Association between added sugar intake and mortality is nonlinear and dependent on sugar source in 2 Swedish population-based prospective cohorts

Abstract

Background:
Although sugar consumption has been associated with several risk factors for cardiometabolic diseases, evidence for harmful long-term effects is lacking. In addition, most studies have focused on sugar-sweetened beverages (SSBs), not sugar per se.

Objective:
The aim of this study was to examine the associations between added and free sugar intake, intake of different sugar sources, and mortality risk.

Methods:
Two prospective population-based cohorts were examined: the Malmö Diet and Cancer Study (MDCS; n = 24,272), which collected dietary data by combining a food diary, interview, and food-frequency questionnaire (FFQ), and the Northern Swedish Health and Disease Study (NSHDS; n = 24,475), which assessed diet with an FFQ. Sugar intakes defined as both added and free sugar and different sugar sources were examined. The associations with mortality were examined using a multivariable Cox proportional hazards regression.

Results:
Higher sugar consumption was associated with a less favorable lifestyle in general. The lowest mortality risk was found with added sugar intakes between 7.5% and 10% of energy (E%) intake in both cohorts. Intakes >20E% were associated with a 30% increased mortality risk, but increased risks were also found at intakes <5E% [23% in the MDCS and 9% (nonsignificant) in the NSHDS]. Similar U-shaped associations were found for both cardiovascular and cancer mortality in the MDCS. By separately analyzing the different sugar sources, the intake of SSBs was positively associated with mortality, whereas the intake of treats was inversely associated.

Conclusions:
Our findings indicate that a high sugar intake is associated with an increased mortality risk. However, the risk is also increased among low sugar consumers, although they have a more favorable lifestyle in general. In addition, the associations are dependent on the type of sugar source.

Full text

Association between added sugar intake and mortality is nonlinear and
dependent on sugar source in 2 Swedish population–based
prospective cohorts
Stina Ramne,1Joana Alves Dias,1Esther González-Padilla,1Kjell Olsson,1Bernt Lindahl,4Gunnar Engström,2
Ulrika Ericson,3Ingegerd Johansson,5and Emily Sonestedt1
1Nutritional Epidemiology, 2Cardiovascular Epidemiology, and 3Diabetes and Cardiovascular Disease–Genetic Epidemiology, Department of Clinical Sciences
Malmö, Lund University, Lund, Sweden; and 4Department of Public Health and Clinical Medicine, Occupational and Environmental Medicine, and
5Departments of Odontology and Nutritional Research, Umeå University, Umeå, Sweden
ABSTRACT
Background: Although sugar consumption has been associated
with several risk factors for cardiometabolic diseases, evidence for
harmful long-term effects is lacking. In addition, most studies have
focused on sugar-sweetened beverages (SSBs), not sugar per se.
Objective: The aim of this study was to examine the associations
between added and free sugar intake, intake of different sugar
sources, and mortality risk.
Methods: Two prospective population-based cohorts were exam-
ined: the Malmö Diet and Cancer Study (MDCS; n=24,272),
which collected dietary data by combining a food diary, interview,
and food-frequency questionnaire (FFQ), and the Northern Swedish
Health and Disease Study (NSHDS; n=24,475), which assessed
diet with an FFQ. Sugar intakes dened as both added and free
sugar and different sugar sources were examined. The associations
with mortality were examined using a multivariable Cox proportional
hazards regression.
Results: Higher sugar consumption was associated with a less
favorable lifestyle in general. The lowest mortality risk was found
with added sugar intakes between 7.5% and 10% of energy (E%)
intake in both cohorts. Intakes >20E% were associated with a
30% increased mortality risk, but increased risks were also found
at intakes <5E% [23% in the MDCS and 9% (nonsignicant) in
the NSHDS]. Similar U-shaped associations were found for both
cardiovascular and cancer mortality in the MDCS. By separately
analyzing the different sugar sources, the intake of SSBs was
positively associated with mortality, whereas the intake of treats was
inversely associated.
Conclusions: Our ndings indicate that a high sugar intake is
associated with an increased mortality risk. However, the risk is
also increased among low sugar consumers, although they have a
more favorable lifestyle in general. In addition, the associations
are dependent on the type of sugar source. Am J Clin Nutr
2019;109:411–423.
Keywords: added sugar, free sugar, sugar-sweetened beverages,
mortality, cardiometabolic risk marker, nutritional epidemiology
Introduction
Although sugar consumption has been associated with several
risk factors for cardiometabolic diseases (1), evidence for
harmful long-term effects of high sugar consumption is lacking.
In addition, most studies have focused on sugar-sweetened
beverages (SSBs), not sugar per se. For example, although a
high consumption of SSBs has been associated with an increased
risk of type 2 diabetes and cardiovascular disease (CVD) (2,
3), the evidence for an association with the total intake of
sugars, or added sugar specically, is inconclusive (3–5). A
potential link between sugar and cancer is also much debated.
In the recent third expert report by the World Cancer Research
Fund, the authors concluded that there is limited/suggestive
evidence for an association between fructose-containing foods
and pancreatic cancer and probable evidence between glycemic
load and endometrial cancer. However, they pinpoint SSBs with
regard to overweight and obesity, which are important risk factors
for 12 different types of cancer (6).
Few studies have examined the potential impact of high sugar
consumption on mortality. In the NIH-AARP Diet and Health
Supported by the Swedish Research Council, the Swedish Society for
Medical Research, the Albert Påhlsson Foundation, the Crafoord Foundation,
and the Swedish Heart and Lung Foundation.
Supplemental Figure 1 and Supplemental Tables 1–4 are available from the
“Supplementary data” link in the online posting of the article and from the
same link in the online table of contents at https://academic.oup.com/ajcn/.
Address correspondence to SR (e-mail: stina.ramne@med.lu.se).
Abbreviations used: apo, apolipoprotein; BP, blood pressure; CVD,
cardiovascular disease; E%, percentage of energy; FFQ, food-frequency
questionnaire; ICD, International Classication of Diseases; MDCS, Malmö
Diet and Cancer Study; NSHDS, Northern Swedish Health and Disease
Study; PAL, physical activity level; PY, person-year; SSB, sugar-sweetened
beverage.
Received June 17, 2018. Accepted for publication September 7, 2018.
First published online December 26, 2018; doi: https://doi.org/10.1093/
ajcn/nqy268.
Am J Clin Nutr 2019;109:411–423. Printed in USA. ©2018 American Society for Nutrition. All rights reserved. 411
Downloaded from https://academic.oup.com/ajcn/article/109/2/411/5261249 by guest on 04 January 2021

412 Ramne et al.
Study, many different sugar exposures and mortality outcomes
were examined. For added sugar, the highest quintile [>21% of
energy (E%)] was not associated with an increased mortality risk
compared with the lowest quintile (<5E%). Only total fructose
and fructose from beverages were associated with linearly
increased risk (7). However, in the NHANES, consumption of
>25E% from added sugar was associated with a 2.75-fold greater
risk of CVD mortality compared with consumption of <10E%
(8). Therefore, further studies are necessary to ascertain the role
of sugar in mortality.
The denition of sugar intake and the recommended upper
limit of sugar intake are not unanimous globally. According to
the Nordic Nutrition Recommendations, energy from added sugar
should not exceed 10E%, where added sugars include “sucrose,
fructose, glucose, starch hydrolysates (glucose syrup and high-
fructose syrup), and other isolated sugar preparations used as
such or added during food preparation and manufacturing” (9).
The same upper limit of 10E% has been established in the Dietary
Guidelines for Americans 2015–2020 (10), whereas the Institute
of Medicine suggests an upper limit of 25E% of added sugar (11).
Meanwhile, the WHO recommends a maximum free sugar intake
of 10E% but encourages further reductions to <5E%, which is
supported by Public Health England (12). Free sugar includes the
added sugars mentioned previously in addition to sugars naturally
occurring in fruit juices (13).
Due to inconsistent and scarce evidence for the effect of
added sugar intake on health and disease, and the consequent
difculties in making unanimous dietary recommendations, more
studies investigating the long-term effects of added sugar intake
are needed. Therefore, the aims of this study were to examine
the consumption of added sugar, free sugar, and different sugar
sources and its associations with all-cause (primary aim), cardio-
vascular, and cancer mortality in 2 different Swedish cohorts.
Methods
Study design and subjects
Two Swedish population-based prospective cohorts were
investigated in this study: the Malmö Diet and Cancer Study
(MDCS) and the Northern Swedish Health and Disease Study
(NSHDS).
MDCS.
The MDCS was conducted in the city of Malmö in the south
of Sweden. Between 1991 and 1996, all men born between
1923 and 1945 and women born between 1923 and 1950 were
recruited for baseline examinations. Recruitment was done either
passively by placing invitations and information about the study
in public areas or actively by sending personal invitations to
eligible individuals. In total, 74,318 individuals in Malmö were
invited and 30,447 individuals participated; complete data were
collected for 28,098 individuals. The only exclusion criteria for
participation in the MDCS were inadequate Swedish language
skills and mental disability. In this study, those with a history
of CVD, diabetes, or cancer at baseline, as assessed from the
medical registers, were excluded, resulting in a total of 24,272
participants (Supplemental Figure 1). The MDCS was given
ethical approval from the Ethics Committee at Lund University,
and all participants signed written informed consent before
participating (14–16).
NSHDS.
The NSHDS comprises several population-based cohorts,
including the Västerbotten Intervention Program, in which all
residents in the county of Västerbotten were invited to their
health care center to participate in a health survey when they
turned 40, 50, or 60 y old. The data collection in the NSHDS
started in 1985 and is still ongoing. In January 2014, a total of
100,256 individuals had participated. Here, only data collected in
the Västerbotten Intervention Program from 1991 to 1996 were
included due to a change in the diet questionnaire made in 1996,
making it no longer possible to distinguish fruit juices from SSBs,
and hence calculating added sugar, resulting in 36,826 eligible
participants. After excluding those with a self-reported history
of CVD or diabetes, in total, 24,475 participants formed the
study sample for this study (Supplemental Figure 1). Exclusion
of subjects with a history of cancer was not possible because
this information was not available from the questionnaire. Ethical
approval was given by the Regional Ethical Review Board of
Northern Sweden, and all participants provided written informed
consent before participation.
Assessment of dietary data
MDCS.
Dietary intake was assessed using a modied diet history
method consisting of a 7-d food diary covering all cooked
meals, cold beverages, and dietary supplements; a 168-item
food-frequency questionnaire (FFQ) covering the noncooked
meals for the past 12 mo; and a 1-h interview to discuss food
choices, cooking methods, and portion sizes (using a booklet of
photographs of a variety of foods in different portion sizes) of
the reported foods in the food diary. In 1994, the interview was
shortened to 45 min (17). In the FFQ, subjects estimated their
intake frequencies of the different food items and their usual
portion sizes with the help of photographs. These dietary data
were converted into daily nutrient intake data using the MDCS
food and nutrition database, which originated from the Swedish
National Food Agency. The modied diet history method has
been validated against an 18-d weighed food record, with a
relatively strong sucrose intake correlation showing Pearson’s
correlation coefcients of 0.74 for women and 0.60 for men (18).
NSHDS.
An 84-item self-administered FFQ was used to assess dietary
intake. Intakes of various food items and food groups were
reported on a 9-level scale of intake frequencies, ranging from
never to ≥4 times/d. For staple foods, such as potatoes, rice, pasta,
meat, sh, and vegetables, portion sizes were estimated with the
support of 4 photos showing different amounts of food on a
plate. For other foods, xed portion sizes or age- and sex-specic
portion sizes were used. Estimated daily energy and nutrient
intakes were calculated from the intake frequencies, weighted for
portion sizes, using the food database from the Swedish National
Food Agency. The FFQ was validated against ten 24-h recalls,
Downloaded from https://academic.oup.com/ajcn/article/109/2/411/5261249 by guest on 04 January 2021

Added sugar consumption and mortality 413
resulting in correlation coefcients of 0.65 and 0.37 for total
sucrose intake and 0.69 and 0.62 for candy intake in men and
women, respectively (19).
The added sugar intake was estimated for each individual on
the basis of his or her total intake of sucrose and monosaccharides
(glucose and fructose) from their whole diet. Then, the amounts
of sucrose and monosaccharides naturally occurring in fruit
and vegetables and their juices were subtracted using each
individual’s intake of different fruit and vegetable groups and
the average sugar content of each of these groups. The most
common fruit, vegetables, and juices in each group were used
to estimate the average sugar content. This calculation resulted
in an estimation of all added sugars, including honey and
syrups. Free sugar was estimated by the same calculation
as for added sugar but without subtracting the sucrose and
monosaccharides in fruit juices. These calculations assume that
naturally occurring monosaccharides and sucrose are present
only in fruit and vegetables. The small amounts present in cereal
products are considered negligible. Because sucrose is the only
disaccharide that was addressed in this calculation, naturally
occurring lactose, such as in yogurt, does not contribute to these
sugar measurements, although the added sucrose in avored
yogurt does. Then, the added and free sugar estimates were
converted into percentages of energy intake and subsequently
divided into 6 intake categories: <5E%, 5E% to <7.5E%,
7.5E% to <10E%, 10E% to <15E%, 15E% to <20E%, and
≥20E%. Nonalcoholic energy intake was used in this case
to address the diet quality and to facilitate the comparison
to the Nordic Nutrition Recommendations, which also uses
nonalcoholic energy intake for their recommendations (9).
The consumption of different sugar-rich foods and beverages
was also examined as an exposure and was divided into
the following 3 food groups: treats, toppings, and SSBs.
The SSBs were evaluated separately to be able to dis-
tinguish between differences of consuming sugar in solid
or liquid forms, which were previously shown to differen-
tially associate with health outcomes. SSBs were assessed
by adding the intakes of carbonated soft drinks, noncar-
bonated sweetened drinks, and fruit drinks (not pure fruit
juices). In the MDCS, SSB intake was collected in the 7-d food
diary. Sports and energy drinks were not specically addressed
in the dietary assessments; however, some of this consumption
might have been captured within the other sweetened beverage
categories, depending on how the participants interpreted the
questions and food categories. The solid sugar-rich foods
were divided into treats, calculated by combining all pastries,
candies, chocolate, and ice cream, and into toppings, calculated
by combining all table sugar, sugar cubes, honey, jam, and
marmalade (data were not available for other sweet spreads
such as nut and chocolate spreads). We hypothesized that treats
are more likely to be overconsumed than toppings. Treats are
generally more energy dense (often with a high fat content
as well), which is one reason why treats are believed to be
more easily overconsumed. However, toppings generally have a
higher percentage of calories from sugar (with negligible fat and
protein content). Because of these differences in eating behavior,
macronutrient distribution, and energy density, we believed it
important to investigate differences in the health effect of treats,
toppings, and SSBs separately. In the MDCS, the consumed
amounts of different sugar-rich foods and beverages were recoded
from grams per day to servings per week using the average
serving sizes of the different food items as follows: pastries
(including cakes, pies, cookies, buns) =60 g, candies =60 g,
chocolate =60 g, ice cream (including sherbet) =75 g, table
sugar (including syrup) =10 g, jam (including marmalade, apple
sauce, jelly, and honey) =20 g, and SSBs =280 g. These serving
sizes are based on what several common manufacturers of the
different products have considered a serving size and what a
“normal” serving size is considered to be in the food database
from the National Swedish Food Agency. On the other hand,
in the NSHDS, intakes were already reported as servings (i.e.,
number of intake occasions) per day and were therefore only
converted into servings per week by multiplying by 7.
Assessment of covariates
MDCS.
At baseline, a questionnaire was used to assess participants’
medical history, medication, and diet supplementation as well
as social, socioeconomic, demographic, and lifestyle factors,
such as leisure-time physical activity (5 categories of metabolic
equivalent-hours per week based on the time spent engaging in
17 different activities and their metabolic equivalent intensity
factor), smoking status (smokers, former smokers, and never
smokers at baseline), and alcohol habits (quintiles of consumers
with one category of nonconsumers, where high alcohol con-
sumption was dened as the fth quintile). Disease history was
also assessed through the medical registers. Body weight, height,
and blood pressure (BP) were measured, and blood samples
(nonfasting) were collected. A sample of 6301 participants were
invited to participate in the MDCS cardiovascular subcohort,
in which fasting blood samples were collected for assessing
total cholesterol, triglycerides, HDL cholesterol, LDL cholesterol
(calculated using Friedewald’s equation), glucose, and insulin.
Apolipoprotein (apo) B and apoA-I were analyzed from nonfast-
ing blood samples and were therefore available for almost the
entire MDCS. More detailed descriptions of the data collection
have been published elsewhere (14,20).
NSHDS.
At baseline, a questionnaire was used to assess participants’
medical history, medication, and diet supplementation as well as
social, socioeconomic, demographic, and lifestyle factors, such
as leisure-time physical activity, smoking status, and alcohol
habits. Smoking and alcohol were divided into categories similar
to those in the MDCS, but leisure-time physical activity was
monitored on a 5-level scale ranging from mainly resting to
exercising ≥3 times/wk (denition: with an intensity that requires
a change to workout clothes) and regular physical activity was
dened as at least once a week. Body weight, height, and BP were
measured, and fasting blood samples were collected to obtain
total cholesterol, triglyceride, and blood glucose concentrations.
HDL- and LDL-cholesterol concentrations were collected only
for a few participants during this time period and could therefore
not be examined in this cohort. More detailed descriptions of the
data collection have been published elsewhere (21).
Downloaded from https://academic.oup.com/ajcn/article/109/2/411/5261249 by guest on 04 January 2021

414 Ramne et al.
TABLE 1 Characteristics of the 2 cohorts (MDCS and NSHDS) and their
participants1
MDCS
(n=24,272)
NSHDS
(n=24,475)
Time period of data collection 1991–1996 1991–1996
Dietary assessment method Modied diet history FFQ
Cases of death, n(%) 6309 (26.0) 2881 (11.8)
Wom en , n(%) 14,907 (61.4) 13,145 (53.7)
Smokers, n(%) 6950 (28.6) 6368 (26.0)
University education, n(%) 3591 (14.8) 5517 (22.5)
Age,2y 57.6 (44–73) 48.6 (36–64)
BMI, kg/m225.6 ±3.9325.5 ±3.8
Energy intake, kcal 2288 ±656 1793 ±573
Added sugar, g/d 57.0 ±32 37.3 ±23
Added sugar, E% 10.1 ±4.3 8.2 ±4.0
Free sugar, g/d 61.9 ±33 41.1 ±25
Free sugar, E% 11.0 ±4.5 9.1 ±4.2
Follow-up time, y 19.9 (18.6–21.6)420.1 (18.8–21.7)
Treats, servings/wk 6.16 (3.6–9.5) 5.46 (3.1–9.4)
Toppings, servings/wk 9.18 (4.0–17.3) 5.04 (1.1–14.0)
SSBs, servings/wk 0.25 (0.0–2.4) 1.12 (0.58–3.1)
1E%, percentage of energy; FFQ, food-frequency questionnaire; MDCS, Malmö
Diet and Cancer Study; NSHDS, Northern Swedish Health and Disease Study; SSB,
sugar-sweetened beverage.
2Values are means (ranges).
3Mean ±SD (all such values).
4Median; IQR in parentheses (all such values) unless otherwise specied.
Assessment of mortality
The Swedish National Tax Agency, Statistics in Sweden, and
the National Board of Health and Welfare were used to obtain
the vital status and emigration status until the end of the follow-
up period on 31 December 2014. To identify the cause of death,
the Swedish Cause of Death Register was used. Cardiovascular
death was identied as International Classication of Diseases
(ICD), Ninth Revision (ICD-9), codes 390–459 or ICD, 10th
Revision (ICD-10), code I; and cancer death was identied as
ICD-9 codes 140–239 or ICD-10 code C-D48. Cause-specic
analysis was considered possible only in the MDCS (number of
death cases =6309) because of the low mortality rate in the
NSHDS (number of death cases =2281).
Statistical analysis
All of the statistical analyses were performed by using either
IBM SPSS Statistics (version 24 or 25; IBM Corp.) or Stata/SE
(version 15; StataCorp). The differences in the characteristics of
the 6 intake categories of added sugar were tested using a chi-
square test for categorical variables and using a general linear
model adjusted for age and sex for continuous variables. Skewed
continuous variables are expressed as medians (IQRs) instead of
means ±SDs and were log-transformed in the general linear
model to improve normality. A constant was added for dietary
variables with many zero values: 0.01 in the MDCS and 0.1 in the
NSHDS. Different constants provided the best normality because
the dietary variables are expressed in different units between
the 2 cohorts (grams per day or servings per day). The median
E% in each intake category was used instead of values 1–6 to
test for trends over the different intake categories. For the cross-
sectional analysis of the baseline measured cardiometabolic risk
markers, the means and tests for trends were adjusted for sex,
educational level, leisure-time physical activity level (PAL),
smoking status, and alcohol habits as categorical variables and
age (years), total energy intake (kilocalories), fruit and vegetable
intake, processed meat intake, coffee intake (grams per day
or servings per day), SFA intake (E%), ber density (grams
per 1000 kcal), and BMI (kg/m2) as continuous variables. This
model was considered the fully adjusted model. However, in
the MDCS, season (categorical) and screening date (continuous)
were also adjusted to control for the seasonal variation in dietary
intake and the minor change in the diet assessment method
implemented in 1994. Adjustment for the dietary assessment
method (before and after 1994) instead of screening date was
also evaluated, but because it did not alter the association, the
screening date was chosen to account for the storage time for the
apolipoprotein samples. In the NSHDS, screening year was tested
as a confounder but did not alter the results and was therefore
not included in the model because no methodologic reason was
present. Additional adjustment for the use of antihypertensive
drugs was made when examining systolic and diastolic BP, and
adjustment for the use of lipid-lowering drugs was made for total
cholesterol, triglycerides, HDL cholesterol, LDL cholesterol, and
the apoB-to-apoA-I ratio.
Cox proportional hazards regression was used to examine
the associations between the intake of added sugar, free sugar,
treats, toppings, and SSBs and the risk of all-cause mortality,
cardiovascular mortality, and cancer mortality. The added and
free sugars were investigated as nutrient densities in E%. Years
of follow-up was used as the time variable. Age was also tested
as a time variable; however, it did not alter the association.
The basic model was adjusted for age, sex, and total energy
intake, and the fully adjusted model was adjusted as described
above. The dietary covariates included in the model (fruit and
vegetables, processed meat, coffee, SFAs, and ber) were chosen
because they had previously been associated with mortality or
CVD risk and/or sugar intake. The associations were tested
for proportional-hazard assumptions, and signicant covariates
were stratied to achieve proportionality. Thus, the fully adjusted
model was stratied for age groups, sex, and alcohol habits in
the NSHDS and additionally for smoking status in the MDCS.
Additional analyses of the relation between added and free sugars
and all-cause mortality were performed using the residual method
instead of the nutrient-density method to adjust for energy intake.
The intake category of added and free sugars with the lowest
risk of all-cause mortality was used as the reference category
when a clear U-shape could be observed in both cohorts.
Restricted cubic splines with 4 knots (3 knots for SSBs in the
MDCS), placed according to Harrell’s default percentiles in Stata
(22), were used to plot the fully adjusted association between
sugar intake and HRs for all-cause mortality. For treats, toppings,
and SSBs, intake categories were determined by examining the
curves using restricted cubic splines. Treats were categorized as
≤2,>2–5, >5–8, >8–14, and >14 servings/wk; toppings were
categorized as ≤2, >2–7, >7–14, >14–28, and >28 servings/wk;
and SSBs were categorized as ≤1, >1–3, >3–5, >5–8, and >8
servings/wk. The overall joint effects were tested over the intake
categories, as well as the linearity of the associations over the
intake categories, using the median E% or servings per week in
each category.
Downloaded from https://academic.oup.com/ajcn/article/109/2/411/5261249 by guest on 04 January 2021

Added sugar consumption and mortality 415
TABLE 2 Baseline characteristics according to added sugar intake in 6 intake E% categories in the MDCS and NSHDS1
Added sugar intake categories
<5E% 5E% to <7.5E%
7.5E% to
<10E%
10E% to
<15E%
15E% to
<20E% ≥20E% P2
MDCS
Median added sugar intake, E% 4.0 6.4 8.8 11.9 16.6 22.3
n2236 4725 6264 8181 2233 644
Wom en , n(%) 1420 (63.5) 2926 (61.9) 3877 (61.9) 4997 (61.1) 1318 (59.0) 369 (58.3) 0.018
Smokers, n(%) 772 (34.5) 1351 (28.6) 1661 (26.5) 2180 (26.6) 717 (32.1) 269 (42.5) <0.001
High alcohol consumers, n(%) 595 (26.6) 1150 (24.3) 1237 (19.7) 1343 (16.4) 302 (13.5) 79 (12.5) <0.001
University education, n(%) 456 (20.4) 788 (16.7) 992 (15.8) 1040 (12.7) 243 (10.9) 72 (11.4) <0.001
Underreporters, n(%) 640 (28.6) 947 (20.0) 923 (14.7) 933 (11.4) 236 (10.6) 57 (9.0) <0.001
Overreporters, n(%) 31 (1.4) 96 (2.0) 154 (2.5) 292 (3.6) 125 (5.6) 47 (7.4) <0.001
Dietary changers, n(%) 615 (27.5) 1052 (22.3) 1251 (20.0) 1625 (19.9) 455 (20.4) 169 (26.7) <0.001
Age, y 55.8 ±7.0356.7 ±7.4 57.7 ±7.5 58.2 ±7.7 57.5 ±7.7 57.9 ±7.3 <0.0014
BMI, kg/m226.2 ±4.3 25.8 ±3.9 25.6 ±3.8 25.4 ±3.8 25.1 ±3.7 25.2 ±4.2 <0.0014
Energy intake, kcal 2030 ±639 2185 ±634 2286 ±632 2367 ±654 2424 ±667 2466 ±719 <0.0014
Physical activity, MET-h/wk 25.3 (13.8–40.4)526.0 (15.4–39.9) 26.9 (16.1–41.2) 26.6 (15.6–42.6) 26.7 (14.9–41.9) 22.7 (11.9–39.9) 0.2186
Treats, servings/wk 2.38 (1.1–3.9) 4.49 (2.7–6.6) 6.91 (4.1–9.0) 7.86 (5.1–11.3) 8.69 (5.3–13.5) 9.22 (4.7–15.1) <0.0016
Toppings, servings/wk 1.78 (0.23–3.7) 5.01 (2.4–8.5) 8.72 (4.8–14.0) 14.3 (8.1–22.6) 21.5 (11.4–35.4) 28.1 (11.6–47.0) <0.0016
SSBs, servings/wk 0.00 (0.0–0.0) 0.00 (0.0–0.7) 0.07 (0.0–1.5) 1.18 (0.0–3.6) 3.57 (0.5–7.8) 7.86 (2.3–15.3) <0.0016
NSHDS
Median added sugar intake, E% 4.0 6.3 8.6 11.7 16.5 22.6
n4886 7318 5703 5141 1097 330
Wom en , n(%) 3061 (62.6) 4260 (58.2) 2990 (52.4) 2263 (44.0) 448 (40.8) 123 (37.3) <0.001
Smokers, n(%) 1509 (30.9) 1756 (24.0) 1313 (23.0) 1360 (26.5) 338 (30.8) 92 (27.9) <0.001
High alcohol consumers, n(%) 914 (18.7) 1437 (19.6) 1049 (18.4) 889 (17.3) 162 (14.8) 39 (11.5) <0.001
University education, n(%) 1164 (23.8) 1786 (24.4) 1351 (23.7) 996 (19.4) 178 (16.2) 42 (12.7) <0.001
Underreporters, n(%) 1089 (22.3) 1122 (15.3) 693 (12.2) 574 (11.2) 143 (13.0) 49 (14.8) <0.001
Overreporters, n(%) 81 (1.7) 171 (2.3) 157 (2.8) 165 (3.2) 35 (3.2) 2 (0.6) <0.001
Regular physical activity, n(%) 1501 (30.7) 2206 (30.1) 1662 (29.1) 1352 (26.3) 209 (19.1) 60 (18.2) <0.001
Age, y 48.1 ±7.6 47.9 ±7.6 48.4 ±7.7 49.6 ±7.9 51.2 ±8.2 51.0 ±7.4 <0.0014
BMI, kg/m225.8 ±3.9 25.4 ±3.8 25.3 ±3.7 25.3 ±3.6 25.5 ±3.9 25.7 ±3.9 <0.0014
Energy, kcal 1631 ±538 1759 ±560 1850 ±563 1904 ±589 1902 ±597 1864 (578) <0.0014
Treats, servings/wk 2.66 (1.70–4.50) 5.04 (3.08–8.54) 7.00 (4.20–10.6) 8.54 (4.62–13.3) 9.1 (5.04–15.4) 8.12 (3.66–12.6) <0.0016
Toppings, servings/wk 0.98 (0.56–1.54) 3.08 (1.12–7.56) 7.56 (3.50–18.1) 17.5 (7.42–22.4) 18.5 (7.98–24.5) 18.1 (7.56–24.5) <0.0016
SSBs, servings/wk 0.58 (0.04–1.12) 1.12 (0.58–1.54) 1.54 (1.00–3.08) 3.08 (1.12–7.00) 7.42 (3.08–9.80) 18.1 (14.0–20.0) <0.0016
1E%, percentage of energy; GLM, general linear model; MDCS, Malmö Diet and Cancer Study; MET-h, metabolic equivalent task hours; NSHDS, Northern Swedish Health
and Disease Study; SSB, sugar-sweetened beverage.
2Pvalues were determined by using a chi-square test for categorical variables.
3Mean ±SD (all such values).
4P-trend determined by using a GLM adjusted for age and sex using the median E% within each category of added sugar intake.
5Median; IQR in parentheses (all such values).
6P-trend determined by using a GLM adjusted for age and sex using the median E% within each category of added sugar intake with a log10-transformed dependent variable
[log10 (x+0.01) in the MDCS and log10 (x+0.1) in the NSHDS] for normality.
Potential effect modication was evaluated for sex, BMI, and
smoking and alcohol habits by introducing a cross-product be-
tween the variable and added sugar intake on all-cause mortality
risk. Because the association between added sugar intake and all-
cause mortality was found to be U-shaped and testing interaction
requires linearity, intakes from <5E% to 7.5E% to <10E% and
10E% to <15E% to ≥20E% were tested separately. Sensitivity
analysis was performed excluding energy misreporters. In the
MDCS, energy misreporters were dened according to Black
and Goldberg using individual values for PAL (23). This method
has been described more extensively in a previous publication
(24). On the other hand, in the NSHDS, because no PAL values
were available, subjects with the lowest 15% and top 2.5%
values of food intake level (i.e., energy intake/basal metabolic
rate) were excluded. These percentages were chosen to match
the proportions of under- and overreporters in the MDCS. An
additional sensitivity analysis was also performed for those
who died ≤2 y after baseline examinations in both cohorts
and those who had reported that they recently changed their
dietary habits at baseline in the MDCS (data unavailable in the
NSHDS).
Results
Cohort characteristics
The 2 examined cohorts are similar to a great extent (Tab l e 1).
Both cohorts were collected between 1991 and 1996 and followed
until 31 December 2014, resulting in a median follow-up time
of ∼20 y. However, the population in the NSHDS is younger,
with a mean age of 48.6 y (range: 36–64 y), whereas in the
MDCS, the mean age was 57.6 y (range: 44–73 y) at baseline.
Hence, a smaller proportion of the population in the NSHDS
has died [2281 cases (12%)] than in the MDCS [6309 cases
(26%)]. The percentage of female participants was 61% in the
MDCS and 54% in the NSHDS. The reported energy intakes,
added and free sugar intakes, and intakes of different sugar-
rich foods and beverages were, on average, lower from the FFQ
Downloaded from https://academic.oup.com/ajcn/article/109/2/411/5261249 by guest on 04 January 2021

416 Ramne et al.
TABLE 3 Cross-sectionally analyzed fully adjusted means (SEs) of cardiometabolic risk markers according to added sugar intake in 6 intake energy
percentage categories in the MDCS and NSHDS1
Added sugar intake categories
n<5E%
5E% to
<7.5E%
7.5E% to
<10E%
10E% to
<15E%
15E% to
<20E% ≥20E% βP-trend
MDCS
Median added sugar intake, E% — 4.0 6.4 8.8 11.9 16.6 22.3
Fasting plasma glucose,2mmol/L 4651 5.67 (0.04) 5.68 (0.03) 5.62 (0.03) 5.65 (0.02) 5.61 (0.04) 5.70 (0.08) −0.002 0.519
Fasting plasma insulin,2pmol/L 4500 47.5 (2.40) 52.2 (1.71) 50.3 (1.52) 51.4 (1.38) 55.1 (2.45) 54.6 (4.75) 0.001∗0.277∗
HOMA-IR24300 1.66 (0.06) 1.70 (0.04) 1.66 (0.04) 1.71 (0.04) 1.77 (0.06) 1.84 (0.12) 0.000∗0.634∗
apoB:apoA-I323,709 0.724 (0.006) 0.730 (0.005) 0.739 (0.005) 0.742 (0.005) 0.758 (0.006) 0.756 (0.009) 0.002 <0.001
Total cholesterol,3mmol/L 4656 6.31 (0.08) 6.34 (0.07) 6.40 (0.07) 6.29 (0.06) 6.32 (0.08) 6.22 (0.13) −0.006 0.218
Triglycerides,3mmol/L 4656 1.48 (0.05) 1.49 (0.05) 1.48 (0.04) 1.49 (0.04) 1.59 (0.05) 1.50 (0.09) 0.005 0.113
HDL cholesterol,3mmol/L 4615 1.40 (0.02) 1.33 (0.02) 1.35 (0.02) 1.33 (0.02) 1.27 (0.02) 1.28 (0.04) −0.007 <0.001
LDL cholesterol,3mmol/L 4558 4.25 (0.07) 4.33 (0.06) 4.37 (0.06) 4.29 (0.06) 4.36 (0.07) 4.28 (0.12) 0.000 0.944
Systolic BP,4mm Hg 24,249 143.2 (0.41) 143.2 (0.30) 142.5 (0.27) 142.9 (0.26) 143.2 (0.41) 141.7 (0.73) 0.027 0.408
Diastolic BP,4mm Hg 24,247 87.2 (0.22) 87.4 (0.16) 86.9 (0.15) 87.2 (0.14) 87.4 (0.22) 87.2 (0.38) 0.006 0.728
NSHDS
Median added sugar intake, E% 4.0 6.3 8.6 11.7 16.5 22.6
Fasting blood glucose,5mmol/L 24,382 5.33 (0.01) 5.31 (0.01) 5.31 (0.01) 5.32 (0.01) 5.31 (0.03) 5.29 (0.04) −0.001 0.515
Total cholesterol,3mmol/L 24,377 6.36 (0.14) 6.09 (0.12) 6.11 (0.13) 6.06 (0.13) 5.80 (0.23) 6.12 (0.52) −0.028 0.055
Triglycerides,3mmol/L 19,824 1.83 (0.12) 1.58 (0.10) 1.63 (0.12) 1.56 (0.11) 1.46 (0.19) 1.45 (0.41) −0.022 0.075
Systolic BP,4mm Hg 24,095 138.8 (0.95) 137.6 (0.83) 136.7 (0.92) 137.4 (0.93) 138.0 (1.6) 142.6 (3.0) −0.008 0.939
Diastolic BP,4mm Hg 24,092 85.7 (0.55) 84.5 (0.48) 84.6 (0.53) 84.8 (0.53) 85.4 (0.94) 89.6 (1.7) 0.040 0.485
1Trends were determined by using a GLM with the use of the median E% in each category of added sugar intake. ∗Determined using a log10-transformed variable. apo,
apolipoprotein; BP, blood pressure; E%, percentage of energy; GLM, general linear model; MDCS, Malmö Diet and Cancer Study; NSHDS, Northern Swedish Health and Disease
Study.
2MDCS fully adjusted model: adjusted for age, sex, season, screening date, energy intake, educational level, physical activity, smoking status, alcohol habits, dietary habits
(fruit and vegetables, processed meat, coffee, SFAs, and ber density), and BMI.
3Additional adjustment for the usage of lipid-lowering drugs.
4Additional adjustment for the usage of antihypertensive medication.
5NSHDS fully adjusted model: adjusted for age, sex, energy intake, educational level, leisure-time physical activity, smoking status, alcohol habits, dietary habits (fruit and
vegetables, processed meat, coffee, SFAs, and ber density), and BMI.
used in the NSHDS than those obtained using the modied
diet history method in the MDCS. The average added sugar
intakes and free sugar intakes were 10.1E% and 11.0E% in
the MDCS and 8.2E% and 9.1E% in the NSHDS, respectively,
indicating an average of 1E% coming from fruit juices in both
cohorts.
Baseline characteristics according to added sugar
intake
Among MDCS participants, higher added sugar intake was
associated with older age, higher energy intake, and higher
intakes of all different sugar-rich foods and beverages but also
with a lower BMI. Having a higher intake of added sugar was
also related to higher proportions of male participants and current
smokers and with lower proportions of high alcohol consumers
and university graduates (Tab l e 2). The characteristics of the
participants in the NSHDS according to added sugar intakes
show similar trends as in the MDCS, with the addition of
lower proportions of regular physical activity observed with
higher added sugar intakes (Tab le 2). A higher intake of added
sugar also tended to be associated with less favorable dietary
habits in both examined cohorts (Supplemental Table 1).
After full adjustment, a signicant positive linear trend for the
apoB-to-apoA-I ratio and a signicant negative linear trend for
HDL cholesterol were observed with increasing added sugar
intake in the MDCS. Data for these particular risk markers
were not available in the NSHDS, in which no signicant
trends were found for other cardiometabolic risk markers
(Tabl e 3 ).
Association between sugar intake and mortality
Per 1000 person-years (PYs), 13 deaths occurred in the lowest
added sugar intake category (<5E%) in the MDCS (calculated
from 552 cases for a total of 41,638 PYs), whereas 20 deaths
occurred in the highest category (>20E%). In the NSHDS, these
crude numbers were 6 and 11 deaths/1000 PYs, respectively.
As shown in Tab l e 4 and Figure 1, for both added sugar
and free sugar, the intake category of 7.5E% to <10E% was
associated with the lowest risk of all-cause mortality and was
thus used as a reference in the Cox regression model. Added
sugar intakes of both >20E% and <5E% were signicantly
associated with a higher risk of all-cause mortality in the
MDCS, with HRs (95% CIs) of 1.30 (1.12, 1.51) and 1.23
(1.11, 1.35), respectively, after full adjustment. Adjustment for
BMI barely altered the association. No considerable difference
was found for the association with free sugar, with HRs (95%
CIs) of 1.26 (1.10, 1.44) and 1.22 (1.09, 1.37), respectively.
A similar U-shaped association was observed in the NSHDS;
however, the associations were weaker within the lower intakes
(with signicantly increased risk only after basic adjustment).
A signicantly increased risk of 31% was observed only in the
highest intake category of added sugar after full adjustment,
despite the low number of individuals in this group (n=330).
The overall effect was signicant for both added and free sugars
Downloaded from https://academic.oup.com/ajcn/article/109/2/411/5261249 by guest on 04 January 2021

Added sugar consumption and mortality 417
TABLE 4 Associations between the intakes of added and free sugar and all-cause mortality in the MDCS and NSHDS1
MDCS NSHDS
Sugar intake categories Sugar intake categories
All-cause mortality <5E%
5E% to
<7.5E%
7.5E% to
<10E%
10E%to
<15E%
15E% to
<20E% ≥20E% P<5E%
5E% to
<7.5E%
7.5E% to
<10E%
10E% to
<15E%
15E% to
<20E% ≥20E% P
Added sugar
n/Cases/PYs 2236/552/
41,638
4725/1138/
89,302
6264/1547/
119,457
8181/2203/
153,752
2233/645/
41,447
633/224/
11,216
4886/553/
94,150
7318/747/
142,067
5703/625/
111,128
5141/710/
99,452
1097/177/
20,913
330/69/
6235
Basic adjusted 1.29
(1.17, 1.43)
1.08
(1.00, 1.17)
11.05
(0.99, 1.12)
1.14
(1.04, 1.24)
1.62
(1.41, 1.87)
0.033,2
<0.0013
1.13
(1.00, 1.27)
1.03
(0.92, 1.14)
11.08
(0.97, 1.21)
1.05
(0.89, 1.25)
1.49
(1.16, 1.91)
0.284,2
0.0233
Fully adjusted 1.23
(1.11, 1.35)
1.06
(0.98, 1.14)
11.03
(0.96, 1.10)
1.04
(0.95, 1.15)
1.30
(1.12, 1.51)
0.920,2
<0.0013
1.09
(0.97, 1.22)
1.05)
(0.95, 1.17)
11.04
(0.93, 1.16)
0.92
(0.78, 1.10)
1.31
(1.01, 1.70)
0.664,2
0.0053
Free sugar
n/Cases/PYs 1554/403/
28,782
3687/909/
69,668
5606/136/
107,005
9376/2426/
176,540
3154/877/
58,797
895/297/
16,020
3308/395/
63,611
6398/671/
124,180
6265/648/
121,831
6447/829/
124,913
1582/247/
30,313
475/91/
9096
Basic adjusted 1.32
(1.18, 1.47)
1.11
(1.02, 1.20)
11.05
(0.98, 1.12)
1.11
(1.01, 1.20)
1.54
(1.35, 1.74)
0.027,2
<0.0013
1.14
(1.00, 1.30)
1.05
(0.95, 1.18)
11.10
(0.99, 1.22)
1.13
(0.97, 1.31)
1.47
(1.18, 1.84)
0.064,2
0.0113
Fully adjusted 1.22
(1.09, 1.37)
1.05
(0.96, 1.14)
11.03
(0.97, 1.11)
1.04
(0.95, 1.14)
1.26
(1.10, 1.44)
0.511,2
<0.0013
1.07
(0.94, 1.21)
1.06
(0.95, 1.18)
11.06
(0.95, 1.18)
1.02
(0.88, 1.19)
1.29
(1.03, 1.63)
0.487,2
0.3373
1Values are HRs (95% CIs) unless otherwise indicated. Multivariable Cox proportional hazards regression was used to examine the associations. Basic adjusted: adjusted for age, sex, and energy intake (and season and screening date in
the MDCS). Fully adjusted: adjusted for energy intake, age, sex, educational level, leisure-time physical activity, smoking status, alcohol habits, dietary habits (fruit and vegetables, processed meat, coffee, SFAs, and ber density), and BMI
(and season and screening date in the MDCS). E%, percentage of energy; MDCS, Malmö Diet and Cancer Study; NSHDS, Northern Swedish Health and Disease Study; PY, person-year.
2P-trend for intake categories using the median in each category.
3P-overall for intake categories.
in the MDCS but only for added sugar in the NSHDS. In addition,
none of these associations were found to be linear when tested for
linearity across the 6 categories, which was expected because the
associations appeared to be U-shaped. When using the residual
method instead of the nutrient-density method to adjust for
energy intake, the associations were attenuated, probably mostly
due to reduced variability. However, similar trends were found
with an increased risk at low sugar intakes.
Differences between sugar sources
The associations with all-cause mortality differed substantially
depending on the type of sugar-rich food or beverage examined
(Tabl e 5 ,Figure 2). In both cohorts, a positive association
between SSB intake and mortality was observed. The overall
effect and the linear positive trend for the 5 categories of SSB
intake were signicant in the MDCS (P-overall =0.035 and
P-trend =0.009, respectively) in the fully adjusted model.
Increased risk was observed in the highest intake category of
>8 servings/wk, with HRs (95% CIs) of 1.14 (1.03, 1.26) in
the MDCS and 1.10 (0.90, 1.35) in the NSHDS. However, as
visible in the cubic spline, the risk from SSB intake appeared
to be signicant over ∼16 servings/wk in the NSHDS. On the
other hand, an inverse association was found for treats, with
signicantly reduced risk in all intake categories above the lowest
intake category of ≤2 servings/wk. The HRs (95% CIs) for
the highest intake category of >14 servings/wk compared with
≤2 servings/wk were 0.83 (0.74, 0.93) in the MDCS and 0.66
(0.56, 0.78) in the NSHDS. The trend of this inverse association
was signicant over the 5 categories in both the MDCS and
NSHDS (P-trend =0.001 and <0.001, respectively) as well as
the overall effects (P-overall <0.001 and <0.001, respectively).
After studying these different food groups separately, the inverse
association was particularly driven by pastries (cookies, cakes,
pies, and buns; data not shown). For toppings, the highest risk
was observed in the lowest intake category of ≤2 servings/wk in
both cohorts. No signicant risk reductions were observed for any
intake category in the NSHDS, whereas a slight risk reduction
could be observed in the middle intake categories of >5–8
and >8–14 servings/wk in the MDCS. These 3 different sugar
sources were only weakly correlated with each other, although
all signicantly and positively, with Spearman’s correlation
coefcients ranging from 0.13 to 0.19 in the MDCS and 0.23 to
0.25 in the NSHDS.
Cause-specic mortality
The specic causes of death were examined in the MDCS and
showed a signicantly increased risk of cardiovascular mortality
of 40% for added sugar and 31% for free sugar at intakes >20E%,
as well as increased risks of 22% and 24%, respectively, at
intakes <5E% in the fully adjusted model with 7.5E% to <10E%
as a reference (Tabl e 6 ). A similar U-shaped association was
also observed for cancer mortality, but only intakes of added
sugar <5E% showed a signicant increase in risk of 20% after
complete adjustment compared with 7.5E% to <10E%.
Downloaded from https://academic.oup.com/ajcn/article/109/2/411/5261249 by guest on 04 January 2021

418 Ramne et al.
HR
HR
HR
HR
FIGURE 1 Restricted cubic splines of all-cause mortality risk (HRs) from added and free sugar intakes in the MDCS (A, C) and NSHDS (B, D) examined
by using a Cox proportional hazards regression with continuous sugar exposure variables with 10E% as the reference value, adjusted for energy intake,age,
sex, educational level, leisure-time physical activity, smoking status, alcohol habits, dietary habits (fruit and vegetables, processed meat, coffee, SFAs, and ber
density), and BMI (and season and screening date in the MDCS). The solid lines represent the HR, and the dashed lines represent the 95% CI. E%, percentage
of energy; MDCS, Malmö Diet and Cancer Study; NSHDS, Northern Swedish Health and Disease Study.
Additional analyses
No signicant effect modications were observed for sex,
BMI, smoking, or alcohol habits on the association between
added sugar intake and all-cause mortality. Sensitivity analyses
without adjusting for energy (above being expressed as nutrient
densities) were performed for added and free sugars and all-cause
mortality and the results remained similar to those from the fully
adjusted model, indicating that the potential direct association
between energy intake and mortality is not very strong. After
excluding those who underreported or overreported their energy
intakes, the associations between added and free sugar intakes
and all-cause mortality were prominently attenuated in the
highest intake category in the MDCS. However, in the NSHDS,
these associations were, as expected, attenuated in the low intake
category, making the U-shape almost negligible (Supplemental
Tabl e 2 ). After the exclusion of “dietary changers” in the MDCS,
the risk was no longer signicantly increased in any intake
category of added sugar but still appeared somewhat U-shaped. In
addition, the exclusion of those who died ≤2 y from the baseline
examination did not alter the association (data not shown).
SSB consumption, but not treat consumption, was sig-
nicantly associated with higher triglycerides, apoB:apoA-I,
fasting insulin, and HOMA-IR in the MDCS. Higher intakes
of all sugar sources were signicantly associated with lower
HDL-cholesterol values. Higher treat intake was additionally
associated with lower systolic BP, and there was a tendency
for lower HOMA-IR values with higher intakes (nonsignicant).
Topping intake was also associated with lower total cholesterol
but higher apoB:apoA-I (Supplemental Table 3). In the NSHDS,
there were tendencies for elevated fasting glucose, triglycerides,
and total cholesterol in the highest SSB-intake category, but
this tendency was observed in the lowest treat intake category
(nonsignicant) (Supplemental Table 4).
Discussion
This study indicates a nonlinear association between sugar
intake and mortality, which is supported by similar results from 2
separate cohorts. In both cohorts, the 2% of the population with
intakes of added or free sugars >20E% had an ∼30% increased
mortality risk compared with those with intakes just below the
recommended upper limit of added sugar of 10E%. In contrast
to our hypothesis, those with the lowest intakes of added or free
sugars also had an increased risk of >20% in the MDCS and ∼8%
(nonsignicant) in the NSHDS.
Downloaded from https://academic.oup.com/ajcn/article/109/2/411/5261249 by guest on 04 January 2021

Added sugar consumption and mortality 419
TABLE 5 Associations between the intake of different sugar-rich foods and beverages and all-cause mortality in the MDCS and NSHDS1
MDCS NSHDS
Intake categories Intake categories
All-cause mortality C1 (lowest) C2 C3 C4 C5 (highest) PC1 (lowest) C2 C3 C4 C5 (highest) P
Treats
n/Cases/PYs 2751/781/
50,260
6741/1657/
126,986
6319/1567/
119,791
6280/1670/
118,910
2181/634/
40,866
2836/478/
54,059
7993/935/
154,370
4975/490/
96,983
6157/693/
119,352
2514/285/
49,180
Basic adjusted 1 0.78
(0.72, 0.86)
0.72
(0.66, 0.79)
0.69
(0.63, 0.75)
0.76
(0.67, 0.84)
<0.001,2
<0.0013
10.77
(0.69, 0.86)
0.64
(0.57, 0.73)
0.67
(0.59, 0.76)
0.62
(0.53, 0.73)
<0.001,2
<0.0013
Fully adjusted 1 0.86
(0.79, 0.93)
0.83
(0.76, 0.90)
0.77
(0.70, 0.85)
0.83
(0.74, 0.93)
0.001,2
<0.0013
10.85
(0.76, 0.95)
0.71
(0.63, 0.81)
0.75
(0.66, 0.85)
0.66
(0.56, 0.78)
<0.001,2
<0.0013
Toppings
n/Cases/PYs 3250/733/
61,389
6530/1501/
124,102
6408/1642/
120,970
5571/1590/
104,394
2513/843/
45,958
8648/920/
167,072
4655/555/
90,104
5327/614/
103,330
4500/608/
87,192
1345/184/
26,247
Basic adjusted 1 0.90
(0.83, 0.99)
0.87
(0.80, 0.95)
0.86
(0.79, 0.95)
1.07
(0.97, 1.20)
0.022,2
<0.0013
10.97
(0.87, 1.08)
0.91
(0.82, 1.01)
1.08
(0.97, 1.20)
1.17
(0.99, 1.14)
0.019,2
0.0123
Fully adjusted 1 0.92
(0.84, 1.01)
0.90
(0.82, 0.99)
0.86
(0.78, 0.94)
0.93
(0.78, 1.03)
0.316,2
0.0333
11.00
(0.90, 1.12)
0.95
(0.85, 1.05)
0.99
(0.88, 1.10)
0.97
(0.82, 1.15)
0.697,2
0.8643
SSBs
n/Cases/PYs
14,236/3718/
267,908
5003/1178/
95,504
2103/541/
39,673
1468/408/
27,420
1462/464/
26,307
8435/1113/
162,361
8956/953/
174,147
3033/332/
59,153
3108/367/
60,208
943/116/
18,074
Basic adjusted 1 0.97
(0.90, 1.03)
1.01
(0.92, 1.10)
1.11
(1.00, 1.23)
1.24
(1.13, 1.37)
<0.001,2
<0.0013
10.95
(0.87, 1.04)
0.99
(0.88, 1.12)
1.10
(0.97, 1.24)
1.26
(1.04, 1.54)
0.002,2
0.0243
Fully adjusted 1 0.97
(0.90, 1.03)
1.00
(0.91, 1.10)
1.09
(0.98, 1.21)
1.14
(1.03, 1.26)
0.009,2
0.0353
10.96
(0.88, 1.05)
0.97
(0.85, 1.10)
1.03
(0.91, 1.10)
1.10
(0.90, 1.35)
0.177,2
0.5493
1Values are HRs (95% CIs) unless otherwise indicated. Multivariable Cox proportional hazards regression was used to examine the associations. Basic adjusted: adjusted for age, sex, and energy intake (and season and screening date in
the MDCS). Fully adjusted: adjusted for energy intake, age, sex, educational level, leisure-time physical activity, smoking status, alcohol habits, dietary habits (fruit and vegetables, processed meat, coffee, SFAs, and ber density), and BMI
(and season and screening date in the MDCS). Categorizations were performed as follows—treats: ≤2, >2–5, >5–8, >8–14, and >14 servings/wk; toppings: ≤2, >2–7, >7–14, >14–28, and >28 servings/wk; and SSBs: ≤1, >1–3, >3–5,
>5–8, and >8 servings/wk. C, category; MDCS, Malmö Diet and Cancer Study; NSHDS, Northern Swedish Health and Disease Study; PY, person-year; SSB, sugar-sweetened beverage.
2P-trend for intake categories using the median in each category.
3P-overall for intake categories.
Downloaded from https://academic.oup.com/ajcn/article/109/2/411/5261249 by guest on 04 January 2021

420 Ramne et al.
HR HR
HRHR
HR
HR
FIGURE 2 Restricted cubic splines of all-cause mortality risk (HRs) from intakes of treats, toppings, and sugar-sweetened beverages in the MDCS (A,
C, E) and NSHDS (B, D, F) examined by using a Cox proportional hazards regression with continuous sugar exposure variables with 0 servings/wk as the
reference value, adjusted for energy intake, age, sex, educational level, leisure-time physical activity, smoking status, alcohol habits, dietary habits (fruit and
vegetables, processed meat, coffee, SFAs, and ber density), and BMI (and season and screening date in the MDCS). The solid lines represent the HR, and the
dashed lines represent the 95% CI. MDCS, Malmö Diet and Cancer Study; NSHDS, Northern Swedish Health and Disease Study.
This nonlinear association between sugar intake and mortality
has been previously observed in the NIH-AARP Diet and Health
Study in which the lowest HRs (signicantly decreased) between
added sugar intake and all-cause and cardiovascular mortality
were observed in the third and fourth quintiles (7). In the
NHANES study of added sugar and cardiovascular mortality, the
risk appeared to increase linearly, with the highest risk at intakes
>25E%. However, the reference category in that study was
<10E% instead of <5E%, which includes the category of 7.5E%
to <10E%, where the lowest mortality risk was found in the
present study. Thus, it is possible that this reference group was too
broad to show an elevated risk with the low intakes. Furthermore,
in the NHANES, the mean sugar intake was ∼15E% (8), whereas
it was 8E% in the NIH-AARP Diet and Health Study (7), which
is comparable to the MDCS and NSHDS, in which intakes
were 10E% and 8E%, respectively. Thus, these previous ndings
Downloaded from https://academic.oup.com/ajcn/article/109/2/411/5261249 by guest on 04 January 2021

Added sugar consumption and mortality 421
TABLE 6 Associations between added and free sugar intake and cardiovascular mortality and cancer mortality in the MDCS1
Sugar intake categories
<5E% 5E% to <7.5E% 7.5E% to <10E% 10E%to <15E% 15E% to <20E% ≥20E% P
Cardiovascular mortality
Added sugar
n/Cases/PYs 2236/164/41,638 4725/359/89,303 6264/479/119,457 8181/678/153,752 2233/195/41,447 633/77/11,216
Basic adjusted 1.30 (1.08, 1.55) 1.12 (0.97, 1.28) 1 1.03 (0.92, 1.16) 1.07 (0.90, 1.26) 1.82 (1.43, 2.32) 0.326,2
<0.0013
Fully adjusted 1.22 (1.02, 1.47) 1.09 (0.95, 1.25) 1 1.01 (0.89, 1.13) 0.97 (0.82, 1.16) 1.40 (1.09, 1.82) 0.744,2
0.0203
Free sugar
n/Cases/PYs 1554/122/28,782 3687/301/69,668 5606/414/69,668 9376/755/176,540 3154/259/58,797 895/98/16,020
Basic adjusted 1.35 (1.10, 1.65) 1.21 (1.04, 1.40) 1 1.05 (0.93, 1.18) 1.05 (0.90, 1.23) 1.69 (1.35, 2.10) 0.561,2
<0.0013
Fully adjusted 1.24 (1.01, 1.53) 1.15 (0.99, 1.33) 1 1.02 (0.90, 1.15) 0.98 (0.83, 1.15) 1.31 (1.03, 1.66) 0.562,2
0.0233
Cancer mortality
Added sugar
n/Cases/PYs 2236/228/41,638 4725/488/89,303 6264/616/119,457 8181/893/153,752 2233/250/41,447 633/79/11,216
Basic adjusted 1.26 (1.08, 1.47) 1.14 (1.01, 1.28) 1 1.09 (0.99, 1.21) 1.13 (0.98, 1.31) 1.40 (1.11, 1.77) 0.622,2
0.0093
Fully adjusted 1.20 (1.03, 1.41) 1.10 (0.98, 1.24) 1 1.07 (0.97, 1.24) 1.03 (0.89, 1.21 1.11 (0.87, 1.42) 0.476,2
0.2733
Free sugar
n/Cases/PYs 1554/162/28,782 3687/378/69,668 5606/551/69,668 9376/1018/176,540 3154/331/58,797 895/114/16,020
Basic adjusted 1.25 (1.05, 1.49) 1.12 (0.98, 1.27) 1 1.08 (0.99, 1.22) 1.06 (0.93, 1.22) 1.44 (1.17, 1.76) 0.327,2
0.0063
Fully adjusted 1.17 (0.97, 1.39) 1.05 (0.92, 1.20) 1 1.09 (0.98, 1.21) 1.01 (0.88, 1.17) 1.19 (0.96, 1.48) 0.801,2
0.2933
1Values are HRs (95% CIs) unless otherwise indicated. Multivariable Cox proportional hazards regression was used to examine the associations. Basic adjusted: adjusted for
age, sex, energy intake, season, and screening date. Fully adjusted: adjusted for age, sex, energy intake, season, screening date, educational level, leisure-time physical activity,
smoking status, alcohol habits, dietary habits (fruit and vegetables, processed meat, coffee, SFAs, and ber density), and BMI. E%, percentage of energy; MDCS, Malmö Diet and
Cancer Study; PY, person-year.
2P-trend for intake categories using the median in each category.
3P-overall for intake categories.
were examined in 2 American populations under different
circumstances, which makes it difcult to conclude the role
of sugar in mortality. Yet, given the results of this study, the
increased risk with high intakes is supported, but the complexity
of the association between sugar and mortality in epidemiologic
settings is also shown.
Why the mortality risk is increased among low sugar
consumers is difcult to explain biologically or physiologically
in ways other than the possibility of malnutrition, which is very
unlikely in this population, especially considering that sugar
intake is examined as a percentage of energy intake. However,
with regard to malnutrition in terms of micronutrient dilution,
an increased risk has been found at free sugar intakes both
>25E% and <5E% in Australian adults, suggesting a U-shaped
association and (by the authors) an optimal intake between 5E%
and 15E% (25).
The misreporting of dietary intakes is probably a more
plausible explanation for some of the observed increased risk at
low intakes because it is known that underreporting increases
with BMI (26) and that unhealthy foods, such as sugars,
are underreported to a greater extent (27). In this study, the
proportion of energy underreporters was higher in the lowest
intake category of added sugar. However, measures were taken
to minimize this form of bias, such as adjusting for total energy
intake, investigating BMI as a possible effect modier (with-
out signicant results), and excluding misreporters of energy
from the regression model, resulting in an attenuation of the
all-cause mortality risk in the highest intake category in the
MDCS, although, as expected, in the lowest intake category in
the NSHDS, which weakens the U-shape. This indicates that
the elevation in risk at low intakes could partly be explained
by misreporting in the NSHDS but not in the MDCS. However,
misreporting of energy does not automatically mean that sugar
intake is underreported. The low consumers in the MDCS
had the most benecial lipoprotein prole (lowest apoB-to-
apoA-I ratio and highest HDL cholesterol), supporting the
high risk found at high intakes and therefore supporting that
misreporting alone could not explain the high risk in the
low intake category. Previous MDCS ndings support this,
in which disaccharide intake was associated with atherogenic
lipoprotein concentrations (28) and sucrose intake was associated
with an increased risk of coronary events (29), which, in
turn, supports the high mortality risk at high sugar intakes by
increasing the CVD risk, although few other cardiometabolic
risk markers were associated with sugar intake in the present
study.
Low sugar consumption was also associated with many
favorable lifestyle factors, with the exception of higher alcohol
consumption. In addition, the intakes of fat and protein were
higher among those who consumed less sugar. Fiber density was
an important confounder because the high-sugar consumers had
less ber-dense diets and ber has been associated with reduced
risks of chronic disease and mortality (30–32). There is also a
possibility that further adjustments for other potential lifestyle
Downloaded from https://academic.oup.com/ajcn/article/109/2/411/5261249 by guest on 04 January 2021

422 Ramne et al.
confounders, such as stress or sleeping habits, could attenuate the
associations further.
The differences in the associations with mortality between
added and free sugars were practically negligible, but the slightly
weaker association for free sugar found in both cohorts could
possibly be partially explained by benets on cardiovascular risk
factors from vitamins and antioxidants in fruit juices (33–35). In
addition, high juice intake might not be as strongly associated
with unfavorable lifestyle characteristics as a high added sugar
intake.
SSB intake was positively associated with mortality, whereas
treat intake was inversely associated. This discrepancy between
sugar sources has been previously found in the MDCS and other
cohorts, in which pastries were associated with a reduced risk
of chronic disease (29,36,37). Furthermore, in a cross-sectional
analysis of American children and weight status, SSB intake
correlated positively with obesity, whereas solid added sugar
sources correlated negatively (38). This agrees with the usual
conclusion that is drawn about sugar; the evidence for total
sugar intake is rather weak, whereas for SSBs there are clear
associations with adverse health (4,5,39).
SSBs are strongly associated with adverse health likely
due to their liquid state. Even though SSBs may not be as
energy dense as treats, the liquid state results in less satiety
in relation to energy content (40). Therefore, the compensation
of energy intake is generally inadequate, which results in
positive energy balance and weight gain over time (41). The
caffeine content of some SSBs could also potentially contribute
to the palatability and “addictiveness” of SSBs (42). One
could also assume that SSB intake would be more easily
assessed, because serving sizes are often standardized in cans
or bottles, whereas other sweet snacks could be more difcult to
recall.
A possible explanation that could contribute to the increased
risk observed with low treat intake is the less favorable prole
of risk markers found with low treat intake (i.e., the negative
association with systolic BP in the MDCS and tendencies
for higher blood lipids and impaired glycemic control in the
NSHDS). This could indicate that subjects are avoiding treats for
some health reason or that a high treat intake is not associated
with an unfavorable lifestyle to the same extent as for SSBs.
Treat consumption is often more culturally attached than is
the consumption of SSBs. In Sweden, taking breaks at work
with coffee and pastries is a strong social tradition (43). Thus,
one could attempt to explain the negative association for treats
by speculating that low consumers might have fewer social
connections, which, in turn, has been associated with higher
mortality (44).
One could also assume that the dietary habits during the
data collection (1990s) somewhat differed from current di-
etary habits and also from the habits during the participants’
adolescence. Thus, the association would perhaps differ if
later generations were studied. Nevertheless, dietary habits will
constantly change, and cohort studies require long follow-up,
which unavoidably will challenge the generalizability of the
results to the present. This limitation as well as uncertain dietary
data due to misreporting and estimations represent the main
limitations of this study, whereas the major strength is that the 2
cohorts show similar results, despite different dietary assessment
methods.
In conclusion, this study indicates that a high sugar consump-
tion of >20%E (reached by 2% of this population) is associated
with an increased mortality risk. However, the risk was also
increased for low sugar consumers, despite their more favorable
lifestyle in general. The 2 independent cohorts show the same
results, which do not support lowering the recommended upper
limits of sugar intake. However, potential confounding from
unmeasured lifestyle factors cannot be excluded. In addition,
the associations were dependent on the type of sugar-rich
products consumed, and future research and policy decisions
may emphasize consideration of the differences between sugar
sources.
The authors’ responsibilities were as follows: SR and ES: planned the
study; SR: performed all the statistical analyses and wrote the manuscript
under the supervision of ES; IJ: provided the data from NSHDS; JAD, EG-P,
KO, BL, GE, UE, IJ, and ES: provided useful comments on the manuscript,
tables, and gures; and all authors: read and approved the nal manuscript.
None of the authors had any conicts of interest to declare.
References
1. Te Morenga LA, Howatson AJ, Jones RM, Mann J. Dietary sugars
and cardiometabolic risk: systematic review and meta-analyses of
randomized controlled trials of the effects on blood pressure and lipids.
Am J Clin Nutr 2014;100(1):65–79.
2. Imamura F, O’Connor L, Ye Z, Mursu J, Hayashino Y, Bhupathiraju
SN, Forouhi NG. Consumption of sugar sweetened beverages,
articially sweetened beverages, and fruit juice and incidence of type 2
diabetes: systematic review, meta-analysis, and estimation of population
attributable fraction. BMJ 2015;351:h3576.
3. Sonestedt E, Overby NC, Laaksonen DE, Birgisdottir BE. Does high
sugar consumption exacerbate cardiometabolic risk factors and increase
the risk of type 2 diabetes and cardiovascular disease? Food Nutr Res
2012;56.
4. Hauner H, Bechthold A, Boeing H, Bronstrup A, Buyken A, Leschik-
Bonnet E, Linseisen J, Schulze M, Strohm D, Wolfram G. Evidence-
based guideline of the German Nutrition Society: carbohydrate
intake and prevention of nutrition-related diseases. Ann Nutr Metab
2012;60(Suppl 1):1–58.
5. European Food Safety Authority Panel on Dietetic Products, Nutrition,
and Allergies. Scientic opinion on Dietary Reference Values for
carbohydrates and dietary bre. EFSA J 2010;8(3): 1462.
6. World Cancer Research Fund/American Institute for Cancer Research.
Third expert report: diet, nutrition, physical activity and cancer: a global
perspective. London: World Cancer Research Fund International; 2018.
7. Tasevska N, Park Y, Jiao L, Hollenbeck A, Subar AF, Potischman N.
Sugars and risk of mortality in the NIH-AARP Diet and Health Study.
Am J Clin Nutr 2014;99(5):1077–88.
8. Yang Q, Zhang Z, Gregg EW, Flanders WD, Merritt R, Hu FB. Added
sugar intake and cardiovascular diseases mortality among US adults.
JAMA Intern Med 2014;174(4):516–24.
9. Nordic Council of Ministers. Nordic Nutrition Recommendations
2012: integrating physical activity and nutrition. 5th ed.
Copenhagen(Denmark): Nordic Council of Ministers; 2014.
10. US Department of Health and Human Services; USDA.2015–2020
Dietary guidelines for Americans. 8th ed. Washington (DC): US
Department of Health and Human Services, USDA; 2015.
11. Institute of Medicine. Dietary References Intakes for energy,
carbohydrate, ber, fat, fatty acids, cholesterol, protein, and amino
acids, Washington (DC): National Academies Press; 2005.
12. Public Health England; Scientic Advisory Committee on Nutrition.
Why 5%? An explanation of SACN’s recommendation about sugars and
health, London: Public Health England; 2015.
13. WHO. Guideline: Sugars Intake for Adults and Children. Geneva
(Switzerland): WHO; 2015.
14. Manjer J, Carlsson S, Elmstahl S, Gullberg B, Janzon L, Lindstrom
M, Mattisson I, Berglund G. The Malmo Diet and Cancer Study:
representativity, cancer incidence and mortality in participants and non-
participants. Eur J Cancer Prev 2001;10(6):489–99.
Downloaded from https://academic.oup.com/ajcn/article/109/2/411/5261249 by guest on 04 January 2021

Added sugar consumption and mortality 423
15. Manjer J, Elmstahl S, Janzon L, Berglund G. Invitation to a population-
based cohort study: differences between subjects recruited using various
strategies. Scand J Public Health 2002;30(2):103–12.
16. Berglund G, Elmstahl S, Janzon L, Larsson SA. The Malmo Diet and
Cancer Study: design and feasibility. J Intern Med 1993;233(1):45–51.
17. Wirfalt E, Mattisson I, Johansson U, Gullberg B, Wallstrom P,
Berglund G. A methodological report from the Malmo Diet and Cancer
study: development and evaluation of altered routines in dietary data
processing. Nutr J 2002;1:3.
18. Riboli E, Elmstahl S, Saracci R, Gullberg B, Lindgarde F. The Malmo
Food Study: validity of two dietary assessment methods for measuring
nutrient intake. Int J Epidemiol 1997;26(Suppl 1):S161–73.
19. Johansson I, Hallmans G, Wikman A, Biessy C, Riboli E, Kaaks
R. Validation and calibration of food-frequency questionnaire
measurements in the Northern Sweden Health and Disease cohort.
Public Health Nutr 2002;5(3):487–96.
20. Hedblad B, Nilsson P, Engstrom G, Berglund G, Janzon L. Insulin
resistance in non-diabetic subjects is associated with increased
incidence of myocardial infarction and death. Diabetic Med
2002;19(6):470–5.
21. Norberg M, Wall S, Boman K, Weinehall L. The Vasterbotten
Intervention Programme: background, design and implications. Global
Health Action 2010;3.
22. StataCorp. mkspline—Linear and restricted cubic spline construction.
[cited 2018 May 30]. Available from: https://www.stata.com/manuals1
3/rmkspline.pdf.
23. Black AE. Critical evaluation of energy intake using the Goldberg
cut-off for energy intake:basal metabolic rate: a practical guide to
its calculation, use and limitations. Int J Obes Relat Metab Disord
2000;24(9):1119–30.
24. Mattisson I, Wirfalt E, Aronsson CA, Wallstrom P, Sonestedt
E, Gullberg B, Berglund G. Misreporting of energy: prevalence,
characteristics of misreporters and inuence on observed risk estimates
in the Malmo Diet and Cancer cohort. Br J Nutr 2005;94(5):
832–42.
25. Mok A, Ahmad R, Rangan A, Louie JCY. Intake of free sugars
and micronutrient dilution in Australian adults. Am J Clin Nutr
2018;107(1):94–104.
26. Subar AF, Kipnis V, Troiano RP, Midthune D, Schoeller DA, Bingham
S, Sharbaugh CO, Trabulsi J, Runswick S, Ballard-Barbash R, et al.
Using intake biomarkers to evaluate the extent of dietary misreporting
in a large sample of adults: the OPEN study. Am J Epidemiol
2003;158(1):1–13.
27. Poppitt SD, Swann D, Black AE, Prentice AM. Assessment of selective
under-reporting of food intake by both obese and non-obese women in
a metabolic facility. Int J Obes Relat Metab Disord 1998;22(4):303–11.
28. Sonestedt E, Wirfalt E, Wallstrom P, Gullberg B, Drake I, Hlebowicz
J, Nordin Fredrikson G, Hedblad B, Nilsson J, Krauss RM, et al. High
disaccharide intake associates with atherogenic lipoprotein prole. Br J
Nutr 2012;107(7):1062–9.
29. Warfa K, Drake I, Wallstrom P, Engstrom G, Sonestedt E. Association
between sucrose intake and acute coronary event risk and effect
modication by lifestyle factors: Malmo Diet and Cancer Cohort Study.
Br J Nutr 2016;116(9):1611–20.
30. Threapleton DE, Greenwood DC, Evans CE, Cleghorn CL, Nykjaer C,
Woodhead C, Cade JE, Gale CP, BurleyVJ. Dietary bre intake and risk
of cardiovascular disease: systematic review and meta-analysis. BMJ
2013;347:f6879.
31. Liu L, Wang S, Liu J. Fiber consumption and all-cause, cardiovascular,
and cancer mortalities: a systematic review and meta-analysis of cohort
studies. Mol Nutr Food Res 2015;59(1):139–46.
32. Bradbury KE, Appleby PN, Key TJ. Fruit, vegetable, and ber intake
in relation to cancer risk: ndings from the European Prospective
Investigation into Cancer and Nutrition (EPIC). Am J Clin Nutr
2014;100(Suppl 1):394S–8S.
33. Dohadwala MM, Holbrook M, Hamburg NM, Shenouda SM, Chung
WB, Titas M, Kluge MA, Wang N, Palmisano J, Milbury PE,
et al. Effects of cranberry juice consumption on vascular function in
patients with coronary artery disease. Am J Clin Nutr 2011;93(5):
934–40.
34. Habauzit V, Verny MA, Milenkovic D, Barber-Chamoux N,
Mazur A, Dubray C, Morand C. Flavanones protect from arterial
stiffness in postmenopausal women consuming grapefruit juice for
6 mo: a randomized, controlled, crossover trial. Am J Clin Nutr
2015;102(1):66–74.
35. Moazzen H, Alizadeh M. Effects of pomegranate juice on
cardiovascular risk factors in patients with metabolic syndrome:
a double-blinded, randomized crossover controlled trial. Plant Foods
Hum Nutr 2017;72(2):126–33.
36. von Ruesten A, Feller S, Bergmann MM, Boeing H. Diet and risk of
chronic diseases: results from the rst 8 years of follow-up in the EPIC-
Potsdam study. Eur J Clin Nutr 2013;67(4):412–9.
37. Ericson U, Hellstrand S, Brunkwall L, Schulz CA, Sonestedt E,
Wallstrom P, Gullberg B, Wirfalt E, Orho-Melander M. Food sources
of fat may clarify the inconsistent role of dietary fat intake for incidence
of type 2 diabetes. Am J Clin Nutr 2015;101(5):1065–80.
38. Welsh JA, Wang Y, Figueroa J, Brumme C. Sugar intake by type (added
vs. naturally occurring) and physical form (liquid vs. solid) and its
varying association with children’s body weight, NHANES 2009–2014.
Pediatr Obesity 2018;13(4):213–21.
39. Malik VS, Pan A, Willett WC, Hu FB. Sugar-sweetened beverages
and weight gain in children and adults: a systematic review and meta-
analysis. Am J Clin Nutr 2013;98(4):1084–102.
40. Pan A, Hu FB. Effects of carbohydrates on satiety: differences between
liquid and solid food. Curr Opin Clin Nutr Metab Care 2011;14(4):385–
90.
41. DiMeglio DP, Mattes RD. Liquid versus solid carbohydrate: effects
on food intake and body weight. Int J Obes Relat Metab Disorder
2000;24(6):794–800.
42. Keast RS, Swinburn BA, Sayompark D, Whitelock S, Riddell
LJ. Caffeine increases sugar-sweetened beverage consumption in
a free-living population: a randomised controlled trial. Br J Nutr
2015;113(2):366–71.
43. Ax E, Warensjo Lemming E, Becker W, Andersson A, Lindroos AK,
Cederholm T, Sjogren P, Fung TT. Dietary patterns in Swedish adults:
results from a national dietary survey. Br J Nutr 2016;115(1):95–104.
44. Holt-Lunstad J, Smith TB, Layton JB. Social relationships and mortality
risk: a meta-analytic review. PLoS Med 2010;7(7):e1000316.
Downloaded from https://academic.oup.com/ajcn/article/109/2/411/5261249 by guest on 04 January 2021