
Health Study. Further adjustment for intake levels of flavonoid subtypes
(flavonols, flavones, flavanones, flavan-3-ols, and anthocyanins) in
the final model did not appreciably attenuate the associations for
individual fruits (see supplementary table 4).
View this table:
Table 2
Pooled hazard ratios (95% confidence intervals) of type 2 diabetes for
total whole fruit consumption in Nurses’ Health Study, Nurses’ Health
Study II, and Health Professionals Follow-up Study
View this table:
Table 3
Pooled hazard ratios (95% confidence intervals) of type 2 diabetes for
individual whole fruit consumption in Nurses’ Health Study, Nurses’
Health Study II, and Health Professionals Follow-up Study
The
goodness of fit of model was significantly improved by adding
consumption of individual fruits to the model with total whole fruit
consumption and other covariates (P<0.001 for likelihood ratio test
in each cohort), indicating that heterogeneity in the associations with
risk of type 2 diabetes among individual fruits was significant.
In
the sensitivity analyses, the associations for individual fruits did
not change appreciably with adjustment for major dietary factors instead
of the modified alternate healthy eating index score, or using baseline
consumption levels of individual fruits instead of cumulative average
of intake levels (see supplementary table 5). When we adjusted for
baseline body mass index instead of updated body mass index or stopped
updating diet after diagnosis of gestational diabetes, cardiovascular
disease, or cancer when calculating the cumulative average of dietary
intake, the inverse associations became weaker, although the
associations for blueberries, grapes and raisins, and apples and pears
remained statistically significant.
In the secondary
analysis examining the associations between fruit consumption and risk
of type 2 diabetes by glycemic index/glycemic load values of fruits,
greater consumption of high glycemic load fruits was associated with a
lower risk of type 2 diabetes, but not moderate and low glycemic load
fruits (table 4⇓).
In contrast, greater consumption of moderate glycemic index fruits, but
not high and low glycemic index fruits, was inversely associated with
risk of type 2 diabetes. Fruit juice consumption was associated with an
increased risk of type 2 diabetes. The associations for low glycemic
load fruits were heterogeneous among cohorts (P for heterogeneity
=0.04): a significant, inverse association was found in the Nurses’
Health Study, but not in the other two cohorts.
View this table:
Table 4
Pooled hazard ratios (95% confidence intervals) of type 2 diabetes* for
consumption of fruit juice and fruits grouped by their glycemic
index/glycemic load in Nurses’ Health Study, Nurses’ Health Study II,
and Health Professionals Follow-up Study
Replacing
each three servings/week of fruit juice consumption with the same
amount of total or individual whole fruits, the risk of type 2 diabetes
in the pooled analysis was 7% (95% confidence interval 4% to 9%) lower
for total whole fruits, 33% (24% to 40%) lower for blueberries, 19% (14%
to 24%) lower for grapes and raisins, 14% (11% to 18%) lower for apples
and pears, 13% (9% to 16%) lower for bananas, and 12% (8% to 17%) lower
for grapefruit after adjustment for personal factors, lifestyle, and
the modified alternate health eating index score (figure⇓).
Additionally, we found that replacing fruit juice with oranges,
peaches, plums, and apricots was also associated with a lower risk of
type 2 diabetes: 18% (8% to 28%) lower for prunes, 11% (5% to 16%) lower
for peaches, plums, and apricots, and 8% (4% to 12%) lower for oranges.
Pooled
multivariate adjusted hazard ratios and 95% confidence intervals (error
bars) of type 2 diabetes for substituting three servings/week of total
or specific fruit for the same amount of fruit juice
Discussion
In
three prospective cohorts of US men and women, we found that the
associations with risk of type 2 diabetes differed significantly among
individual fruits: greater consumption of blueberries, grapes, apples,
bananas, and grapefruit were significantly associated with a reduced
risk of type 2 diabetes. Most of these associations were quite
consistent among three cohorts. Additionally, differences in the
glycemic index/glycemic load values of fruits did not account for the
association of specific fruits with risk of type 2 diabetes. Moreover,
greater fruit juice consumption was associated with an increased risk,
and substitution of whole fruits for fruit juice was associated with a
lower risk, except for strawberries and cantaloupe.
Results in relation to other studies
In eight previous prospective studies, the association between total fruit consumption and risk of type 2 diabetes was examined,2 3 4 5 6 7 8 9 and the results were mixed. Similar to previous analyses in the Nurses’ Health Study3 and the Finnish Mobile Clinic Health Examination Survey study,2
the current findings supported an inverse association between total
fruit consumption and risk of type 2 diabetes, but not in other studies.4 5 6 7 8 9
In contrast to total fruit consumption, evidence on consumption of
individual fruits or fruit groups with risk of type 2 diabetes is
limited and incomplete. In four prospective studies, consumption of
citrus fruit was not associated with a lower risk of type 2 diabetes.5 6 7 8 Apple consumption was inversely associated with risk in the Women’s Health Study29 and in the Finnish study,30 but not in the Iowa Women’s Health Study.31 In addition, greater consumption of berries was associated with a lower risk in the Finnish study,2 but not in the Iowa Women’s Health Study.31
In our previous analyses that focused on anthocyanin rich fruits,
intakes of blueberries, strawberries, and apples were associated with a
lower risk of type 2 diabetes.32 Consistently, in a clinical trial, increased consumption of berries improved glycemic control among people with diabetes.33
Our current investigation extended the evidence in this regard and
found novel, inverse associations for grapes, bananas, and grapefruit.
The
different associations of individual fruits with diabetes risk may be
due to the heterogeneous composition of these foods. Firstly,
blueberries, apples, and red or black grapes contain high levels of
anthocyanins.12
In mice with diabetes, bilberry extract rich in anthocyanins can
activate adenosine monophosphate-activated protein kinase, enhance
glucose uptake and utilization in white adipose tissue and skeletal
muscle, and reduce glucose production in the liver.34 Our previous analyses also showed that levels of anthocyanin intake were inversely associated with risk of type 2 diabetes.32
In the current study, further adjustment for anthocyanins did not
substantially change the associations for individual fruits, suggesting
that the inverse associations of individual fruits are likely due to
other constituents of these foods. Both red and white grapes contain
high levels of resveratrol in skin.35
In mice, a high fat diet with 0.04% resveratrol increased insulin
sensitivity at 24 months compared with the same diet without
resveratrol.36
However, randomized controlled trials examining the effects of
supplementation of resveratrol on glucose metabolism have generated
inconsistent results.37 38 39 Prunes, peaches, plums, apricots, and apples contain chlorogenic acid,40 41 42 43 which may potentially mediate the beneficial effects of coffee consumption on diabetes risk.44
In rats, chlorogenic acid reduces glucose dependent insulinotropic
peptide secretion by slowing glucose absorption in the intestine.45 Moreover, chlorogenic acid increases muscle glucose uptake in mice with diabetes.46 Finally, grapefruits contain high amounts of naringin.12
In rats, naringin inhibits dipeptidyl peptidase 4 similarly to
sitagliptin, a dipeptidyl peptidase 4 inhibitor used for the treatment
of diabetes.47 Inhibition of dipeptidyl peptidase 4 increases glucagon-like peptide 1, which subsequently leads to improved glucose tolerance.48
In contrast to these specific fruits mentioned above, cantaloupe was
associated with an increased risk of type 2 diabetes in the current
analysis. Melons have lower levels of phytochemicals than the
aforementioned fruits.12
None the less, little evidence exists regarding the effects of melons
on glucose metabolism. Although other fruits may also be beneficial for
glucose metabolism, significant associations between other specific
fruits and risk of type 2 diabetes were not found in the current and
previous investigations.5 6 7 8
The
glycemic index/glycemic load values of fruits did not seem to be the
factor that determined their association with type 2 diabetes in the
current study, although in a clinical trial, increased consumption of
low glycemic index fruits improved glycemic control among people with
diabetes.33 In recent meta-analyses, a higher dietary glycemic index/glycemic load was associated with a greater risk of type 2 diabetes.49 50
In the Nurses’ Health Study and Health Professionals Follow-up Study,
the associations between dietary glycemic index and risk of type 2
diabetes were positive, although the associations for dietary glycemic
load were not significant.51 52 53
None the less, the contribution of total fruit consumption to dietary
glycemic load was rather small (about 10%) in these populations. Of
individual fruits, the top three contributors to dietary glycemic load
were bananas (3-4%), apples (2%), and grapes (1%). In contrast, the
relatively high glycemic load values of fruit juices13 along with reduced levels of beneficial nutrients through juicing processes11 12
(for example, the glycemic load values per serving are 6.2 for raw
oranges and 13.4 for orange juice, and fibre levels per serving are 3.1 g
and 0.5 g, respectively) may explain the positive associations between
fruit juice consumption and risk of type 2 diabetes. Moreover, the
difference in the viscosity of foods is also an important factor
affecting postprandial blood glucose dynamics. Fluids pass through the
stomach to the intestine more rapidly than solids even if nutritional
content is similar.54 For example, fruit juices lead to more rapid and larger changes in serum levels of glucose and insulin than whole fruits.55 56
Although these mechanisms may potentially explain the diverse
associations for individual fruits, further research is apparently
needed to confirm our findings on specific fruits in relation to type 2
diabetes and to further elucidate underlying mechanisms.
Strengths and limitations of this study
The
present study has several limitations. Firstly, measurement errors were
inevitable in the estimates of fruit consumption, especially for
individual fruits with lower consumption levels.17 18
Adjustment for energy intake and use of cumulatively averaged intake
levels can reduce the magnitude of measurement errors to some extent.26
Generally, random errors in exposure assessments attenuate true
associations toward the null. Secondly, the possibility of false
positive findings may exist because we examined the associations of
multiple fruits in the current investigation without adjusting for
multiple comparisons based on a priori hypotheses. Meanwhile, most
associations were consistent across three cohorts, and the associations
for blueberries, grapes, and apples remained statistically significant
even after applying the Bonferroni correction, a conservative method
correcting for multiple comparisons. Thirdly, in our food frequency
questionnaires, intakes of some individual fruits (apples and pears;
peaches, plums, and apricots) were combined because these fruits have
similar nutrient profiles. Therefore we could not determine whether the
associations for these combined fruits can be ascribed to a specific
individual fruit. Fourthly, we cannot exclude the possibility of recall
bias in the assessments of diet based on the food frequency
questionnaires. However, the prospective study design and exclusion of
participants with chronic diseases at baseline should minimize such
bias. Fifthly, although in the multivariable analysis we considered a
multitude of lifestyle and dietary factors, including other individual
fruits, residual or unmeasured confounding may still exist. Finally, our
study populations primarily consist of health professionals with
European ancestry. Thus our findings may not be generalized to other
populations.
Conclusions
Our
findings suggest that there is significant heterogeneity in the
associations between individual fruits and risk of type 2 diabetes.
Greater consumption of specific whole fruits, particularly blueberries,
grapes, and apples, was significantly associated with a lower risk of
type 2 diabetes, whereas greater fruit juice consumption was associated
with a higher risk. The differences in the associations between
individual fruits were not accounted for by variation in the glycemic
index/glycemic load values of individual fruits. Overall, these results
support recommendations on increasing consumption of a variety of whole
fruits, especially blueberries, grapes, and apples, as a measure for
diabetes prevention.
What is already known on this topic
- Total fruit consumption is not consistently associated with a lower risk of type 2 diabetes
- The possible heterogeneity among individual fruits regarding the associations with risk of type 2 diabetes has not been examined
What this study adds
- The associations with risk of type 2 diabetes are different among individual fruits
- Greater consumption of specific whole fruits, particularly blueberries, grapes, and apples, is significantly associated with a lower risk of type 2 diabetes, whereas increased consumption of fruit juices has the opposite association
- In addition, the associations of individual fruits are not determined by their glycemic index or glycemic load values
Notes
Cite this as: BMJ 2013;347:f5001
Footnotes
- Contributors: RMvD, FBH, and WC obtained funding from the National Institutes of Health. IM, RMvD, and QS designed this study. RMvD, QS, JEM, FBH, and WCW were involved in data collection. IM, FI, QS, and RMvD provided statistical expertise. IM analysed the data and wrote the first draft of the manuscript. All authors contributed to the interpretation of the results and critical revision of the manuscript for important intellectual content and approved the final version of the manuscript. IM and QS are the guarantors of this investigation.
- Funding: This study was funded by research grants CA87969, CA176726, CA55075, CA50385, CA167552, DK58845, and DK082486 from the National Institutes of Health. Dr. Sun was supported by a career development award R00HL098459 from the National Heart, Lung, and Blood Institute. The funding sources had no role in study design; in the collection, analysis, and interpretation of data; in the writing of the report; or in the decision to submit the article for publication. The authors are not affiliated with the funding institutions.
- Competing interests: All authors have completed the ICMJE uniform disclosure form at www.icmje.org/coi_disclosure.pdf (available on request from the corresponding author) and declare: no support from any organisation for the submitted work; no financial relationships with any organisations that might have an interest in the submitted work in the previous three years, no other relationships or activities that could appear to have influenced the submitted work.
- Ethical approval: The study protocol was approved by the institutional review boards of the Brigham and Women’s Hospital, and the Harvard School of Public Health. The completion of the self administered questionnaire was considered to imply informed consent.
- Data sharing: No additional data available.
This
is an Open Access article distributed in accordance with the Creative
Commons Attribution Non Commercial (CC BY-NC 3.0) license, which permits
others to distribute, remix, adapt, build upon this work
non-commercially, and license their derivative works on different terms,
provided the original work is properly cited and the use is
non-commercial. See: http://creativecommons.org/licenses/by-nc/3.0/.
