Associations between vitamin D levels and polycystic ovary syndrome phenotypes.

Between January 2008 and June 2012, we identified 168 cases with PCOS and 146 controls with male factor infertility issues undergoing their first IUI cycle at the study site. After excluding women without available serum (22 cases and 43 controls) or without records available for diagnostic confirmation (9 cases), the study population consisted of 240 women, with 137 cases of confirmed PCOS and 103 controls seeking IUI for male factor infertility. Covariate distributions for cases and controls are reported in Table I . The study population was predominately non-Hispanic white (85%), with no racial/ethnic differences observed among the cases and controls. Compared to controls, cases were younger, had higher BMI, and were also more likely to experience infertility for two or more years. Current exercise, alcohol use, and smoking status were similar among the cases and controls. In terms of factors related to the serum collection and storage, season of serum collection and year of IUI treatment did not differ by case status. Women with PCOS had lower median levels of total vitamin D (26 ng/mL, range: 6–70) when compared to controls (30 ng/mL, range: 11–63) However, this difference was not statistically significant (p= 0.07). Concentrations of 25OHD 3 were similar to total vitamin D levels with median concentrations of 25 ng/mL (range: 2–70) for cases and 30 ng/mL (range: 11–63) for controls (p-value=0.13). Approximately 96% of the study population had 25OHD 2 levels below the levels of detection (LOD) (4 ng/mL) (n=230). Therefore, 25OHD 2 levels were not evaluated further. Likewise, since results evaluating 25OHD 3 levels were nearly identical to results analyzed using total vitamin D concentrations, further study results are reported using only total 25OHD levels. A higher proportion of vitamin D deficiency was observed among the cases although it did not reach statistical significance (21.2% vs. 13.6%, p-value=0.13) ( Table II ). Additionally, 59.9% of the cases lacked sufficient vitamin D levels (i.e. combining insufficient and deficient categories) compared to 47.6% of the controls (p-value=0.06). Using vitamin D sufficiency as the referent, the unadjusted odds of vitamin D deficiency were 2.03 (95% CI: 0.97, 4.26) times higher in women with PCOS compared to controls ( Table II ). Significant associations with vitamin D insufficiency were not observed but the point estimates were in the same direction ( unadj POR=1.49, 95% CI: 0.84, 2.63). Adjusting for BMI and race/ethnicity attenuated the results ( Table II ). A pattern of increasing odds ratios with decreasing vitamin D levels was noted; however the confidence intervals contained the null value. Associations between vitamin D exposure and PCOS phenotypes are reported in Table III . Vitamin D deficiency and insufficiency were positively associated with phenotype 3 (ovulatory dysfunction, polycystic appearing ovaries, and androgen excess) in the unadjusted analyses, but associations were attenuated after adjusting for BMI and race/ethnicity. Vitamin D insufficiency, but not deficiency, was associated with an increased odds of phenotype 2 (ovulatory dysfunction and androgen excess) in the unadjusted analyses, but again the odds ratio was attenuated after controlling for BMI and race/ethnicity. The precision of associations with phenotype 2, however, was limited by the small number of cases with this phenotype (n=15). When phenotypes 2 and 3 were combined to examine all 82 cases with androgen excess, crude associations with both insufficient ( unadj POR=2.35, 95% CI: 1.21, 4.54) and deficient ( unadj POR=2.93, 95% CI: 1.27, 6.77) vitamin D levels were observed. However, the magnitude of association decreased and the confidence intervals included 1.0 after adjusting for BMI and race/ethnicity (Insufficient: adj POR=1.72, 95% CI: 0.84, 3.50; Deficient: adj POR=2.03, 95% CI: 0.79, 5.19). In contrast, no associations with vitamin D deficiency or insufficiency were observed for phenotype 1 (ovulatory dysfunction and polycystic appearing ovaries).

We conducted a prevalence case-control study among women undergoing fertility treatment at a university-based fertility clinic. We identified women with PCOS (cases) and non-PCOS ovulatory women with male factor infertility (controls) who underwent an IUI from January 2008 to June 2012. Subjects were ineligible if records were unavailable or incomplete, frozen serum prior to IUI treatment was unavailable, or the patients had other female infertility diagnoses (e.g., endometriosis, tubal issues, unexplained infertility, fibroids, recurrent miscarriage) or underlying known medical conditions identified as affecting vitamin D levels and absorption (e.g., anti-phospholipid antibody syndrome, hyperprolactinemia, or hypothyroidism). The case women were diagnosed with PCOS using the Rotterdam criteria (PCOS Consensus Workshop 2004) (n=137). Based on the 2012 NIH-sponsored Evidence-based Methodology Workshop on Polycystic Ovary Syndrome 28 , the clinical records for each potential PCOS case were reviewed for the symptoms identified at diagnosis to differentiate between specific PCOS phenotypes: 1) ovulatory dysfunction and polycystic appearing ovaries on ultrasound (US) (n=55); 2) ovulatory dysfunction and androgen excess (n=15); 3) ovulatory dysfunction, polycystic ovaries, and androgen excess (n=67). No women with the Rotterdam phenotype of androgen excess and polycystic ovaries were identified because all patients with these characteristics also exhibited oligo or anovulation. The control group were women attending the same fertility clinic due to male factor infertility issues (n=103). This included women whose male partners had WHO IV defined semen deficiencies 29 – 31 , as well as women seeking donor sperm insemination. Vitamin D concentrations were analyzed using serum collected at the initial fertility clinic appointment and stored at −80 o C until analysis. Serum was analyzed by the Diagnostic Laboratory of Oklahoma (Quest Diagnostics) for 25-hydroxyvitamin D (25OHD) concentrations, using tandem mass spectrometry (LC-MS/MS). Total 25OHD concentrations (the sum of 25OHD 2 and 25OHD 3 ), as well as separate concentrations of 25OHD 2 and 25OHD 3, were measured. The analysis was conducted in three batches, with the first analysis completed as part of a pilot study in October 2011 (n=42). The remaining samples were separated into two batches (n=62 and 136) for analysis in June 2013. Statistical analyses examined the potential impact of batch variability and long-term freezer storage on serum 25OHD measurements. Vitamin D concentrations were categorized using Institute of Medicine (IOM) recommended cut-points for total 25OHD: ≤20 ng/mL (deficient), 21–29 ng/mL (insufficient), and ≥30 ng/mL (sufficient) 32 , 33 . Covariates were extracted from patient medical records. These variables were recorded at the first clinical evaluation for each patient and included demographic factors such as age (≥35, <35), race/ethnicity (non-Hispanic white, other), and body mass index (BMI) (<25, 25–29, 30–34, ≥35), as well as fertility factors such as years of infertility (<2, ≥2), parity (yes/no), and gravidity (yes/no). Tobacco use (yes/no), alcohol use (yes/no), and exercise habits (yes/no) were also available, but these were limited to binary indicators of current behaviors. Factors such as season of blood draw (Winter, Spring, Summer, Fall), IUI treatment year (2008–2012), and batch of serum analysis (1–3) were also considered as covariates. Data analyses were conducted using SAS v9.2 (Cary, NC, USA). Median (range) vitamin D levels for women with PCOS and women undergoing IUI due to male factor infertility were compared using a Wilcoxon Rank Sum Test. Covariate distributions were examined by case status and compared by calculating prevalence odds ratios. Differences in the proportion of cases and controls with sufficient, insufficient or deficient vitamin D levels were evaluated using chi-square tests. Associations between vitamin D concentrations and PCOS were evaluated using logistic regression, with prevalence odds ratios (POR) and 95% confidence intervals (CI) reported. Using directed acyclic graphs (DAGs), we selected age, race/ethnicity, alcohol, smoking, and BMI for evaluation as potential confounders. Quantitative assessment for confounding included these covariates in the logistic regression model and used backward elimination to evaluate the change in the POR for vitamin D. Only adjustment for BMI accompanied by race/ethnicity altered the POR by more than 15% and met our criteria for inclusion in the adjusted models. To acknowledge the uncertainty surrounding obesity as a potential consequence of inadequate vitamin D levels 34 – 36 , which would introduce bias by inappropriately adjusting for causal intermediates, the results are reported for both unadjusted associations and for associations controlling for BMI and race/ethnicity. Effect modification between the exposure and covariates was explored by entering interaction terms in the logistic regression models. No interaction terms had p-values ≤0.05; thus no covariates were identified as effect modifiers. Similar analyses were also conducted for the PCOS diagnostic phenotypes with unadjusted and adjusted results reported for each. Variability by batch of serum analysis was assessed using mixed effects logistic regression, modeling vitamin D exposure as a fixed effect and assay batch as a random effect. Accounting for laboratory batch had minimal impact on the study results; therefore batch was not included in the final models.

The prevalence of vitamin D deficiency was greater among PCOS subjects when compared to controls; however, this relationship, which approached marginal significance in unadjusted analyses, was not statistically significant in adjusted models. We chose to present the study results both with and without adjustment for BMI due to concerns that the temporal relationship between obesity, vitamin D levels and PCOS remains unclear. For instance, individuals with higher BMI have been reported to have lower vitamin D levels due to the sequestration of vitamin D in the larger adipocytes 37 , 38 . In contrast, associations between lower vitamin D concentrations and incident obesity have also been observed 34 – 36 . Additionally, since the etiology of PCOS remains unknown, the directionality of the relationship between PCOS and obesity remains unclear 5 , 17 – 19 . Thus, if BMI resides on the causal path between vitamin D and development of PCOS, it would not be appropriate to control for BMI as a confounder and doing so would lead to overadjustment bias toward the null value. Ultimately the results of our study suggest that BMI adjusted associations for vitamin D deficiency and insufficiency were similar to the unadjusted associations, but attenuated. Four previously conducted studies also reported lower serum 25OHD levels in women with PCOS 17 , 23 – 25 , 27 , however only one of those studies calculated a measure of association 23 . The case-control study by Mazloomi, et al. reported that the odds of vitamin D deficiency were 1.6 times higher among women with PCOS when compared to control women (95% CI: 1.2, 2.0) who were described as healthy, ovulatory women, and free of any systemic disease that may affect their reproductive physiology 23 . Associations with vitamin D insufficiency were not reported nor were any analyses for confounding variables. Despite our study having similar sample sizes of approximately 200 women, our unadjusted measures of association failed to reach statistical significance and when adjusted for BMI and race/ethnicity the results were further attenuated. These differences in results may be due, in part, to differences in our control populations. In the Mazloomi study, controls were recruited from an endocrine clinic and the infertility status was unknown 23 . Although it is difficult to discern, it appears that these controls were not women seeking infertility treatment like those selected for our study population. We used a similar study design to Li et al., which included a control group of women also seeking fertility treatments, but lacking a female infertility diagnosis 17 . Bloom et al. 39 addresses the common misconception that healthy controls are optimal for PCOS case-control comparisons, when in fact control selection should represent the exposure distribution among the source population at risk in order to avoid introducing bias. The source population at risk for PCOS is, thus, determined by the case definition 39 . Our controls who were seeking IUI due to male infertility were selected to represent the vitamin D exposure distribution in the source population from which the PCOS cases were drawn. Controls selected from the general population of fertile women may not reflect the exposure experience of women who would have sought treatment if they experienced an inability to conceive a pregnancy, thereby introducing a potential selection bias. Therefore, the use of a community-based control group in other studies that ascertained PCOS cases from an infertility clinic 21 may have introduced a selection bias if factors that influence access to fertility treatment, such as socioeconomic factors are also associated with vitamin D status. The generalizability of our results from this clinic-based population, however, is limited to women seeking treatment for infertility. The heterogeneous nature of PCOS and use of various diagnostic criteria 28 , 40 , 41 contributes to the difficulty in elucidating its etiology. To address the complexity of PCOS, a 2012 NIH-sponsored Evidence-based Methodology Workshop on Polycystic Ovary Syndrome reviewed the criteria used in diagnosing PCOS, recommending the use of the more broad Rotterdam diagnostic criteria 28 , 40 . Our study follows the panel’s recommendations for specifying the PCOS diagnostic phenotype in an effort to improve outcome classification and to shed light on the associations between etiologically distinct PCOS subgroups. In our study population, the majority of cases were characterized by all three Rotterdam criteria (ovulatory dysfunction, polycystic ovaries by US, and androgen excess) (49%), followed closely by cases with ovulatory dysfunction and polycystic ovaries (40%). No women were identified as having the PCOS phenotype of polycystic ovarian appearance on US and androgen excess only. Since this study population was recruited from a fertility clinic, it is reasonable that ovulation issues would be a defining feature of PCOS for these women. The results from our analyses of PCOS phenotypes indicate that women diagnosed with PCOS due to androgen excess (either increased serum testosterone levels or evidence of hirsutism) had higher odds of vitamin D insufficiency and deficiency when compared to controls. In our study 60% of PCOS women had phenotypes 2 or 3, which included androgen excess. Further support regarding the relationship between the PCOS symptom of androgen excess and decreased vitamin D levels is demonstrated by studies that observed negative correlations between PCOS symptoms, such as higher hirsutism scores and increased testosterone levels, and vitamin D levels 5 , 18 , 19 . Yet, no other studies have addressed the association between vitamin D levels and PCOS phenotypes. Additional research is needed to further investigate this association between androgen excess and vitamin D concentration. While the prevalence case-control design was a practical choice for studying women with PCOS, use of this design leaves us unable to establish the temporality of the association between PCOS and serum 25OHD. Evaluating risk factors for chronic disorders such as PCOS is challenging, given that disease onset is difficult to establish and thus incident cases are rarely identified or diagnosed. Thus, this study is limited by the measurement of serum 25OHD at the time of PCOS evaluation, which may not accurately reflect vitamin D concentrations present at the time of PCOS development. Studies of vitamin D variability, however, have reported a relatively high degree of correlation (>0.70) between 25OHD measures when repeated in women over one, two and three years 42 , 43 . An additional concern regarding the vitamin D measurement results from the potential overestimation of total 25OHD levels that may result when using the tandem mass spectrometry analytical method. Tandem mass spectrometry is the method of choice for large population studies, such as the NHANES 44 , and allows for individual quantification of the 25OHD 3 form. During the high-pressure liquid chromatography (HPLC) process, analyses completed without a specialized column result in the co-elution of 25OHD 3 and the epimeric version of this molecule. This co-elution results in inflated measures of total 25OHD, with an estimated 3% misclassification of vitamin D sufficiency 45 . However, our assessment of this potential source of misclassification using quantitative bias analyses 46 indicated it had no meaningful impact on study conclusions (data not shown). Serum 25OHD is known to be extremely stable, with reports of continued stability after undergoing up to 4 freeze-thaw cycles 47 , 48 . Additionally, the Vitamin D External Quality Assessment Scheme (DEQAS), a vitamin D quality assessment organization, ships serum samples to quality assessment laboratories by ground mail with no refrigeration, failing to observe issues with 25OHD concentrations 49 . Long-term freezer storage stability has also been observed when stored for up to 50 years 50 . We observed no association between vitamin D levels and year of IUI treatment with serum stored for up to 5 years. This suggests that sample degradation or desiccation during the storage period was not likely to contribute to exposure misclassification.

Although we failed to observe a statistically significant association between vitamin D exposure and PCOS among the entire case group, positive, but less precise, associations were observed in PCOS women exhibiting androgen excess. Thus, a mixing of etiologically distinct PCOS conditions may obscure associations with vitamin D deficiency and other potential risk factors. Future research should continue to evaluate relationships with more homogenous subgroups of PCOS women sharing similar phenotypes. The use of refined case definitions may provide more definitive information regarding PCOS development and potential treatment for women suffering from this disorder.

Polycystic ovary syndrome (PCOS) is the most common cause of anovulatory infertility, affecting approximately 6 to 12% of reproductive aged women 1 , 2 . First described by Stein and Leventhal in 1935, PCOS results from a malfunctioning hypothalamus-pituitary-gonadal (HPG) axis and produces symptoms such as oligo- or amenorrhea, hirsutism, and increased acne 3 , 4 . Women with PCOS may also exhibit symptoms of metabolic syndrome, including central obesity and insulin resistance 3 , 5 . Typically, PCOS symptoms begin around menarche 6 ; however women may not be diagnosed until years later when seeking medical treatment for infertility. Short and long-term health concerns with PCOS include obesity 7 , hypertension 8 , diabetes 9 , 10 , and endometrial cancer 11 . Thus, studies are needed to improve our limited understanding of PCOS and issues that could contribute to the associated co-morbidities. Vitamin D is considered essential for reproductive health and for achieving and supporting a healthy and viable pregnancy. Fertility issues, poor pregnancy outcomes, and pregnancy complications are associated with decreased levels of vitamin D 12 , 13 . According to the National Health and Nutrition Examination Survey (NHANES) 2003–2006 data, vitamin D deficiency (≤20 ng/ml) and insufficiency (21–30 ng/ml) are relatively high in women of reproductive age (20–44 years) with a prevalence of 11% and 26%, respectively 14 , 15 . Therefore, this lack of vitamin D sufficiency may negatively impact the health of reproductive age women. The prevalence of vitamin D deficiency has been shown to be even higher in women with PCOS. Li et al found that 72% of PCOS women were either deficient or insufficient 16 . Further support for this relationship is demonstrated by associations between vitamin D deficiency and symptoms of PCOS, including increased BMI 5 , 17 – 19 , higher hirsutism score, and hyperandrogenemia 5 , 18 , 19 . However, the existing literature comparing vitamin D concentrations in women with and without PCOS is inconsistent 20 , with studies reporting higher 21 , 22 , lower 23 , 24 , and no observed differences 17 , 25 – 27 in serum 25 hydroxyvitamin D (25OHD) levels. The discrepancy in these results could be due to variation in study design including differences in case definition, control selection, type of serum vitamin D analyses, and analytical methods. To address this unresolved question, we conducted a case-control study to evaluate the relationship between vitamin D deficiency and PCOS among infertility patients undergoing intrauterine insemination (IUI). The objective of our study was to compare serum 25OHD concentrations in women with PCOS to levels in non-PCOS ovulatory women undergoing IUI for male factor infertility. We aimed to evaluate associations with PCOS diagnoses overall and by diagnostic phenotypes.