Six hundred sixty-four subjects were enrolled. Thirty subjects were excluded because race/ethnicity data were not recorded, leaving a final cohort of 634 subjects. The demographic makeup of the cohort is described in Table 1. Mean age was 29 ± 8 years (range, 18–48 years).
The mean 25(OH)D level for the cohort was 27 ± 14 ng/mL. Figure 1 shows the median and interquartile ranges of 25(OH)D levels for specified subgroups. 25(OH)D levels varied with sex, race, season, and multivitamin use. Females had significantly higher 25(OH)D levels compared with males (30 ± 15 versus 24 ± 12 ng/mL, P<0.001). White subjects had significantly higher 25(OH)D levels compared with black, Asian, and "other" race subjects (30 ± 14, 17 ± 10, 20 ± 11, and 22 ± 10 ng/mL respectively; P<0.001 for each pairwise comparison). After stratification by race, there were no significant differences between Hispanic and non-Hispanic subjects (Table 2). The 25(OH)D levels were higher in summer (31 ± 14 ng/mL) than in fall (27 ± 13 ng/mL), winter (25 ± 16 ng/mL), and spring (24 ± 12 ng/mL) (P = 0.03, P = 0.002, and P<0.001, respectively). Multivitamin users had higher 25(OH)D levels in comparison with nonusers (35 ± 15 versus 26 ± 14 ng/mL; P<0.001).
Box plot of 25-hydroxyvitamin D [25(OH)D] levels in different subgroups. The center band within each box indicates the median; the boundaries of the box indicate the 25th and 75th percentiles; whiskers represent minimum and maximum values less than 1.5 times the interquartile range; outliers are indicated by circles. A = Asian; B = black; F = female; M = male; MVI = multivitamin; O = other; Spr = spring; Sum = summer; W = white; Win = winter. *P<.05, †P<0.01, and ‡P<0.001. To convert 25(OH)D to nmol/L, multiply ng/mL by 2.496.
Seasonal Variation, Race, and Multivitamin Use
Figure 2A shows the mean 25(OH)D levels by season after stratification by race. Seasonal variation of 25(OH)D levels was present in each racial group (P = 0.45 for the interaction of season and race by two-way ANOVA). This seasonal variation, however, was quantitatively less among black subjects when compared with white and Asian subjects. For example, the difference between spring and summer 25(OH)D levels was 8 ng/mL (95% confidence interval [CI]: 5–11 ng/mL) among white subjects, 11 ng/mL (95% CI: 6–17 ng/mL) among Asian subjects, and 2 ng/mL (95% CI: 5–9 ng/mL) among black subjects.
Mean 25-hydroxyvitamin D [25(OH)D] ± standard error of the mean (SEM) by season stratified by race (panel A) and by multivitamin (MVI) use (panel B). (†P = 0.01 and ‡P<0.001 for difference between the no multivitamin and multivitamin groups at the indicated season after Bonferroni correction for multiple comparisons). To convert 25(OH)D to nmol/L, multiply ng/mL by 2.496.
Figure 2B shows the mean 25(OH)D levels by season after stratification by multivitamin use. There was a significant effect of multivitamin use on seasonal variation (P = 0.01 for the interaction of season and multivitamin use by two-way ANOVA). Specifically, there was significant seasonal variation in 25(OH)D levels among subjects who did not use multivitamins (P = 0.001 by one-way ANOVA), but seasonal variation was not observed among multivitamin users. In pair-wise comparisons, multivitamin users had higher 25(OH)D levels in winter and spring in comparison with nonusers (P<0.001 and P = 0.01, respectively). Multivitamin users also had quantitatively higher 25(OH)D levels than nonusers in summer and fall, though these differences were not statistically significant.
Prevalence of Vitamin D Deficiency
Table 3 shows the prevalence of vitamin D deficiency defined by three different thresholds of 25(OH)D: ≤ 10, ≤20, and ≤30 ng/mL.[12,17] Consistent with the mean 25(OH)D data, the prevalence of vitamin D deficiency among male subjects was higher than among female subjects at each threshold (P = 0.05, P<0.001, and P<0.001, respectively). Similarly, the prevalence of vitamin D deficiency differed by race and by season for each threshold (P<0.001 at all thresholds for the effect of race and for the effect of season). The prevalence of vitamin D deficiency was lower among multivitamin users when compared to nonusers at each threshold (P = 0.002, P<0.001, and P<0.001, respectively).
25(OH)D and PTH
As shown in Figure 3A, lower 25(OH)D levels were associated with higher levels of PTH (r = −0.28; P<0.001). As shown in Figure 3B, when stratified by the 25(OH)D thresholds described above, mean PTH levels decreased from 48 ± 14 pg/mL for 25(OH)D ≤10 ng/mL to 36 ± 16 pg/mL for 25(OH)D >30 ng/mL (P for ANOVA <0.001). As shown in Figure 3C, the prevalence of hyperparathyroidism (PTH >60 pg/mL) decreased from 25% to 3% in the groups with the lowest to highest 25(OH)D levels, respectively (P for ANOVA = 0.009).
A, Scatterplot and regression line of 25-hydroxyvitamin D [25(OH)D] with parathyroid hormone (PTH). B, Mean PTH ± standard error of the mean (SEM) by 25(OH)D at the thresholds indicated. C, Percentage of subjects at each 25(OH)D threshold with PTH ≤60 pg/mL (light gray) and with PTH >60 pg/mL (dark gray).
We developed a linear regression model for 25(OH)D levels using sex, age, self-identified race, self-identified ethnicity, season, and multivitamin use as potential predictors. After backwards step-wise selection, all variables except self-identified ethnicity (P = 0.22 in the initial model) were found to be independent clinical predictors. We randomly divided our sample into a training set (75% of the cohort) and a validation set (25% of the cohort), and performed logistic regression analysis using these variables in the training set to predict vitamin D deficiency at 2 thresholds: 25(OH)D ≤20 and ≤30 ng/mL (Table 4). Goodness-of-fit, assessed by the Hosmer-Lemeshow test, showed that the models fit the data well [P = 0.29 for the 25(OH)D ≤20 ng/mL model and P = 0.91 for the 25(OH)D ≤30 ng/mL model, indicating that the number of deficient subjects observed is not significantly different from that predicted]. Using our model, areas under the receiver operating characteristic (ROC) curves in the validation set were 0.76 and 0.80 for the 25(OH)D ≤20 ng/mL and 25(OH)D ≤30 ng/mL models, respectively, indicating good discrimination of 25(OH)D deficient versus sufficient subjects.
Table 5 shows the predicted probability of vitamin D deficiency (defined as 25(OH)D ≤20 ng/mL) in hypothetical 30-year-old patients of different race, sex, and multivitamin use status, if tested in the summer [highest predicted 25(OH)D level] or in the spring [lowest predicted 25(OH)D level]. In this model, the probability of vitamin D deficiency ranges from 5% (for a white female, taking a multivitamin, tested in the summer) to 87% (for a black or Asian male, not taking a multivitamin, tested in the spring).
Endocr Pract. 2012;18(6):914-923. © 2012 American Association of Clinical Endocrinologists