American Journal of Epidemiology
© The Author 2014. Published by Oxford University Press on behalf of the Johns Hopkins Bloomberg School of
Public Health. All rights reserved. For permissions, please e-mail: [email protected].
Vol. 179, No. 7
DOI: 10.1093/aje/kwt446
Advance Access publication:
February 25, 2014
Original Contribution
The Contributions of Solar Ultraviolet Radiation Exposure and Other Determinants to
Serum 25-Hydroxyvitamin D Concentrations in Australian Adults: The AusD Study
Michael G. Kimlin*, Robyn M. Lucas, Simone L. Harrison, Ingrid van der Mei, Bruce K. Armstrong,
David C. Whiteman, Anne Kricker, Madeleine Nowak, Alison M. Brodie, and Jiandong Sun
* Correspondence to Prof. Michael Kimlin, Institute of Health and Biomedical Innovation, Queensland University of Technology, 60 Musk Avenue,
Kelvin Grove, Queensland, Australia 4059 (e-mail: [email protected]).
Initially submitted September 26, 2013; accepted for publication December 27, 2013.
The Quantitative Assessment of Solar UV [ultraviolet] Exposure for Vitamin D Synthesis in Australian Adults
(AusD) Study aimed to better define the relationship between sun exposure and serum 25-hydroxyvitamin
D (25(OH)D) concentration. Cross-sectional data were collected between May 2009 and December 2010 from
1,002 participants aged 18–75 years in 4 Australian sites spanning 24° of latitude. Participants completed the following: 1) questionnaires on sun exposure, dietary vitamin D intake, and vitamin D supplementation; 2) 10 days of
personal ultraviolet radiation dosimetry; 3) a sun exposure and physical activity diary; and 4) clinical measurements
and blood collection for 25(OH)D determination. Our multiple regression model described 40% of the variance in
25(OH)D concentration; modifiable behavioral factors contributed 52% of the explained variance, and environmental and demographic or constitutional variables contributed 38% and 10%, respectively. The amount of skin
exposed was the single strongest contributor to the explained variance (27%), followed by location (20%), season
(17%), personal ultraviolet radiation exposure (8%), vitamin D supplementation (7%), body mass index (weight (kg)/
height (m)2) (4%), and physical activity (4%). Modifiable behavioral factors strongly influence serum 25(OH)D
concentrations in Australian adults. In addition, latitude was a strong determinant of the relative contribution of
different behavioral factors.
determinants; sunlight; ultraviolet; vitamin D
Abbreviations: AusD, A Quantitative Assessment of Solar UV [ultraviolet] Exposure for Vitamin D Synthesis in Australian Adults;
BMI, body mass index; CI, confidence interval; SED, standard erythemal dose; 25(OH)D, 25-hydroxyvitamin D; UVR, ultraviolet
radiation.
Past research suggests that most Australians obtain more
than 90% of their vitamin D from endogenous synthesis of
vitamin D3 (cholecalciferol) (13), initiated by exposure of 7dehydrocholesterol in the skin to ultraviolet B radiation (14).
Consequently, factors that influence an individual’s UVR exposure (e.g., time spent outdoors) affect cutaneous penetration of UVR (e.g., darker skin pigmentation) or the skin’s
ability to utilize UVR to synthesize vitamin D (e.g., skin
aging) are likely to affect serum 25(OH)D concentration.
The primary aim of this article is to describe comprehensively the environmental, constitutional, and behavioral
determinants of serum 25(OH)D concentration. We also
Vitamin D plays an essential role in maintaining bone
health (1), with deficiency causing osteomalacia in adults and
rickets in children (2). Low vitamin D status (measured as the
serum concentration of 25-hydroxyvitamin D (25(OH)D))
has also been linked to increased risk of nonskeletal health
outcomes, including cardiovascular disease (3), diabetes
(4), multiple sclerosis (5, 6), cognitive dysfunction (7), and
several cancers (8, 9). Vitamin D insufficiency is reported
to be common in many countries, including some with high
ambient solar ultraviolet radiation (UVR), such as Australia
(10–12), which suggests that factors other than ambient UVR
may influence vitamin D status.
864
Am J Epidemiol. 2014;179(7):864–874
Determinants of 25-Hydroxyvitamin D in Australia 865
aimed to investigate the contribution of potentially modifiable behaviors to variance in serum 25(OH)D concentration
and to examine whether the determinants of 25(OH)D concentration varied across a broad latitude range.
METHODS
Study design
The AusD Study was a multicenter, cross-sectional study
of 1,002 Australian adults (54.2% female) aged 18–75 years.
Data were collected in the following 4 Australian cities:
Townsville (19.3° south, 146° east), Brisbane (27.5° south,
153° east), Canberra (35.3° south, 149° east), and Hobart
(42.8° south, 147° east) between May 2009 and December
2010. Study participants were randomly recruited from the
Australian electoral roll. The ethics committees of all participating institutions (Queensland University of Technology
(Brisbane, Queensland); James Cook University (Townsville, Queensland); Australian National University (Canberra,
Australian Capital Territory); and University of Tasmania
(Hobart, Tasmania)) approved the study. Detailed methodology is available elsewhere (15).
Data collection
Participants attended 2 interviews, 10 days apart, and
provided data through questionnaires and physical measurements. Between the 2 interviews—the index exposure
period—participants completed a daily sun diary that provided detail of time spent outdoors between 6 AM and 6 PM
and wore a new polysulphone UVR dosimeter on a wristband
each day (15).
At the second interview, a nonfasting blood sample was
collected. Blood samples were centrifuged within 4 hours
of collection, and serum was stored immediately at −80°C.
At study completion, all stored sera were analyzed for
25(OH)D concentration (in nmol/L) using the Liaison semiautomated chemiluminescence assay (DiaSorin, S.p.A., Saluggia, Italy). Internal and external vitamin D standards were
included in each batch. Intraassay and interassay variations
were 3%–6% and 6%–9%, respectively. Low (<15 nmol/L)
and high (>125 nmol/L) serum 25(OH)D concentrations were
reassayed. This assay consistently returns results within 2%
of the Vitamin D External Quality Assessment Scheme alllaboratory trimmed mean.
Daily ambient (environmental) UVR and meteorological
data (maximum temperature and relative humidity) for each
day of each participant’s index exposure period were obtained from the Australian Radiation Protection and Nuclear
SafetyAgency(Canberra, Australian Capital Territory)and the
Bureau of Meteorology (Melbourne, Victoria), respectively.
Predictor variables
Potential predictors were chosen on the basis of their documented effects on 25(OH)D concentration (2, 11). The analyzed variables are described below.
Demographic and constitutional variables. We recorded
sex, age, country of birth, educational level, employment
Am J Epidemiol. 2014;179(7):864–874
status, indoor/outdoor occupation, self-reported skin color,
natural hair color (at age 18 years), eye color, perceived general health, previous cancer history, previous diagnosis of
high blood pressure or cholesterol, and melanin density
of the upper arm, dorsum of the hand, and cheek measured
using reflectance spectrophotometry and a published algorithm (16). Skin reflectance of the dorsum of the hand is indicative of tan (acquired skin color), which is a mixture of
behavior and constitution (ability to tan); on the inner
upper arm, it indicates natural skin color (i.e., constitution).
Environmental variables. We considered study location,
season during which the participant entered the study, daily
total standard erythemal dose (SED) of ambient UVR (1 SED =
100 J/m2), daily maximum temperature (in °C), and daily average relative humidity (as %) during the index exposure period.
Behavioral variables. We recorded body mass index
(BMI) (weight (kg)/height (m)2), physical activity level (inactive, mildly active, moderately active, highly active (17)),
smoking (never, ever), alcohol consumption (<1 drink/week,
1–6 drinks/week, ≥7 drinks/week), vitamin D supplementation (ever vs. never in the past month), estimated vitamin D
intake from oily fish (<25 mg/month, ≥25 mg/month), sunscreen use (yes, no), time spent outdoors (in minutes), proportion of skin covered by clothing (as %), and personal
UVR exposure (as SED) as measured by UVR dosimetry.
Calculation of selected predictor variables
All person-day data, including ambient UVR, temperature,
humidity, personal UVR exposure, and variables calculated
using the diary data (daily outdoor time and daily proportion
of skin area covered by clothing) (18, 19) were averaged
across the index exposure period for each person. Some participants had fewer than 10 days of data because of noncompliance or missing data; however, most participants (>92%)
had at least 7 days of data for all variables.
Time spent outdoors. For each hour from 6 AM to 6 PM
(Australian Eastern Standard Time) in the index exposure period, participants reported their time spent outdoors (in minutes) by indicating 1 of 5 categories (0, <15, 15–29, 30–44 or
45–60 minutes). Total daily time spent outdoors was calculated by summing midpoint values (0, 7.5, 22.5, 37.5 and
52.5 minutes, respectively) of each of the 5 time categories
for the 12 hourly intervals (18).
Body surface area covered by clothing. Clothing cover on
the upper and lower body, head, hands, and feet was recorded
for each hour with reference to the clothing guide provided.
Each garment was assigned a relative body surface area
(whole body surface area = 1) using the relative surface proportions reported by Pearl and Scott (20) and the body surface area of finer body-site divisions used previously (18,
21). The proportion of skin covered during each hourly interval was calculated by summing the relative proportions at
each body site. The average proportion (as %) of skin area
covered when outdoors each day (Propoutdoor) was calculated
as follows (18, 21):
P17
Propoutdoor ¼
propt × outdoort
;
P17
t¼6 outdoor t
t¼6
ð1Þ
866 Kimlin et al.
Table 1. Mean 25(OH)D Values of Variables Related to Serum
25(OH)D Concentration in Univariate Analyses (P < 0.10) in Australian
Adults, AusD Study, Australia, 2009–2010
Variable
No.
%
25(OH)D,
nmol/L
Mean (SD)
P Value
Location
Townsville
259
25.9
73.6 (21.0)
Brisbane
254
25.4
62.3 (24.8)
Canberra
252
25.2
48.5 (18.4)
Hobart
237
23.7
54.2 (23.4)
Winter (June–August)
322
32.1
50.9 (23.2)
Spring (September–
November)
380
37.9
60.5 (21.9)
<0.001
105
10.5
74.1 (23.6)
Autumn (March–May)
195
19.5
65.2 (23.5)
543
54.2
58.5 (24.4)
Male
459
45.8
<0.001
0.064
61.3 (23.3)
243
24.3
63.1 (25.3)
35–64
564
56.3
58.2 (23.1)
65–75
195
19.5
60.2 (24.1)
Australia
806
80.4
60.7 (23.8)
Other countrya
196
19.6
55.9 (23.9)
Full-time
481
48.1
57.2 (22.9)
Part-time
173
17.3
59.8 (23.3)
Retired
191
19.1
61.8 (25.4)
Otherb
156
15.6
65.1 (25.0)
Mainly indoors
686
71.0
57.1 (23.4)
Mainly or half outdoors
280
29.0
66.2 (24.2)
Fair
628
63.4
60.2 (24.2)
Medium
258
26.0
61.0 (23.6)
Dark/black/olive
105
10.6
55.1 (22.7)
Blond/red
237
23.8
60.4 (23.8)
Brown
655
65.9
60.6 (23.5)
Black
102
10.3
52.8 (26.6)
%
25(OH)D,
nmol/L
Mean (SD)
P Value
<0.001
Eye color
Blue/grey/green
544
54.8
59.0 (23.2)
Hazel
247
24.9
65.3 (25.2)
Brown
201
20.3
55.4 (23.3)
<25.0
386
38.6
62.3 (24.7)
25.0–29.9
348
34.8
60.4 (24.2)
≥30.0
265
26.5
55.1 (21.9)
Inactive
284
28.5
54.2 (22.8)
Minimally active
344
34.5
59.3 (22.0)
Highly active
369
37.0
64.7 (25.4)
None
635
64.5
56.3 (23.9)
Any
349
35.5
65.7 (23.0)
<25
625
62.8
58.1 (23.0)
≥25
370
37.2
62.8 (25.5)
No
558
55.7
58.1 (24.1)
Yes
443
44.3
61.9 (23.6)
Low (0–1 SEDd)
466
46.8
55.3 (24.5)
Medium (1.01–
2.0 SED)
268
26.9
59.9 (21.7)
High (≥2.01 SED)
261
26.2
67.6 (23.0)
0.001
Physical activity
<0.001
Vitamin D oral
supplements
Age group, years
18–34
No.
BMI
Sex
Female
Variable
c
Season of participation
Summer (December–
February)
Table 1. Continued
0.027
<0.001
Vitamin D intake from
oily fish, mg/month
0.003
Recent sunscreen use
Country of birth
0.011
0.013
Personal daily average
UVR exposure
Employment status
0.002
Occupation type
<0.001
Self-reported skin color
0.087
Natural hair color
0.008
<0.001
Abbreviations: AusD, A Quantitative Assessment of Solar UV
Exposure for Vitamin D Synthesis in Australian Adults; BMI, body
mass index; 25 (OH)D, 25-hydroxyvitamin D; SD, standard deviation; SED, standard erythemal dose; UVR, ultraviolet radiation.
a
Including Northern Europe (n = 117), Southern Europe (n = 8),
Asia (n = 28), North America (n = 9), South America (n = 3), Middle
East (n = 2), New Zealand (n = 20), Papua New Guinea/Solomon
Islands/Vanuatu (n = 6), and Africa (n = 3).
b
Including unemployment (n = 11), home duties (n = 59), students
(n = 42), sole-parent pension (n = 3), disability pension (n = 21), and
other (n = 20).
c
Weight (kg)/height (m)2.
d
1 SED = 100 J/m2.
Table continues
where propt is the proportion covered by clothing at hour t (6–
7 AM to 5–6 PM), and outdoort is the time (in minutes) spent
outdoors during hour t. Daily average personal UVR exposure was the average of each participant’s UVR exposure
for the days of their personal dosimeter use.
Vitamin D intake was estimated (in mg/month) from oily
fish intake only. Previous research suggests that, for Austra-
lians not taking supplements, fish is the major dietary source
of vitamin D (13).
Statistical analyses
Analyses were undertaken using R software (The R Project, University of Auckland, Auckland, New Zealand). Serum
25(OH)D concentration (in nmol/L) was the dependent
Am J Epidemiol. 2014;179(7):864–874
Determinants of 25-Hydroxyvitamin D in Australia 867
B)
150
Serum 25(OH)D Concentration,
nmol/L
Serum 25(OH)D Concentration,
nmol/L
A)
100
50
0
1
2
3
4
5
6
7
8
9
150
100
50
0
10
1
Decile Groups of Personal UVR Exposure
2
3
4
5
6
7
8
9
10
Decile Groups of Personal UVR Exposure
D)
150
Serum 25(OH)D Concentration,
nmol/L
Serum 25(OH)D Concentration,
nmol/L
C)
100
50
0
1
2
3
4
5
6
7
8
9
150
100
50
0
10
1
Decile Groups of Personal UVR Exposure
2
3
4
5
6
7
8
9
10
Decile Groups of Personal UVR Exposure
Figure 1. Serum 25-hydroxyvitamin D (25(OH)D) concentration (in nmol/L) in relation to deciles of estimated personal ultraviolet radiation (UVR)
exposure (as standard erythemal dose (SED)), in the following Australian sites: A) Townsville (Spearman’s ρ = 0.16, 95% confidence interval: 0.04,
0.28); B) Brisbane (Spearman’s ρ = 0.05, 95% confidence interval: −0.09, 0.18); C) Canberra (Spearman’s ρ = 0.26, 95% confidence interval: 0.13,
0.38); and D) Hobart (Spearman’s ρ = 0.28, 95% CI: 0.16, 0.40). Quantitative Assessment of Solar UV [ultraviolet] Exposure for Vitamin D Synthesis
in Australian Adults (AusD Study), Australia, 2009–2010. Ranges of deciles (in SED) are as follows: first decile, 0.0–0.29; second decile, 0.3–0.49;
third decile, 0.5–0.59; fourth decile, 0.6–0.79; fifth decile, 0.8–1.09; sixth decile, 1.1–1.39; seventh decile, 1.4–1.79; eighth decile, 1.8–2.29; ninth
decile, 2.3–3.19; tenth decile, 3.2–21. 1 SED = 100 J/m2.
variable. Records with missing variables were excluded from
analyses involving those variables.
A 1-way analysis of variance or independent-sample t test
was used to identify categorical variables associated with
25(OH)D level. Univariate relationships between continuous
variables and 25(OH)D were assessed using correlation
analysis.
A multiple regression model of serum 25(OH)D concentration was developed using backward stepwise elimination
with location, sex, and age group forced into the model at
the final step (22). All independent variables with P values
of less than 0.10 in univariate analyses were included in the
initial model, and then a variable was removed in each step if
there was a fall in the model Akaike information criterion
value. The final model was that with the lowest Akaike information criterion value. The predictive ability of the model
was expressed as the proportion of total variance explained
(R 2), and the relative importance of each factor was estimated
using the R 2 contribution averaged over orderings among
regressors calculated using the “relaimpo” package for R
software (23). We used the variable inflation factor to
check for multicollinearity, the Breusch-Pagan test and examination of the residual plot to test for heteroscedasticity,
Am J Epidemiol. 2014;179(7):864–874
and a Bonferroni outlier test to identify possible outliers. We
used White-Huber corrected standard errors (24) to adjust for
heteroscedasticity.
RESULTS
Sample description and overall 25(OH)D concentration
In each study site, numbers recruited per week varied from
1 to 8, with the low recruitment weeks concentrated in and
around holiday periods (late December to mid-January). A
total of 1,002 participants (543 women) with a mean age of
48.15 (standard deviation, 15.68) years completed this study.
Participants’ characteristics are described in detail elsewhere
(15). Serum 25(OH)D concentrations ranged from 2–
178 nmol/L with a slightly positively skewed distribution
(skewness = 0.63, kurtosis = 1.01). The mean 25(OH)D concentration was 59.8 (standard deviation, 23.9) nmol/L, and its
distribution was as follows: less than 25.0 nmol/L, 5.6%;
25.0–49.9 nmol/L, 29.3%; 50.0–74.9 nmol/L, 42.4%; and
75.0 nmol/L or more, 22.7%. The median dose of vitamin
D in those taking oral supplements (n = 349) was 401 IU
per day (range, 10–10,000 IU/day).
868 Kimlin et al.
Table 2. Final Multiple Regression Model for the Determinants of 25(OH)D Concentration in Australian Adults, AusD
Study, Australia, 2009–2010
βa
Independent Variable by Model
95% CI
P Value
Overall model
R2
0.40
Intercept
63.41
39.01, 87.81
RC, %b
100
<0.001
Environmental variables
37.82
Study center (degrees latitude)
19.70
Townsville (19.3° south)
Referent
Brisbane (27.5° south)
−7.70
Canberra (35.3° south)
−13.26
−17.73, −8.78
<0.001
−9.53
−14.09, −4.97
<0.001
Hobart (42.8° south)
−11.57, −3.83
<0.001
Season entered the study
16.74
Winter (June–August)
Referent
−0.34, 6.41
Spring (September–November)
3.04
Summer (December–February)
19.77
14.62, 24.91
<0.001
Autumn (March–May)
11.64
7.74, 15.53
<0.001
0.02, 0.51
0.037
Ambient UVR, SED
NR
Daily maximum temperature (°C)
NR
Relative humidity (%)
0.26
0.078
Demographic and constitutional variables
Male sex
1.38
10.03
1.87
−0.96, 4.70
0.194
Age group, years
0.53
2.29
15–34
Referent
35–64
−5.56
−8.88, −2.24
0.001
65–75
−5.85
−10.41, −1.28
0.012
Country of birth (other countries)
−2.91
−6.29, 0.47
0.092
0.98
Employment status (not full-time)
2.73
−0.14, 5.60
0.062
1.31
Occupation type (some outdoor)
2.80
−0.47, 6.08
0.094
2.47
Skin color (medium to dark)
NR
Hair (blown or black)
NR
Eye color (hazel or brown)
NR
Table continues
Potential determinants of 25(OH)D concentration
Variables associated with 25(OH)D concentration (P <
0.10) are shown in Table 1. Mean 25(OH)D concentration
was highest in Townsville (latitude, 19.3° south) and lowest
in Canberra (latitude, 35.3° south), highest in summer and
lowest in winter in all sites, and slightly higher in men than
in women. There was a nonlinear association with age, with
the highest 25(OH)D concentration in the youngest age group
(18–34 years) and the lowest in those aged 35–64 years. Participants who reported their occupations as “other” (mostly
undertook home duties or were students) had higher 25(OH)D
concentrations than retirees, part-time workers, or full-time
workers. Overall, 25(OH)D concentrations were highest
in Australian-born participants who were outdoor workers,
physically active, and who reported a recent history of sunscreen use, had a lower BMI, took oral vitamin D supplements, or consumed an estimated 25 mg/month or more of
vitamin D from oily fish. Concentrations of 25(OH)D did
not vary according to educational level, perceived health, history of cancer, history of hypertension or high cholesterol,
smoking status, or alcohol consumption (data not shown).
Concentration of 25(OH)D was directly correlated with
melanin density (inner upper arm, r = 0.17, P < 0.001; dorsum of the hand, r = 0.14, P < 0.001; cheek, r = 0.10, P =
0.002), ambient UVR (r = 0.38, P < 0.001), daily maximum
temperature (r = 0.45, P < 0.001), daily average humidity
(r = 0.08, P = 0.011), and time spent outdoors daily (r = 0.16,
P < 0.001) and was inversely correlated with clothing cover
(body surface area covered, r = −0.50, P < 0.001).
Daily average personal UVR exposure (data available for
995 (99%) participants) varied from 0.01 to 20.66 SED, with
a median of 1.08 (interquartile range, 0.55–2.06) SED. Personal UVR exposure and 25(OH)D concentration were positively correlated (Pearson’s r = 0.18, P < 0.001; Spearman’s
ρ = 0.24, P < 0.001). The large difference between the Pearson
and Spearman coefficients indicates a rank-linear relationship between personal UVR and 25(OH)D concentration.
Am J Epidemiol. 2014;179(7):864–874
Determinants of 25-Hydroxyvitamin D in Australia 869
Table 2. Continued
Independent Variable by Model
βa
95% CI
P Value
R2
RC, %b
Melanin density (%)
Inner upper arm
Hand
Cheek
NR
2.45
0.11, 4.80
0.040
Modifiable behavioral variables
52.16
Body mass indexc
4.14
<25.0
Referent
25.0–29.9
−2.89
−5.99, 0.21
0.068
≥30.0
−7.83
−11.19, −4.47
<0.001
Physical activity
3.56
Inactive
Minimally active
Highly active
Vitamin D supplement use (yes)
Vitamin D intake from oily fish (≥25 mg/month)
Recent sunscreen use (yes)
2.45
NR
Referent
2.67
−0.59, 5.92
0.108
5.05
1.63, 8.46
0.004
8.67
6.00, 11.35
<0.001
7.35
NR
NR
Outdoor time, minutes
−0.02
−0.04, 0.00
0.098
1.94
Clothing cover (%)
−0.51
−0.68, −0.34
<0.001
26.91
1.31, 2.55
<0.001
8.26
Personal UVR level, SEDd
1.93
Abbreviations: AusD, A Quantitative Assessment of Solar UV Exposure for Vitamin D Synthesis in Australian
Adults; CI, confidence interval; 25(OH)D, 25-hydroxyvitamin D; NR, not retained (in the final model); RC, relative
contribution; SED, standard erythemal dose; UVR, ultraviolet radiation.
a
Regression coefficient.
b
The relative contribution (as %) to the overall R 2 of the model adjusted for all other variables.
c
Weight (kg)/height (m)2.
d
1 SED = 100 J/m2.
Therefore, a new variable (from 1 to 10) corresponding to the
10 deciles of personal UVR exposure was created (Pearson’s
r = 0.24 with 25(OH)D, P < 0.001) and used in most analyses
(Figure 1).
Multiple regression models for 25(OH)D concentration
Only participants without missing data were included in
this analysis. There was no material difference in 25(OH)D
concentration or personal UVR dose (P > 0.05) between
participants with at least 1 missing variable (n = 117) and
those with no missing variables (n = 885); however, the former were older (mean = 53.29 (standard deviation, 15.93)
years vs. mean = 47.47 (standard deviation, 15.53) years,
P < 0.001). We removed from the model ambient UVR;
daily maximum temperature; self-reported skin, hair, and
eye color; melanin density for the cheek and upper arm;
vitamin D intake from oily fish; and sunscreen use by backwards elimination, because the model Akaike information
criterion was not reduced by their retention. Table 2 shows
the results of the final model with the variables grouped
into the following 3 broad categories: 1) environmental factors,
2) demographic and constitutional factors, and 3) potentially
modifiable behavioral factors.
Am J Epidemiol. 2014;179(7):864–874
The model explained 40.2% of the variance in serum
25(OH)D concentration (R 2 = 0.402, P < 0.0001). Collectively, potentially modifiable factors (BMI, physical activity,
vitamin D supplementation, time spent outdoors, clothing
cover, and personal UVR exposure) contributed to 52% of
the explained variance, whereas environmental factors (season, location, ambient UVR, and humidity) and demographic
and constitutional factors (sex, age, country of birth, employment status, occupation type, and melanin density on the
hand) contributed 38% and 10%, respectively. Clothing
cover was the single greatest contributor to the explained variance (27%), followed by location (20%) and season (17%).
In the final model, after controlling for all other variables,
a large seasonal difference between summer and winter
25(OH)D concentrations was noted (19.8 nmol/L, 95% confidence interval (CI): 14.6, 24.9). A location effect was also
observed, with the mean 25(OH)D concentrations in Brisbane,
Hobart, and Canberra between 7.7 nmol/L and 13.3 nmol/L
lower than the Townsville mean.
Oral vitamin D supplementation was associated with
higher 25(OH)D concentration (9.0 nmol/L, 95% CI: 6.4,
11.6), whereas obesity (BMI > 30 vs. BMI ≤ 25) was associated with lower 25(OH)D concentration (−8.0 nmol/L,
95% CI: −11.2, −4.8). On average, 25(OH)D concentration increased by 1.9 nmol/L (95% CI: 1.3, 2.3) for every
Study Sitea
Independent Variable
Intercept
Townsville
βb
95% CI
36.03
−9.30, 81.36
Brisbane
P Value
0.119
β
95% CI
46.05
−4.33, 96.43
Canberra
P Value
0.073
β
64.23
95% CI
29.51, 98.95
Hobart
P Value
<0.001
β
94.21
P Value
95% CI
36.50, 151.93
0.002
Season
Winter
Referent
Referent
Referent
Spring
NR
1.26
0.744
3.95
−2.11, 10.01
0.200
−2.99
Summer
NR
15.28
3.46, 27.10
0.012
17.02
5.38, 28.67
0.004
3.15
Autumn
1.48, 18.17
0.021
5.83
−0.14, 11.79
0.056
17.91
7.94, 27.88
0.001
0.55
0.05, 1.06
0.031
−0.43
−1.02, 0.16
NR
9.83
Ambient UVR (SED)
NR
NR
Daily max temperature (°C)
NR
Relative humidity (%)
0.73
−6.35, 8.87
Referent
NR
NR
0.27, 1.20
0.002
0.81
NR
0.23, 1.39
0.007
−13.87, 7.90
0.589
−15.25, 21.55
0.736
NR
NR
0.149
Sex
Female
Referent
−1.39
−6.89, 4.10
35–64
−6.11
65–75
−1.39
Male
Referent
0.618
0.04
−5.90, 5.98
−13.16, 0.94
0.089
−2.84
−10.10, 7.32
0.754
−4.97
Referent
0.990
−0.75
−5.48, 3.99
−9.81, 4.12
0.422
1.25
−14.34, 4.39
0.296
Referent
0.756
1.24
−4.69, 7.17
−4.50, 7.01
0.668
0.87
−6.47, 8.22
0.815
5.26
−2.46, 12.98
0.181
−2.60
−12.23, 7.03
0.595
−6.89
−12.04, −1.73
0.009
1.24
−4.69, 7.17
0.681
Age, years
18–34
Referent
Referent
Referent
Referent
Country of birth
Australia
Other countryc
Referent
Referent
NR
Referent
NR
Referent
0.681
Employment
Full-time
Not full-time
Referent
Referent
NR
9.26
3.45, 15.07
Referent
Am J Epidemiol. 2014;179(7):864–874
0.002
NR
0.125
NR
Referent
NR
Skin color
Fair
Medium to dark
Referent
Referent
−4.44
NR
−10.12, 1.25
Referent
Referent
5.95
−0.62, 12.52
0.075
Hair color
Blond or red
Brown or black
Referent
4.46
−1.44, 10.36
Referent
0.138
NR
Referent
NR
Referent
NR
Eye color
Blue, grey, or green
Hazel or brown
Referent
NR
Referent
NR
Referent
NR
Referent
NR
Table continues
870 Kimlin et al.
Table 3. Final Multiple Regression Models for the Determinants of 25(OH)D Concentration in Australian Adults in Each Study Location, AusD Study, Australia, 2009–2010
Am J Epidemiol. 2014;179(7):864–874
Table 3. Continued
Study Sitea
Independent Variable
Townsville
βb
95% CI
Brisbane
P Value
β
95% CI
Canberra
P Value
β
95% CI
Hobart
P Value
β
95% CI
P Value
Melanin density (%)
Inner upper arm
NR
NR
NR
Cheek
NR
NR
NR
Hand
NR
NR
4.63
0.91, 8.34
2.71
−0.66, 6.08
0.114
−3.18
−6.54, 0.18
0.063
0.015
NR
Body mass indexd
<25.0
Referent
Referent
Referent
Referent
25.0–29.9
−2.05
−8.78, 4.68
0.549
−2.22
−8.53, 4.08
0.488
−7.13
−12.53, −1.73
0.010
1.27
−5.51, 8.06
0.711
≥30.0
−6.99
−13.82, −0.17
0.045
−7.88
−15.57, −0.19
0.045
−8.72
−14.41, −3.04
0.003
−10.20
−17.33, −3.07
0.005
4.81
−1.93, 11.55
0.161
NR
5.01
−2.70, 12.72
13.53
6.38, 20.67
<0.001
NR
8.61
1.40, 15.83
0.020
13.46
7.40, 19.52
<0.001
Physical activity level
Inactive
Minimally active
Highly active
Referent
Referent
NR
7.94
Vitamin D intake from oily fish
(≥25 mg/month)
NR
NR
Sunscreen use (yes)
Clothing cover (%)
Personal UVR level (SEDe)
NR
NR
2.53, 13.35
0.004
−0.66, 0.00
0.050
−0.74
1.00
0.01, 2.00
0.047
1.94
9.90
5.18, 14.63
<0.001
NR
NR
−0.33
Referent
NR
NR
−1.09, −0.39
<0.001
−0.52
0.73, 3.15
0.002
1.33
0.201
NR
−0.89, −0.16
0.006
−0.44
0.40, 2.26
0.005
1.08
−0.86, −0.01
0.043
−0.29, 2.45
0.122
Abbreviations: AusD, A Quantitative Assessment of Solar UV Exposure for Vitamin D Synthesis in Australian Adults; CI, confidence interval; NR, not retained (in the final model); SED,
standard erythemal dose; UVR, ultraviolet radiation.
a
R 2 values are as follows: Townsville, 0.216; Brisbane, 0.373; Canberra, 0.386; and Hobart, 0.471.
b
Regression coefficient.
c
Numbers born in other countries are as follows: Townsville, n = 41 (15.8%); Brisbane, n =51 (20.1%); Canberra, n = 73 (29.0%); and Hobart, n = 31 (13.1%).
d
Weight (kg)/height (m)2.
e
1 SED = 100 J/m2.
Determinants of 25-Hydroxyvitamin D in Australia 871
Taking supplements (yes)
Referent
NR
872 Kimlin et al.
decile increase in personal UVR exposure and decreased by
0.5 nmol/L (95% CI: −0.7, −0.3) for every 1% increase in
clothing cover.
There was no evidence of multicollinearity in the final model
(for all independent variables, variance inflation factor < 2.3)
(25). However, both the residual plot and Breusch-Pagan
test strongly suggested the presence of heteroscedasticity (χ2 =
19.97, P < 0.001). A Bonferroni outlier test identified 3 possible outliers (all P < 0.007). The coefficients of all independent
variables were therefore retested using the White-Huber
corrected standard errors (24). Despite minor increases in the
P values, no material change was observed for any variable except for melanin density on the hand (P value changed from
0.040 to 0.052).
After the removal of outliers, the variance explained by the
final model increased from 40.2% to 41.1%. Separately, the
use of log-transformed 25(OH)D concentration instead of original values in the final model increased the R 2 value to 40.8%,
whereas substituting the ambient UVR level with raw ambient
UVR data or log-transformed ambient UVR data slightly decreased the R 2 values to 38.4% and 39.9%, respectively. The
number of variables retained in these alternative models and
their respective P values remained unchanged.
Separate stepwise models for each site are presented in
Table 3. The amount of variance in 25(OH)D concentration
explained by the model increased with increasing latitude
(for Townsville, 21.6%; for Brisbane, 37.3%; for Canberra,
38.6%; and for Hobart, 47.1%). For all locations, personal
UVR exposure was positively associated with 25(OH)D concentration, whereas clothing cover and obesity (BMI ≥ 30)
were negatively associated. There was no material effect of
sex at any location. There were marked differences in each
location in the relative importance of other variables retained
in the models. The seasonal effect was strong and similar in
all sites except tropical Townville. Country of birth was
strongly predictive of 25(OH)D concentration in Canberra,
and employment was retained only in the Brisbane model.
Physical activity was strongly predictive only in Townsville
and Hobart. The contribution of vitamin D supplementation
was highly influential to the model in all sites except
Townsville.
DISCUSSION
The AusD Study collected information on ambient UVR,
personal UVR exposure, dietary vitamin D intake, and environmental and personal characteristics to assess their roles
in determining 25(OH)D concentration. Our final model explained 40.2% of the variance in serum 25(OH)D concentration, with variation by location ranging from 47.1% in Hobart
(latitude, 42.8° south) to 21.6% in Townsville (latitude, 19.3°
south). Other studies investigating the determinants of
25(OH)D concentration have explained between 21% and
54% of the variance (26–34); however, these studies differed
widely by study population, chosen predictors, and measurement rigor. The common major contributors to explained
variance in these models were sun exposure–related factors
(26–30, 33), dietary intake (27–30, 34), season (32–34), use of
vitamin D supplements (28–30, 32–34), BMI or waist circumference (27, 28, 32, 34), and physical activity (27, 28, 32, 34).
The major contributors to explained variance in this study
were clothing cover (27%), location (20%), season (17%),
and personal UVR exposure measured by polysulphone dosimeters (8%). Potentially modifiable factors contributed to
52.2% of the explained variance, with key variables in our
model being clothing cover (contributing 27% to the explained variance in 25(OH)D), personal UVR exposure
(8%), vitamin D supplementation (7%), BMI (4%), and
physical activity (4%). The potential gain from clothing
cover is substantial; for every 10% decrease in clothing
cover, 25(OH)D concentration increased by 5.2 nmol/L without needing to increase the duration of sun exposure of habitually exposed body sites. Additional value would be obtained
by decreasing clothing cover during outdoor exercise in populations prone to 25(OH)D insufficiency, given the association between physical activity and 25(OH)D concentration.
Environmental variables were important contributors to
our model. Latitude of residence contributed 20% to the explained variance. There was a decrease in average 25(OH)D
concentration with increasing latitude—a finding supported
by several recent studies (12, 26, 35). Similarly, season has
been reported as an important predictor of 25(OH)D concentration (34, 36, 37).
Personal UVR exposure, usually reported as “time outdoors” or “time in the sun” (38), contributed only 8% to
the explained variance in our model—a much smaller contribution than that of other UVR-related variables. On average,
25(OH)D concentration increased by only 1.9 nmol/L for
every decile increase in personal UVR exposure. Thus, although personal UVR exposure and clothing cover are intrinsically associated, our results indicate that reducing clothing
cover would be a more effective way to increase serum
25(OH)D concentration than would increasing the duration
of sun exposure of habitually exposed skin.
The predictive ability of our model changed with increasing latitude, being greatest at high latitudes and least at low
latitudes. Apart from BMI, which was strongly negatively
correlated with 25(OH)D concentration across all sites, the
relative contribution of the variables differed by location.
Physical activity contributed strongly at the study site of lowest latitude, whereas season and clothing cover contributed
only weakly (35), possibly because of the light clothing
worn all year round and relatively constant high ambient
UVR levels. Vitamin D supplementation also contributed little at this location, as reported previously (35), whereas its
relative contribution became progressively stronger with increasing latitude. This finding was unrelated to the prevalence of vitamin D supplementation, which did not differ
markedly between the sites (for Townsville, 34.3%; for Brisbane, 39.4%; for Canberra, 35.9%; and for Hobart, 32.1%)
(data not shown). Because most vitamin D is produced
through skin exposure to UVR, it is perhaps not surprising
that the relative contribution of oral supplementation would
be greater at higher latitudes where solar UVR and vitamin
D levels are lower and clothing cover is higher.
Despite our comprehensive analysis, we could account for
only 40.2% of the variance in 25(OH)D concentration. Other
factors that probably contributed to the remaining variance include interactions between exposures, measurement error, and
variables not measured, including genetic variables (common
Am J Epidemiol. 2014;179(7):864–874
Determinants of 25-Hydroxyvitamin D in Australia 873
single nucleotide polymorphisms have been shown to contribute 4%–10% to variance in 25(OH)D (26, 39, 40)) and factors
as yet unknown. There is also evidence that sun exposure in the
6–8 weeks prior to blood collection is a strong predictor of
25(OH)D concentration (26, 41); hence, it is possible that confining our UVR measurement to the 10 days prior to blood
collection may have contributed to the unexplained variance.
Several factors identified in preliminary analysis as contributing to 25(OH)D concentration were not significant in
our final model, including dietary intake of vitamin D–rich
foods; sex; self-reported skin, hair, and eye color; and use
of sunscreen. Natural skin color (melanin density at the
upper inner arm) had only a minor effect on 25(OH)D concentration, possibly because of the small number of darkskinned participants in our study and the use of an algorithm
for melanin density (16) that was validated only in whites.
This study may have been limited by our categorization of
sunscreen use, for the purpose of analysis, as “yes” or “no,”
which did not allow consideration of sunscreen coverage or
sun protection factor. Other limitations were a low response
rate and a slight bias toward older, Australian-born women
and indoor workers (15). Although we do not expect these
biases to affect the association between the predictors and
serum 25(OH)D concentration, it is possible that both the exposure and the outcome may have influenced the probability
of participating, and hence the variability of the model between study locations and the generalizability of the findings.
For example, if older women were more aware of the importance of vitamin D and were thus more likely to take supplements, we may have overestimated the role of vitamin D
supplementation as a determinant of 25(OH)D concentration.
However, estimates of associations between variables appear
to be relatively unaffected by participation rates (42).
The study strengths were the size and breadth of its population samples, high compliance, broad range of latitudes,
cross-seasonal recruitment strategy, and coordination of protocols among the study sites (15). Our use of UVR dosimetry
combined with diary records (time spent outdoors and clothing cover) and contemporaneous data on ambient UVR is
probably the “gold standard” for assessing personal UVR exposure in noninstitutionalized populations.
Our findings suggest that modifiable factors relating to
UVR exposure could help in maintaining healthy vitamin
D status, with decreasing clothing cover being a more effective means than increasing duration of UVR exposure. We
also provide strong evidence for customizing approaches to
maintaining healthy vitamin D status according to season
and latitude. The role of such measures in ameliorating or
preventing mild to moderate levels of vitamin D insufficiency
when vitamin D supplementation is readily available is currently a matter of debate.
ACKNOWLEDGMENTS
Author affiliations: AusSun Research Laboratory, Institute
of Health and Biomedical Innovation, Queensland University
of Technology, Brisbane, Australia (Michael G. Kimlin,
Simone L. Harrison, Alison M. Brodie, Jiandong Sun); National Health and Medical Research Council Centre for
Am J Epidemiol. 2014;179(7):864–874
Research Excellence in Sun and Health, Institute of Health
and Biomedical Innovation, Queensland University of Technology, Brisbane, Australia (Michael G. Kimlin, Robyn
M. Lucas, David C. Whiteman, Jiandong Sun); National
Centre for Epidemiology and Population Health, College of
Medicine, Biology, and Environment, Australian National
University, Canberra, Australia (Robyn M. Lucas); Faculty
of Medicine, Health and Molecular Sciences, School of Public Health, Tropical Medicine, and Rehabilitation Sciences,
James Cook University, Townsville, Australia (Simone
L. Harrison, Madeleine Nowak); Menzies Research Institute,
Hobart, Australia (Ingrid van der Mei); Sydney School of
Public Health, University of Sydney, Sydney, Australia
(Bruce K. Armstrong, Anne Kricker); The Sax Institute, Sydney, Australia (Bruce K. Armstrong); and Population Health
Division, QIMR Berghofer Medical Research Institute, Brisbane, Australia (David C. Whiteman).
This work was supported by the National Health and Medical Research Council (grant 497220). M.G.K. was supported
through a Cancer Council Queensland senior research fellowship. R.M.L. was supported through a National Health and
Medical Research Council career development fellowship.
S.L.H. and M.N. received salary support from Queensland
Health. I.v.d.M. and D.C.W. were supported through Australian Research Council Future Fellowships. J.S. was supported
through a National Health and Medical Research Council
Centre for Research Excellence in Sun and Health postdoctoral fellowship.
We thank the local research officers for their work in collecting data for this project.
Conflict of interest: none declared.
REFERENCES
1. Nowson CA, McGrath JJ, Ebeling PR, et al. Vitamin D and
health in adults in Australia and New Zealand: a position
statement. Med J Aust. 2012;196(11):686–687.
2. Holick MF. Vitamin D deficiency. N Engl J Med. 2007;357(3):
266–281.
3. Correia LCL, Sodré F, Garcia G, et al. Relation of severe
deficiency of vitamin D to cardiovascular mortality during
acute coronary syndromes. Am J Cardiol. 2013;111(3):
324–327.
4. Khan H, Kunutsor S, Franco OH, et al. Vitamin D, type 2
diabetes and other metabolic outcomes: a systematic review and
meta-analysis of prospective studies. Proc Nutr Soc. 2013;
72(1):89–97.
5. Lucas RM, Ponsonby AL, Dear K, et al. Sun exposure and
vitamin D are independent risk factors for CNS demyelination.
Neurology. 2011;76(6):540–548.
6. Salzer J, Hallmans G, Nystrom M, et al. Vitamin D as a
protective factor in multiple sclerosis. Neurology. 2012;79(21):
2140–2145.
7. Dickens AP, Lang IA, Langa KM, et al. Vitamin D, cognitive
dysfunction and dementia in older adults. CNS Drugs. 2011;
25(8):629–639.
8. Schottker B, Haug U, Schomburg L, et al. Strong
associations of 25-hydroxyvitamin D concentrations with
all-cause, cardiovascular, cancer, and respiratory disease
mortality in a large cohort study. Am J Clin Nutr. 2013;97(4):
782–793.
874 Kimlin et al.
9. Gandini S, Boniol M, Haukka J, et al. Meta-analysis of
observational studies of serum 25-hydroxyvitamin D levels and
colorectal, breast and prostate cancer and colorectal adenoma.
Int J Cancer. 2011;128(6):1414–1424.
10. Kimlin M, Harrison S, Nowak M, et al. Does a high UV
environment ensure adequate vitamin D status? J Photochem
Photobiol B. 2007;89(2-3):139–147.
11. Lips P. Worldwide status of vitamin D nutrition. J Steroid
Biochem Mol Biol. 2010;121(1-2):297–300.
12. Daly RM, Gagnon C, Lu ZX, et al. Prevalence of vitamin D
deficiency and its determinants in Australian adults aged
25 years and older: a national, population-based study. Clin
Endocrinol (Oxf ). 2012;77(1):26–35.
13. Nowson CA, Margerison C. Vitamin D intake and vitamin D
status of Australians. Med J Aust. 2002;177(3):149–152.
14. Holick MF. Vitamin D: a millenium perspective. J Cell
Biochem. 2003;88(2):296–307.
15. Brodie AM, Lucas RM, Harrison SL, et al. The AusD Study: a
population-based study of the determinants of serum
25-hydroxyvitamin D concentration across a broad latitude
range. Am J Epidemiol. 2013;177(9):894–903.
16. Dwyer T, Blizzard L, Ashbolt R, et al. Cutaneous melanin
density of Caucasians measured by spectrophotometry and risk
of malignant melanoma, basal cell carcinoma, and squamous cell
carcinoma of the skin. Am J Epidemiol. 2002;155(7):614–621.
17. Craig CL, Marshall AL, Sjöström M, et al. International
physical activity questionnaire: 12-country reliability and
validity. Med Sci Sports Exerc. 2003;35(8):1381–1395.
18. Harrison SL. The Development of Melanocytic Naevi (Moles) in
Children Born and Raised in Tropical Australia [dissertation].
Townsville, Australia: James Cook University; 1999.
19. Harrison SL, Buettner PG, MacLennan R. The North
Queensland “Sun-Safe Clothing” Study: design and baseline
results of a randomized trial to determine the effectiveness of
sun-protective clothing in preventing melanocytic nevi. Am J
Epidemiol. 2005;161(6):536–545.
20. Pearl DK, Scott EL. The anatomical distribution of skin
cancers. Int J Epidemiol. 1986;15(4):502–506.
21. Harrison SL, Buettner PG, MacLennan R. Body-site
distribution of melanocytic nevi in young Australian children.
Arch Dermatol. 1999;135(1):47–52.
22. Sakamoto Y, Ishiguro M, Kitagawa G. Akaike Information
Criterion Statistics. Tokyo, Japan: KTK Scientific Publishers;
1986.
23. Relaimpo: relative importance of regressors in linear models.
http://cran.r-project.org/web/packages/relaimpo/. Accessed
June 13, 2013.
24. White H. A heteroskedasticity-consistent covariance matrix
estimator and a direct test for heteroskedasticity. Econometrica.
1980;48(4):817–838.
25. O’Brien RM. A caution regarding rules of thumb for variance
inflation factors. Qual Quant. 2007;41(5):673–690.
26. Lucas RM, Ponsonby AL, Dear K, et al. Vitamin D status:
multifactorial contribution of environment, genes and other
factors in healthy Australian adults across a latitude gradient.
J Steroid Biochem Mol Biol. 2013;136:300–308.
27. Millen AE, Wactawski-Wende J, Pettinger M, et al. Predictors
of serum 25-hydroxyvitamin D concentrations among
28.
29.
30.
31.
32.
33.
34.
35.
36.
37.
38.
39.
40.
41.
42.
postmenopausal women: the Women’s Health Initiative
Calcium Plus Vitamin D Clinical Trial. Am J Clin Nutr. 2010;
91(5):1324–1335.
Bertrand KA, Giovannucci E, Liu Y, et al. Determinants of
plasma 25-hydroxyvitamin D and development of prediction
models in three US cohorts. Br J Nutr. 2012;108(10):
1889–1896.
Brustad M, Alsaker E, Engelsen O, et al. Vitamin D status of
middle-aged women at 65–71°N in relation to dietary intake
and exposure to ultraviolet radiation. Public Health Nutr. 2004;
7(2):327–335.
van der Meer IM, Boeke AJP, Lips P, et al. Fatty fish and
supplements are the greatest modifiable contributors to
the serum 25-hydroxyvitamin D concentration in a
multiethnic population. Clin Endocrinol (Oxf ). 2008;68(3):
466–472.
Barth J, Gerlach B, Knuschke P, et al. Serum 25(OH)D3 and
ultraviolet exposure of residents in an old people’s home in
Germany. Photodermatol Photoimmunol Photomed. 1992–
1993;9(5):229–231.
Freedman DM, Cahoon EK, Rajaraman P, et al. Sunlight and
other determinants of circulating 25-hydroxyvitamin D levels in
black and white participants in a nationwide US study. Am J
Epidemiol. 2013;177(2):180–192.
Hedlund L, Brembeck P, Olausson H. Determinants of vitamin
D status in fair-skinned women of childbearing age at northern
latitudes. PLoS One. 2013;8(4):e60864.
McCullough ML, Weinstein SJ, Freedman DM, et al. Correlates
of circulating 25-hydroxyvitamin D: Cohort Consortium
Vitamin D Pooling Project of Rarer Cancers. Am J Epidemiol.
2010;172(1):21–35.
Nowak M, Harrison SL, Buettner PG, et al. Vitamin D status of
adults from tropical Australia determined using two different
laboratory assays: implications for public health messages.
Photochem Photobiol. 2011;87(4):935–943.
Greene-Finestone LS, Berger C, de Groh M, et al.
25-Hydroxyvitamin D in Canadian adults: biological,
environmental, and behavioral correlates. Osteoporos Int. 2011;
22(5):1389–1399.
Janssen HCJP, Emmelot-Vonk MH, Verhaar HJJ, et al.
Determinants of vitamin D status in healthy men and women
aged 40–80 years. Maturitas. 2013;74(1):79–83.
Whiteman DC, Stickley M, Watt P, et al. Anatomic site, sun
exposure, and risk of cutaneous melanoma. J Clin Oncol. 2006;
24(19):3172–3177.
Hiraki LT, Major JM, Chen C, et al. Exploring the genetic
architecture of circulating 25‐hydroxyvitamin D. Genet
Epidemiol. 2013;37(1):92–98.
Wang TJ, Zhang F, Richards JB, et al. Common genetic
determinants of vitamin D insufficiency: a genome-wide
association study. Lancet. 2010;376(9736):180–188.
Nair-Shalliker V, Clements M, Fenech M, et al. Personal sun
exposure and serum 25-hydroxy vitamin D concentrations.
Photochem Photobiol. 2013;89(1):208–214.
Mealing NM, Banks E, Jorm LR, et al. Investigation of relative
risk estimates from studies of the same population with
contrasting response rates and designs. BMC Med Res
Methodol. 2010;10:26.
Am J Epidemiol. 2014;179(7):864–874