Vitamin D status and its determinants in healthy pregnant women living in Switzerland in the first trimester of pregnancy

The study design was described elsewhere [20]. In summary, pregnant women in their first trimester of pregnancy (between week 6 and 12 of pregnancy) were conveniently recruited between September 2014 through December 2015 while attending their first routine antenatal care appointment at the Clinic of Obstetrics of the University Hospital Zurich (USZ). Inclusion criteria were pregnancy, plans to deliver at the Clinic of Obstetrics of the USZ, minimum age of 18 years, current residence in Switzerland for at least 6 months before the start of pregnancy, and fluency in German, French, Italian or English. Exclusion criteria were multiple pregnancies, HIV infection, history of parathyroid, renal or liver diseases, chronic malabsorption syndromes or granuloma-forming disorders, age below 17 years or known (or suspected) drug or alcohol abuse. For eligible participants, an informed consent was obtained by the physician. A 10 mL blood sample was drawn during routine blood collection. Study participants and the responsible physician completed together a questionnaire (see Additional file 1) aiming at gathering socio-demographic information as well as data related to the pregnancy and lifestyle. Our final sample consisted of 204 women. The study was not representative of the whole Swiss population, but the choice of the USZ was driven by the heterogeneity in the socio-demographic and cultural background of women attending this hospital. In general, the city of Zurich has a large foreign population (32%) [21]. A true sample size calculation was not available in practice, because the coefficients of determination between covariates could not be determined. However, a minimum of ten participants per covariate were included in the analysis, as suggested by Agresti [22] for multivariable logistic regressions.

Approval (KEK-ZH-Nr. 2013–0213) was provided by the cantonal ethics committee of Zurich, Switzerland.

The Institute of Clinical Chemistry of the University Hospital Zurich analysed all blood samples within hours following collection: after centrifugation and serum extraction, total 25-hydroxyvitamin D (25(OH)D) was analyzed with the vitamin D total-analysis Roche Cobas® electro-chemo-luminescence immunoassay (Roche Diagnostics, Basel, Switzerland; detection range: 3.0–70.0 ng/mL for 25(OH)D; above 15 ng/mL, inter-assay coefficient of variation: 11.5% and intra-assay coefficient: 6.5% [23]). Women with serum 25(OH)D concentrations strictly below 20 ng/mL were considered as vitamin D deficient, as recommended by the Endocrine Society [24]. Women with serum 25(OH)D concentrations above 20 ng/mL were considered as non-deficient (which includes both insufficient (20 to 30 ng/mL) and sufficient (above 30 ng/mL) as defined by the Endocrine Society [24]).

The variables and potential determinants of vitamin D status were retrieved from the questionnaire (Additional file 1) and medical records. In summary, we collected age, week of pregnancy, parity (nulliparous yes, no), gravidity or first pregnancy (yes, no), Body Mass Index (BMI) before pregnancy (self-reported), BMI at enrolment (self-reported, mentioned in tables as BMI current), skin colour type, country of origin categorized into five groups, education level of the mother and of the partner in line with the International Standard Classification of Education [20] (less than compulsory education and compulsory education; secondary education; tertiary education), smoking status (never, ever, current), season at enrolment (winter [December 21st – March 20th], spring [March 21st – June 20th], summer [June 21st – September 20th], autumn [September 21st – December 20th]), sun exposure as the average number of days per week spent at least 1 h outdoor between 10 am and 4 pm in the past 6 months, use of sun protection (such as use of sunscreen, wearing long sleeves, trousers, hat (never, sometimes, always), fish consumption (herring, salmon, mackerel, sardine or tuna) at least once a week (yes, no) and intake of vitamin supplements containing vitamin D (yes, no). Vitamin D containing supplements were either pure vitamin D supplements or multivitamin supplements containing vitamin D. Dosage, adherence to treatment and start of intake were not recorded. We grouped countries of origin in five categories based on the regions defined by the world National Bank and previously reported [20] (group 1: Switzerland and Germany; group 2: Northern America, Northern Europe, Central Asia and New-Zealand; group 3: Southern Europe, Australia and Latin America; group 4: South-East Asia Pacific; group 5: Africa and Middle East). We have previously shown that dark skin colour was associated with a higher prevalence of vitamin D deficiency [20]. Eventually, we decided to assess vitamin D concentrations under the angle of the country of origin and compare the differences between subgroups. Country of origin is informative and important in clinical practice as it encompasses aspects such cultural behaviours, physical activity, eating habits and attitude towards sun exposure in addition to skin colour. We then proceeded to identify the significant determinants of vitamin D deficiency in early pregnancy and finally established a prospective prediction model. Several models were compared with the objective to identify the best compromise between statistical validity, ease of data collection and practicality in a clinical environment.

Women’s skin colour was assessed according to the classification by Fitzpatrick [25] using a five-level scale [20]. The classification into skin types was assessed in two steps. In a first step, study participants were shown a picture of the different skin colour types (I-V) and rated their own type. The same evaluation was done in parallel by the physician. In a second step, women were asked to classify in one of five predefined categories what happens to their untanned skin when exposed to the sun under specific conditions [20]. Based on their skin colour type and skin tanning evaluations, women estimated their own skin phototypes. When the classification of a pregnant woman and the interviewer disagreed, the upper rounded arithmetic mean was used to determine the skin colour type.

Additionally, we defined a melanin index [26] on the basis of measurements conducted with a narrow band spectroscopy instrument (DSM II ColourMeter, Cortex Technology, Hadsund, Denmark; green diode 568 nm, red diode 655 nm). Measurements were repeated 3 times on the inner underarm and the arithmetic mean was calculated. The instrument was calibrated on a weekly basis using a white balance.

Statistical analyses and graphing were carried out using R version 3.3.2 for Windows. Boxplots represent the medians and first and third quartiles. Prior to starting statistical analyses, categorical variables, for which the size of certain subsets was insufficient to guarantee the strength of statistical testing, were transformed and levels combined when deemed necessary. Skin colour phototypes were dichotomized into light skin colour (Fitzpatrick scale I to III) vs. dark skin colour (Fitzpatrick scale IV and V). For the variables education level of the mother and education level of the partner, individuals with “less than compulsory education” and “compulsory education” were grouped into one level (level 1). Finally, melanin index was taken into account both as a continuous variable and split into tertiles.

Univariable logistic regression analyses were performed with vitamin D status (deficient versus non-deficient) as dependent variable to estimate associations between vitamin D deficiency and following potential determinants: age, week of pregnancy, nulliparity, gravidity, BMI before pregnancy, BMI at enrolment, skin colour, country of origin, education level achieved by the mother and by the partner, smoking status, season, sun exposure, use of sun protection, fish consumption and intake of vitamin D-containing supplements. Multivariable logistic regressions with vitamin D status (deficient versus non-deficient) as dependent variable were subsequently performed. The Akaike information criterion (AIC) score and the area under the curve of the receiver operating characteristics (c-index) and likelihood ratio test Chi-square test (LRT) were used for selecting the final model. Calibration curves were realized to display how much the predictive values calculated with our models differed from the observed values. These curves were obtained by binning our data by halves: data was first split in upper and lower half, these halves were in their turn split in halves and finally, extreme halves were split recursively. Both predictive and observed values were binned and averaged within each bin. Dots sizes are proportional to the number of observations within a bin.

Model 1, also referred to as full model, included all probable determinants, excluding variables highly associated with each other. Associations between variables were tested using Cramér’s V for categorical and Pearson’s correlation coefficient for continuous variables. Model 2 was the most parsimonious model. Finally, model 3 was determined using backward-forward AIC selection principles. For all models, the measure of association was the odds ratio (OR) and its corresponding 95% confidence interval (95% CI). P value threshold was set at p < 0.05. The goodness of fit and best prediction power assessments were performed by comparing the c-statistics of all three models.

Among the 204 pregnant women in their first trimester participating in the study (Table 1), 22.1% originated from Switzerland or Germany (group 1), 34.3% from North America, Northern Europe, Caucasus, Central Asia and New Zealand, excluding Germany and Switzerland (group 2), 13.7% from Southern Europe, Australia, Latin America and the Caribbean (group 3), 12.7% from South East Asia Pacific (group 4) and 17.2% from Africa and Middle East (group 5). The median vitamin D concentration in the overall sample was 17.1 ng/mL with 63.2% of participating women being vitamin D deficient. The highest proportions of vitamin D deficiency were detected in group 4 (88.5% of deficiency, median vitamin D concentration of 8.4 ng/mL) and group 5 (91.4% of deficiency, median vitamin D concentration of 10.7 ng/mL). The overall mean age at blood collection was 30 years, with women in group 1 being the oldest. All women in group 1 were fair skinned, whereas more than 65% of women in groups 4 and 5 were dark skinned. The highest proportion of pregnant women with an education level less than compulsory and compulsory was observed in group 5 (54.3%). More than half of the women in the overall sample had never smoked with the highest proportion of never and former smokers seen in groups 4 and 5. 75% of women ate fish more than once a week, which was similar among the different subsets. Finally, 37.8% of women used vitamin D supplements, with the highest proportion in group 3 (53.6%).

Results obtained through univariable logistic regression analyses (Table 2) revealed that risk factors for vitamin D deficiency were higher BMI before pregnancy and at enrolment, higher melanin index, dark skin colour, and country of origin (women originating from country groups 2, 4, and 5 had higher odds of deficiency than group 1). Conversely, older age, education level of both the mother and her partner (the higher the education level the lower the odds of deficiency), smoking status (former smokers had a lower odds of low vitamin D circulating level), use of sun protection (women never or sometimes using sun protection suffered less from vitamin D deficiency) and the intake of vitamin D-containing supplements, had a protective effect.

Before proceeding to multivariable logistic regressions, we tested the collinearity between statistically significant variables. The following pairs of variables were considered as correlated: gravidity and nulliparity (Cramér’s V = 0.79), education of the mother and education of the partner (Cramér’s V = 0.56), skin colour and country of origin (Cramér’s V = 0.64), and finally BMI before pregnancy and BMI at enrolment (Pearson’s correlation coefficient = 0.98). Consequently, a decision was made to exclude the variables nulliparity, education of the partner, skin colour and BMI before pregnancy from further analyses as we considered them the least informative. As a result, the first model analyzed (model 1) included age, week of pregnancy, BMI at enrolment, gravidity, country of origin, education of the mother, smoking status, season, days spent in the sun, use of sun protection, fish consumption and use of vitamin D containing supplements.

Compared to the univariable logistic regressions results, country of origin, smoking status and use of vitamin D-containing supplements remained significant determinants of vitamin D deficiency, and BMI at enrolment remained close to significance (Model 1; Table 2). All other variables did not reach the significance level (p < 0.05). Concentrations of 25(OH)D across levels of the four statistically significant (or close to significant) determinants of vitamin D deficiency are displayed in Fig. 1. In Model 2, country of origin, BMI at enrolment, smoking status and use of vitamin D-containing supplements best explained vitamin D deficiency within this Swiss cohort. Model 3 was the most parsimonious model. Model 2, with an AUC of 0.796, a good calibration and an AIC value of 220.09 corresponded to the most easily adaptable model for practical prediction of Vitamin D deficiency (Fig. 2 and Table 2). Though it could be argued that model 3 as the most parsimonious model would better serve this purpose, LRT Chi-square tests between model 2 and 3 were performed. Results indicated that model 2 fits the data significantly better than model 3 (LRχ²: 10.46, DF: 3, p = 0.015) recommending the former as the best model to predict vitamin D deficiency in practice. While Receiver Operator Characteristic (ROC) curves supported the discrimination power assessment of the models, we used calibration curves to display the degree of accuracy of predictive values obtained with our models vs. observed values (Fig. 2).