The present study is a cross-sectional survey of 226 women in early pregnancy recruited from five antenatal clinics in primary care in Umeå, Sweden. Participants, who were consecutively recruited between September 2006 and March 2009, were part of a longitudinal study (PregNut) where dietary intake, height, and weight were measured and blood sampled during pregnancy and postpartum. One hundred and forty-three of the women were nullipara, 83 multipara, 222 had a singleton pregnancy, and four had a twin pregnancy. All these women but ten were born in Sweden, 223 were married or cohabiting, and three were single. All women attending the antenatal clinics were invited by midwives, and women who expressed an interest in participating were given verbal and written information. Signed consent was obtained during the first visit to the antenatal welfare program. Three exclusion criteria were used: major medical conditions, unable to attend the ordinary antenatal welfare program, and insufficient competence in the Swedish language.
The pregnant women were asked to answer a 66-item food frequency questionnaire (FFQ). The FFQ is a shortened version of the original Northern Sweden FFQ, but the questions used were the same as the original version. Both the original and shortened versions were designed to be semi-quantitative and optically readable for data input. The Northern Sweden FFQ is used by the Västerbotten Intervention Programme (VIP) , the European Prospective Investigation into Cancer and Nutrition (EPIC) , and the Northern Sweden WHO Multinational Monitoring of Trends and Determinants in Cardiovascular Disease (MONICA) . The original Northern Sweden FFQ has been validated against ten repeated 24 h dietary recalls and selected plasma or erythrocyte biomarkers with regard to intake of food, energy, and macronutrients, vitamins, minerals, and fatty acids ,,. The correlation coefficients for the two recording methods were typically between 0.45 and 0.61, and the median correlation coefficient for all nutrients was 0.50. Consumption frequencies were reported on a nine-level scale, from 0 (never) to 8 (4 times/day or more). The shortened version included eight questions on the frequency of consumption of various types of fats used for spreading on bread or cooking, nine on milk and other dairy products, seven on bread and cereals, six on fruit, greens and root vegetables, and six on soft drinks and sugar-containing snacks. Five questions on spirits, wine and beer consumption were included in a list of beverages. Twenty of the remaining 25 questions recorded, intake of potato, rice, pasta, meat and fish, and five were on varied items, such as salty snacks, coffee, tea and water. The participants indicated their average portion of (i) potato/pasta/rice, (ii) vegetables, and (iii) meat/fish using four colour photographs illustrating four plates with increasing portion sizes of potatoes, vegetables, and meat. For other food items, either gender and age portion sizes or standard portion sizes were used as described previously . The reported consumption frequencies were converted to number of intakes per day. The content of energy and nutrients was calculated by multiplying daily intake frequency by the portion content according to the latest available update for the specific nutrient in the database provided by the National Food Administration (Uppsala, Sweden) . In general the shorter version yields lower total energy intake reports but rank subjects in the same order .
Five pregnant women did not answer the FFQ and 12 more had either (i) left ≥10% food questions unanswered or (ii) lacked at least one portion size indication or (iii) had an estimated food intake level (FIL, reported energy intake/basal metabolic rate) corresponding to the lowest 5% and highest 2.5% as described earlier . These women were excluded, leaving 209 pregnant women for diet intake evaluation. For these women, mean (99% CI) number of days of pregnancy at time of examination was 85 days (82-87 days).
The participants also completed a questionnaire, which in addition to diet intake, covered socioeconomic and psychosocial conditions, marital status, level of education, self-rated health, personal health history, family history, and quality of life . The questionnaire also covered social network and support, working conditions, physical activity, alcohol consumption, tobacco use, and dietary supplement use. For supplements, intake of multimineral, multivitamin, and iron supplement was reported as "Yes" or "No" for intake during the previous two weeks. In addition, body weight (light clothing) and height (no shoes) were measured.
A group of referent women was nested in the current Västerbotten Intervention Program (VIP) . VIP invites all 40-, 50-, and 60-year-old inhabitants in Västerbotten County to a health screening. In some communities, 30-year olds are also invited. The Northern Sweden Diet Database (NSDD) compiles diet data for the VIP, producing approximately 140 000 observations . All 30-year-old women from the larger Umeå area, who had participated in VIP during the same recruitment period as the pregnant women in this study, were included as referents (n = 108). Of these, five women who did not fulfil the FFQ quality criteria described above were excluded. An equally sized group of 40-year-old women (n = 103) in the NSDD with FFQs fulfilling the quality criteria – being from the Umeå area and attending VIP in the same period as the pregnant women in this study – was randomly selected as referents for 40-year-old pregnant women. The referent women also had their height and weight measured during their VIP and completed the same questionnaire as the pregnant women. Thus, information for the referents and pregnant women was obtained using an identical questionnaire and virtually identical routines. The only difference was that the referent women completed the FFQ from the perspective of intake the previous year but not just the previous two weeks.
Data handling and statistical analysis
As a basis for data analysis in the present project, the distribution of reported energy and nutrient intake was evaluated among all 30- and 40-year-old women in NSDD with a diet recording fulfilling FFQ quality criteria and with a screening date within the same period as recruitment of the pregnant women in this study (n = 26 394). The distribution was found to be acceptably normal for all diet variables, except alcohol. Thus, for descriptions, means with 95% confidence interval (CI) are presented for dietary variables, except alcohol intake where median with max-min values are presented. Intake of total fats, carbohydrates, proteins, and alcohol are presented as the proportion energy they provide in per cent of the total reported energy intake (E%). To compensate for the systematic underreporting by the shortened FFQ, reported intakes were extrapolated to an energy level corresponding to the 25% reduction of the original FFQ.
Differences between groups were tested with Student's t-test for normally distributed variables after appropriate adjustments. Mann-Whitney U test was used to test the difference in alcohol intake. In addition to testing for differences in estimated amounts eaten per day, differences in residuals from regressions of the respective nutrient on energy intake was assessed for vitamins and minerals as described by Willet . All residuals were normally distributed. The use of residuals was done to circumvent potential errors from underreporting or over-reporting. In the present population, underreporting has been found common due to the instrument, high BMI, low education, and smoking . Accordingly, for ten-year age groups, means of nutrient intakes were calculated by standardizing for BMI groups, education level, and smoking. When comparing all pregnant and referent women, age group was also included as a covariate.
Participants were classified as normal weight (BMI <25), overweight (BMI ≥25 - <30), or obese (BMI ≥30). No pregnant or referent women had a BMI <18. Use of tobacco (smoking or Swedish snus (snuff)) and alcohol was dichotomized into present-use or no-use. Education was dichotomized into having a university education or not. Physical activity was dichotomized as having a low physical activity at work, leisure-time, both or neither. Differences in the distributions of these variables and proportions of obese, overweight, and normal-weight subjects among pregnant versus referent women were tested with a Chi-square test, unless the number in a cell was five or lower, in which case Fisher's exact test was used.
Use of a supplement was dichotomized into taking a supplement the previous two weeks or not. Vitamin and mineral supplementation was estimated by using the most frequent content in over-the-counter preparations aimed for women. Thus, the following additions were made: calcium, 200 mg; iron, 18 mg; vitamin D, 7.5 μg; vitamin B12, 2.5 μg; and folate, 200 μg. The latter was adjusted for increased bioavailability from folic acid by the factor 1.7 . P-values <0.01 were considered statistically significant. IBM SPSS Statistics version 20 was used for these data analyses.
Multivariate partial least square modelling (PLS) was performed to search for clustering among the women and to identify variables associated with being early pregnant or a referent woman. The software SIMCA P+, version 12.0 (Umetrics AB, Umeå, Sweden) was used. The independent (X) block included the 66 foods/aggregates in the FFQ, estimated nutrients, supplement use, and tobacco use. Variables were autoscaled to unit variance, and cross-validated prediction of Y was calculated. Clustering of participants was displayed in a score loading plot with the two strongest components (t and t) on the x- and y-axis ,. For the theoretical concept behind PLS modelling we refer to the review paper by Haenlein and Kaplan .
The study was approved by the Regional Ethical Review Board at Umeå University Sweden (Dnr 04-171 M). The clinics involved are all part of Västerbotten County Council and under the evaluation by Regional Ethical Review Board at Umeå University. Thus, the given ethical approval from the Regional Ethical Review Board at Umeå University Sweden includes all health centers.