Vitamin D Deficiency Is Associated With Poor Ovarian Stimulation Outcome in PCOS but Not Unexplained Infertility

Samantha F. Butts; David B. Seifer; Nathanael Koelper; Suneeta Senapati; Mary D. Sammel; Andrew N. Hoofnagle; Andrea Kelly; Steven A. Krawetz; Nanette Santoro; Heping Zhang; Michael P. Diamond; Richard S. Legro for the Eunice Kennedy Shriver National Institute of Child Health and Human Development Reproductive Medicine Network


J Clin Endocrinol Metab. 2019;104(2):369-378. 

In This Article

Subjects and Methods

Study Cohorts

Vitamin D status was determined by measuring serum 25(OH)D, which is the primary circulating form of the vitamin. The study design and methods for each of the RMN trials has been previously described.[10,19] PPCOS II was a doubleblind, multicenter trial of 750 women with PCOS who were randomized to receive letrozole or clomiphene citrate to determine rates of ovulation and live birth for up to five treatment cycles.[10] The trial enrolled subjects ages 18 to 40 years meeting modified Rotterdam criteria for PCOS and the exclusion of disorders mimicking PCOS. Additional inclusion criteria were ≥1 patent fallopian tube and a normal uterine cavity, a male partner with a sperm concentration of ≥14 million per milliliter, with documented motility according to World Health Organization cutoff points, in at least one ejaculate during the previous year. The primary trial outcomes demonstrated the superiority of letrozole compared with Clomid in achieving ovulation and live birth. Of the original 750 PPCOS II subjects, 607 had banked serum available to assess vitamin D status and were included in this study.

The AMIGOS trial was a randomized multicenter trial that enrolled 900 subjects to assess whether ovulation induction for up to 4 cycles in couples with unexplained infertility treated with the aromatase inhibitor letrozole reduced multiple gestations while maintaining a comparable pregnancy success rate to that achieved by ovulation induction with either gonadotropins or Clomid.[19] Women were between 18 and 40 years of age with regular menses, had a normal uterine cavity with ≥1 patent fallopian tube, and had a male partner with a semen specimen of ≥5 million sperm/mL. The trial concluded that in women with unexplained infertility, ovarian stimulation with letrozole resulted in a significantly lower frequency of multiple gestation and a lower frequency of live birth compared with gonadotropin but not as compared with Clomid. Of the original 900 AMIGOS subjects, 647 had banked serum available to assess vitamin D status and were included in this study.

Permission was received from the RMN Resource and Data Access committee to access serum and clinical data on study participants from PPCOS II and AMIGOS. The study protocol was submitted to the University of Pennsylvania institutional review board and considered eligible for institutional review board review exemption authorized by 45 CFR 46.101, category 4. Both PPCOS II (NCT00719186) and AMIGOS (NCT01044862) were registered with

Data Collection

For each RMN study, a baseline study visit before randomization and treatment initiation occurred; at this visit, blood was collected from subjects providing consent to have serum stored in the RMN biorepository.[10,19] In properly banked serum, 25(OH)D is stable for at least 6 years and after up to four freeze–thaw cycles.[20,21] Quantification of total 25(OH)D was performed using liquid-liquid extraction and liquid chromatography-tandem mass spectrometry in the Department of Laboratory Medicine at the University of Washington.[22] This assay has been shown to meet the target criteria for precision and bias established by the Vitamin D Standardization Program.[23] Subject-level data regarding demographics and relevant clinical variables including outcomes of interest from PPCOS II and AMIGOS were abstracted and deidentified.

Clinical Measures and Outcomes

The primary study outcome was live birth. Secondary outcomes included ovulation, pregnancy, and pregnancy loss. In PPCOS II subjects, ovulation was defined by a progesterone level ≥3 ng/mL (10 nmol/L) measured at the monthly midluteal visit 3 weeks after initiation of study medication ± 4 days. Pregnancy was defined as a serum level of human chorionic gonadotropin >10 mIU/mL.[10] Vitamin D status was characterized according to Endocrine Society cutoffs.[24] Vitamin D deficiency is characterized as a 25(OH)D level <20 ng/mL (50 nmol/L); a 25(OH)D level of 20 to 29 ng/mL (52 to 72 nmol/L) is consistent with vitamin D insufficiency, and a level ≥30 ng/mL (75 pmol/L) indicates sufficient vitamin D status.[13,24,25] In this analysis, vitamin D status was dichotomized as 25(OH)D < 20 ng/mL (D deficiency) or 25(OH)D ≥ 20 ng/mL groups. These categories reflected the most robust cutpoint when performing univariate tests of the associations between 25(OH)D levels and live birth and based on locally weighted regression of live birth on 25(OH)D levels depicted graphically using a locally weighted scatterplot smoothing curve (Supplemental Figure 1).

Statistical Analysis and Power

Analyses were conducted separately for the PPCOS II and AMIGOS subjects. Comparisons of continuous data were analyzed using Student t test or Wilcoxon rank-sum tests, whereas categorical data were analyzed using the χ2 test. Graphical and statistical tests of normality were used to examine distributional assumptions of continuous variables. Logistic regression was used to calculate ORs to perform initial tests of association between D deficiency and outcomes of interest. Multivariable logistic regression was used to adjust OR estimates for relevant covariates. For the primary study outcome of live birth, prepregnancy body mass index (BMI), and study treatment arm were selected a priori to be included in multivariable models. Otherwise, model covariates were selected if the marginal association between the variable and the outcome was present at the P ≤ 0.1 level of significance. Variables were removed in a backward stepwise fashion if they did not maintain statistical significance at P < 0.05 unless there was strong biological plausibility for retention in the model. For the PPCOS II ovulation and live birth multivariable models, total testosterone was treated as a categorical variable and examined as quartiles to limit the effect of extreme values.A final logistic regressionmodel for live birth associated with D deficiency in PPCOS II subjects was adjusted for age, BMI, race, quartiles of total testosterone, and severity of insulin resistance. Collinearity diagnostics were evaluated for the final multivariable model to ensure the stability of regression coefficients and tests of statistical significance. Likelihood ratio tests were performed to determine whether the associations between D deficiency and chance of live birth were modified across strata of race and treatment.

Kaplan-Meier curves were generated to graphically depict time from randomization to positive pregnancy in subjects who ultimately achieved a live birth according to vitamin D status. A log-rank test was used to determine the significance of this association. A detailed power calculation is described in the Supplemental Methods. SAS, version 9.2, was used for statistical analysis.