Vitamin D Status Relates to Reproductive Outcome in Women With Polycystic Ovary Syndrome: Secondary Analysis of a Multicenter Randomized Controlled Trial

Abstract

Context:

Experimental evidence supports a relevance of vitamin D (VitD) for reproduction; however, data in humans are sparse and inconsistent.

Objective:

To assess the relationship of VitD status with ovulation induction (OI) outcomes in women with polycystic ovary syndrome (PCOS).

Design:

A retrospective cohort.

Setting:

Secondary analysis of randomized controlled trial data.

Participants:

Participants in the Pregnancy in PCOS I (PPCOS I) randomized controlled trial (n = 540) met the National Institutes of Health diagnostic criteria for PCOS.

Interventions:

Serum 25OHD levels were measured in stored sera.

Main Outcome Measures:

Primary, live birth (LB); secondary, ovulation and pregnancy loss after OI.

Results:

Likelihood for LB was reduced by 44% for women if the 25OHD level was < 30 ng/mL (<75 nmol/L; odds ratio [OR], 0.58 [0.35–0.92]). Progressive improvement in the odds for LB was noted at thresholds of ≥38 ng/mL (≥95 nmol/L; OR, 1.42 [1.08–1.8]), ≥40 ng/mL (≥100 nmol/L; OR, 1.51 [1.05–2.17]), and ≥45 ng/mL (≥112.5 nmol/L; OR, 4.46 [1.27–15.72]). On adjusted analyses, VitD status was an independent predictor of LB and ovulation after OI.

Conclusions:

In women with PCOS, serum 25OHD was an independent predictor of measures of reproductive success after OI. Our data identify reproductive thresholds for serum 25OHD that are higher than recommended for the nonpregnant population.

Stored sera from a RCT (Pregnancy in Polycystic Ovary Syndrome I) were assayed for 25OHD; vitamin D status was identified as an independent predictor of live birth following ovulation induction in women with PCOS.

Nearly 10% of the reproductive-age women in the United States (6.1 million) have difficulty achieving pregnancy, with ovulatory dysfunction being a major cause of female infertility (1, 2). Characterized by clinical and biochemical hyperandrogenism, menstrual irregularities, and ovulatory dysfunction, polycystic ovary syndrome (PCOS) is the most common cause of ovulatory infertility (3).

Although vitamin D (VitD) has long been recognized for its importance in skeletal biology (4), an appreciation of its relevance for reproductive physiology is also evident, particularly in animal models (5, 6). However, evidence supporting a critical role for VitD signaling in human reproduction is sparse and almost entirely observational, and data are inconsistent (7–15). A growing body of literature suggests mechanistic implications of VitD deficiency for insulin resistance, inflammation, dyslipidemia, and obesity, ie, clinical and metabolic phenomena commonly encountered in PCOS (16. 17), implying pathophysiological relevance of VitD insufficiency for PCOS. To that end, multiple observational as well as small sample trials have explored the role of VitD in PCOS (18–22).

Given that ovulation induction (OI) is a first-line approach for management of PCOS-related ovulatory infertility (23), and in light of the data summarized above relating VitD to both PCOS and fertility, we hypothesized that in women with PCOS, VitD insufficiency (serum 25-hydroxyvitamin D [25OHD] <30 ng/mL) is associated with lower rates of live birth (LB) after OI.

Subjects and Methods

Serum 25OHD levels were assessed in stored samples from participants in the Pregnancy in PCOS I (PPCOS I) randomized controlled trial (RCT) (24). Briefly, 626 reproductive-age women (ages 18–39 years), with elevated testosterone (T) levels and oligomenorrhea (thus meeting the 1990 National Institutes of Health criteria for PCOS diagnosis) (3) and seeking pregnancy, with at least one patent fallopian tube, normal uterine cavity, and a partner with sperm concentration of at least 20 million/mL in at least one ejaculate, were randomized to one of three different treatment arms: 1) clomiphene citrate (CC) 50 mg every day for 5 days; 2) metformin XR (M) 1000 mg twice daily; or 3) a combination of CC and M. Women continued on study medications for 30 weeks or six treatment cycles; treatment-related LB was the primary outcome of interest. Ovulation (OV; a secondary outcome) was determined by weekly or every other week assessment of serum progesterone levels and was confirmed by a serum progesterone concentration above 5 ng/mL (24). This trial demonstrated superiority of CC (either alone or in combination with M) over M alone as an OI agent, with LB rates of 22.5% in the CC group (47 of 209), 26.8% in the CC+M group (56 of 209), and only 7.2% (15 of 208) in the M alone group (P < .001 for M vs both CC and CC+M) (24).

Stored sera (either unthawed or previously thawed no more than three times) maintained at the Reproductive Medicine Network (RMN) biorepository at −80°C, from a time point before initiation of trial drugs, were utilized for assessment of 25OHD levels for the present study; 25OHD analyte stability over a protracted period of storage at temperatures of −20°C or less and after repeated thaw-freeze cycles is well established (25, 26). Approvals were obtained from the RMN Repository Committee and the Human Investigation Committee at Yale University for utilization of deidentified samples as well as clinical and biochemical data collected as part of the PPCOS I RCT, which included: 1) outcomes—LB (on intent to treat analysis), OV (ovulatory response achieved at least once over six cycles or up to 30 weeks), and attainment of pregnancy (positive pregnancy test); 2) participant characteristics—age, parity, body mass index (BMI), and ovulatory infertility as the only evident contributor to infertility (yes/no); hirsutism (yes/no) was identified based on the modified Ferriman-Gallwey pictorial assessment tool, a visual scoring method to quantify the presence and severity of hair growth in nine androgen-sensitive hair growth areas (a Ferriman-Gallwey score ≥8 was taken as evidence of hirsutism) (27); 3) two time variables (cycles to OV and days in study); and 4) baseline hormonal and metabolic parameters—total T (TT; ng/dL), SHBG (mg/dL), fasting glucose (mg/dL), fasting insulin (mIU/L), homeostasis model of assessment for insulin resistance (HOMA-IR), serum creatinine (mEq/L), and hepatic transaminases (aspartate aminotransferase and alanine aminotransferase; mg/dL). The following variables were created for analyses: pregnancy loss (PL; difference between positive pregnancy test and LB), glucose:insulin ratio (GIR), and free androgen index (FAI) (TT in nmol/L/SHBG in nmol/L × 100).

Assays for 25OHD were performed in duplicate at the Endocrinology and Metabolism Laboratory at Yale University using competitive equilibrium RIA (Diasorin, sensitivity 5ng/ml; intra assay and inter assay coefficients of variation 11% and 17% respectively) and mean value was used for analyses. As per guidelines of the Endocrine Society of North America (28), 25OHD ≥30 ng/mL (≥75 nmol/L) was defined as VitD sufficiency, levels between 20 and 29.9 ng/mL (50–74.9 nmol/L) were considered inadequate, <20 ng/mL (<50 nmol/L) was defined as VitD deficiency, and <10 ng/mL (<25 nmol/L) was characterized as severe VitD deficiency. Multiplication of value in ng/ml by 2.5 allows conversion to SI units.

OV (yes/no), LB (yes/no, primary outcome), and PL (yes/no) attainments were evaluated as outcomes of interest. Relationships between 25OHD (continuous) with dichotomous outcomes (OV, LB, and PL) were graphically assessed using Lowess curves (Supplemental Figures 1 and 2) for visual identification of inflection points (if any), at which association between 25OHD levels with one or more of the specified outcomes became exaggerated; VitD level was then dichotomized at these visually discerned threshold values for individual outcomes (20 ng/mL [50 nmol/L] for OV, 45 ng/mL [112.5 nmol/L) for LB, and 39 ng/mL [97.5 nmol/L] for PL).

Correlation between serum 25OHD levels with baseline endocrine (fasting insulin, TT, SHBG, FAI) and metabolic parameters (fasting glucose, GIR, serum creatinine, hepatic transaminases, and HOMA-IR) was also undertaken using Pearson or Spearman correlation analyses as appropriate (based on data distribution). Univariate analyses examined baseline data (participant characteristics, anthropometric and biochemical data for subjects prior to initiation of study interventions). Parametric tests (Student's t test and ANOVA) compared normally distributed continuous data across two or more groups, respectively; nonparametric tests (Mann-Whitney U test and Kruskal Wallis rank test) compared continuous data of skewed distribution across two or more groups, respectively, and proportions were compared by χ² test. Multivariable logistic regression analyses assessed the relationship between serum 25OHD (as both continuous and categorized variable) with the specified outcomes. Key covariates included in model building were age, BMI, smoking status, race, and hirsutism; additionally, variables demonstrating P ≤ .20 for relationship with outcome of interest on univariate analysis were incorporated in model building. Stepwise backward elimination of nonsignificant variables was then undertaken, and the final model represented one demonstrating the best “fit” for each outcome. Interactions between OI treatment and VitD status and between BMI and VitD for specified outcomes were examined; interaction terms were created and included in respective logistic regression models for each specified outcome. Goodness of model fit was assessed (29). Sensitivity, specificity, positive and negative predictive values (PPV and NPV) for specified 25OHD thresholds were also calculated (30).

Continuous data are presented as mean ± standard deviation or median (interquartile range), and categorical data are presented as percentage. Magnitude of associations are presented as odds ratio (OR) and 95% confidence interval (CI). P value <.05 was deemed statistically significant; P values are reported up to three decimal points, except as otherwise specified. STATA version 12 (StataCorp Inc.) was used for statistical analyses.

Sensitivity analyses

Given the lipophilic propensity of VitD metabolites, we hypothesized that an individual's body mass needs to be considered when interpreting VitD status based on circulating 25OHD level. A new variable, BMI adjusted D (BMI^aD), was created (ratio of serum 25OHD to BMI). Correlation analyses assessed the directionality and magnitude of the relationship of BMI^aD with baseline endocrine and metabolic parameters. Multivariable logistic analyses assessed the relationship of BMI^aD with specified outcomes after adjusting for previously specified covariates.

Power analysis

The overall LB rate for the PPCOS I cohort was almost 19% (118 of 626). Based on an assumption that participants in the lowest quartile of serum 25OHD would have half the LB rate as those in the highest quartile, we hypothesized a LB rate of 13% for subjects in the lowest and 26% for those in the highest 25OHD quartile. These assumptions provided >80% power to demonstrate such a difference, if indeed it existed, in a two-sided test with an α of 0.05.

Results

Stored sera collected before initiation of study drugs were available for 540 of the 626 PPCOS I participants (86%). Baseline data are presented in Table 1. VitD status was comparable for subjects in the three study treatment arms (serum 25OHD levels were 22.85 ± 10.12 ng/mL [57.13 ± 25.3 nmol/L], 24.11 ± 9.8 ng/mL [60.27 ± 24.5 nmol/L], and 23.71 ± 9.76 ng/mL [59.28 ± 24.4 nmol/L] for CC, M, and CC+M groups, respectively; P = .473).

Table 1.

Baseline Data for PPCOS I Subjects for Whom Stored Sera Were Available (n = 540)

Variable
Age, y	28.07 (3.98)
BMI, kg/m2	35.29 (8.66)
≥35	270 (50)
30–34.99	118 (22)
<30	152 (28)
Parity
0	362 (67)
1	178 (37)
Race
White	372 (69)
Black	88 (16)
Asian	16 (2.97)
American Indian	61 (11.34)
Native Hawaiian	1 (0.19)
Ethnicity
Hispanic	148 (27.41)
Non-Hispanic	392 (72.59)
Hirsutisma
Yes	436 (81)
No	104 (19)
Ovulatory dysfunctionb
Yes	415 (77)
No	125 (23)
Smoking history
Yes	205 (38)
No	335 (62)
Baseline TT, ng/dL	61.72 (27.74)
Baseline SHBG, nmol/L	29.53 (18.24)
Baseline FAIc	9.51 (6.45)
Baseline HOMAd	3.52 (1.90–6.15)
Baseline GIR	5.38 (3.24–8.48)
Baseline creatinine	0.76 (0.13)
Serum 25OHD, ng/mLe	23.56 (9.88)
Categories of VitD, ng/mL
<10e	42 (8)
10–19.99e	143 (27)
20–29.99e	207 (38)
≥30e	148 (27)
Serum 25OHD >45 ng/mLe
Yes	10 (2)
No	530 (98)
Study drug allocation
CC alone	175 (32)
CC+M	183 (34)
M alone	182 (34)
LB	102 (19)
OVf	402 (74)
PL	42 (29)
Days in study	180.70 (66.21)
Cycles to OV	2.45 (1.87)
1	235 (43)
2	142 (26)
3	39 (7)
4	37 (7)
5	13 (2)
6	46 (8)
7g	26 (5)
8g	2 (0.4)
> two cycles to OV
Yes	163 (30)
No	377 (70)

Continuous data are presented as mean (standard deviation) or median (interquartile range), and categorical data are presented as number (percentage).

^aFerriman-Gallwey score ≥8.

^bOvulatory dysfunction as attributable cause for infertility.

^cFAI: TT (nmol/L)/SHBG (nmol/L) × 100.

^dHOMA-IR: glucose (mg/dL) × insulin (μIU/mL)/405.

^eMultiply by 2.5 for conversion to SI units (nmol/L).

^fOvulation achieved during the course of the trial.

^gMaximum of six treatment cycles over a 30-week period.

Racial differences in VitD status were apparent; 25OHD levels were highest in White women (26.0 ± 9.2 ng/mL [64.82 ± 23 nmol/L]), and lowest in Black women (16.3 ± 10.2 ng/mL [40.83 ± 25.4 nmol/L]), with intermediate levels for women of other races (P < .01 for racial differences in serum 25OHD). The relationship between race and VitD was independent of BMI. Adjusting for BMI, Black women were 14 times (OR, 14.5; 95% CI, 7.2, 29.5) more likely to be severely deficient and 63% (OR, 0.37; 95% CI, 0.19, 0.70) less likely to have normal (≥30 ng/mL [≥75 nmol/L]) VitD levels.

Association of 25OHD with OV

Evidence of OV was observed in 74% of the cohort (402 of 540) over the 6-month trial duration. The probability of achieving OV varied directly with VitD status (68, 77, and 78% in those with VitD deficiency, insufficiency, and normal status, respectively; P = .050). VitD- deficient women were significantly less likely to achieve OV compared to those with 25OHD levels ≥20 ng/mL (P = .006; Table 2). Advancing age, higher GIR (ie, better insulin sensitivity), higher SHBG, and the use of CC (either alone or with M) were associated with higher chances of OV; conversely, higher BMI, ovulatory dysfunction as the sole attributed cause for infertility, higher FAI, and a longer time to OV were associated with the reduced likelihood of OV (Table 2). Unlike FAI, hirsutism was unrelated to the chance of achieving OV (P = .467). There was no evidence of interaction between VitD status and OI treatment categories or between VitD and BMI for OV outcome.

Table 2.

Predictors of OV on Multivariable Logistic Regression Analysis of PPCOS I RCT Data

Variables	Unadjusted OR	P Value	Adjusted OR	P Value
Age, y	1.07 (1.02–1.123)	.006	1.12 (1.04–1.20)	.002
25OHD <20 ng/mLa	0.58 (0.39–0.86)	.006	0.43 (0.25–0.76)	.003
BMI, kg/m2	0.94 (0.91–0.96)	<.001	0.95 (0.92–0.98)	.003
Ovulatory dysfunctionb	0.48 (0.28–0.81)	.006	0.46 (0.23–0.91)	.025
Baseline FAI	0.999 (0.998–1.00)	.060	1.002 (1.0001–1.003)	.031
Cycles to OV, n	0.44 (0.38–0.50)	<.001	0.45 (0.38–0.52)	<.001
CCc	1.36 (0.89–2.08)	.157	2.68 (1.41–5.09)	.003
CC+Mc	3.07 (1.89–4.98)	<.001	3.69 (1.86–7.30)	<.001

The magnitude of associations is presented as OR (95% CI). Model sensitivity is 89% with all specified covariates included in the model.

^aMultiply value by 2.5 for conversion to SI units (nmol/L).

^bVersus other infertility diagnoses.

^cVersus metformin alone.

On adjusted analyses, VitD deficiency, higher BMI, FAI, and length of time in study were identified as negative predictors of OV response; advancing age and use of CC (either alone or with M) were predictive of a higher likelihood for OV (Table 2). The final statistical model demonstrated an 89% sensitivity for the specified outcome.

Figure 1 demonstrates attainment of OV and time to OV during the course of the PPCOS I trial by VitD status.

Figure 1.

Kaplan-Meier curve for OV in the PPCOS I cohort by VitD status (OV in women with 25OHD <20 ng/mL vs in women with higher levels). For conversion to SI units (nmol/L), multiply value in ng/mL by 2.5.

Association of 25OHD with LB

The overall LB rate was almost 19% (112 of 540). Serum 25OHD was significantly higher in women achieving LB (25.34 ± 10.39) compared to those failing to attain LB (23.16 ± 9.71; P = .046). Each 1 ng/mL (2.5 nmol/L) increase in 25OHD increased the likelihood of LB by 2% (OR, 1.02; 95% CI, 1.00, 1.04; P = .046). A dose-response directionality was suggested when examining the relationship between LB and VitD status as defined by specified thresholds of 25OHD; compared to a LB rate of 26% in those sufficient in VitD (38 of 148), the likelihood of LB declined progressively in the settings of VitD insufficiency (33 of 207; OR, 0.74; 95% CI, 0.57, 0.96), deficiency (25 of 143; OR, 0.61; 95% CI, 0.35, 1.08), and severe deficiency (6 of 42; OR, 0.48; 95% CI, 0.19, 1.23). The proportion of women with VitD inadequacy (<30 ng/mL [<75 nmol/L]) was significantly higher among those failing to achieve LB compared to those attaining LB (75 vs 63%; P = .013).

Supplemental Figure 1 offers a visual representation of the relationship between LB and serum 25OHD and demonstrates an inflection point at and above 38 ng/mL (95 nmol/L), beyond which the association between serum 25OHD with LB is progressively magnified. The likelihood for LB was increased 4-fold (OR, 4.5; 95% CI, 1.27, 15.72; P = .02) for women with 25OHD levels >45 ng/mL (>112.5 nmol/L) (Table 3). Conversely, the likelihood of achieving LB was reduced by 44% for women with 25OHD levels <30 ng/mL (<75 nmol/L; OR, 0.58; 95% CI, 0.35, 0.92). The magnitude of association between 25OHD level and LB was exaggerated when analyses were restricted to participants assigned to CC treatment (alone or in combination with M) wherein each nanogram per milliliter increase in 25OHD increased the likelihood of LB by 3% (OR, 1.03; 95% CI, 1.01, 1.06). Progressive improvement in the odds for LB was noted at 25OHD thresholds of ≥38 ng/mL (>95 nmol/L) (n = 27; OR, 1.42; 95% CI, 1.08, 1.8; P = .013) and ≥40 ng/mL (≥100 nmol/L) (n = 20; OR, 1.51; 95% CI, 1.05, 2.17; P = .027).

Table 3.

Predictors of LB on Multivariable Logistic Regression Analysis of PPCOS I RCT Data

Variables	Unadjusted OR	P Value	Adjusted OR	P Value
Age, y	0.95 (0.90–1.01)	.081	0.94 (0.89–1.00)	.076
25OHD > 45 ng/mLa	4.46 (1.27–15.72)	.020	5.35 (1.02–28.15)	.048
BMI, kg/m2	0.95 (0.92–0.97)	<.001	0.97 (0.94–0.99	.044
Black raceb	0.43 (0.21–0.89)	.023	0.44 (0.20–0.96)	.039
Hirsutism, yes/no	0.95 (0.93–0.98)	.002	0.45 (0.26–0.78)	.005
CCc	1.44 (0.93–2.26)	.104	3.99 (1.92–8.30)	<.001
CC+Mc	2.23 (1.44–3.46	<.001	5.79 (2.82–11.88)	<.001
> Two cycles to OV	0.15 (0.07–0.33)	<.001	0.20 (0.10–0.43)	<.001
Baseline creatinine	0.24 (0.04–1.27)	.092	0.23 (0.03–158)	.135
History of smoking	0.78 (0.49–1.23)	.285	0.85 (0.51–1.40)	.521

The magnitude of associations is presented as OR (95% CI). Model sensitivity is 78% with all specified covariates included in the model.

^aMultiply value by 2.5 for conversion to SI units (nmol/L).

^bVersus other races.

^cVersus metformin alone.

On univariate analyses, advancing age, higher BMI, Black race, hirsutism, attainment of ovulatory cycles after more than two attempts at OI, and higher baseline serum creatinine levels were associated with reduced odds of LB (Table 3). Higher SHBG levels were associated with a significantly higher likelihood for achieving LB (OR, 1.02; 95% CI, 1.01, 1.03). Although smoking history reduced the odds of LB by 22%, this association was not of statistical significance. There was no evidence of interaction between VitD status and OI treatment categories or between VitD and BMI for LB.

On adjusted analyses, 25OHD >45 ng/mL (>112.5 nmol/L) and the use of CC (either alone or in combination with M) were associated with increased likelihood of LB (Table 3); conversely, Black race, increasing BMI, the presence of hirsutism, and attainment of ovulation after more than two attempts at OI were predictive of the reduced likelihood of LB (Table 3). Statistical significance of the associations between advancing age, serum creatinine, and smoking history with LB disappeared on adjusted analyses; however, these variables were retained in the final model due to biological plausibility (age and smoking history) and borderline statistical significance (serum creatinine). Association between SHBG and LB was no longer present on multivariable adjustment (P > .50). The final statistical model presented in Table 3 demonstrated 78% sensitivity for LB.

Figure 2 demonstrates attainment of LB achieved during the course of PPCOS I trial in participants across categories of VitD status (normal, insufficient, deficient, and severely deficient).

Figure 2.

Kaplan-Meier curves for LB in the PPCOS I cohort by VitD status based on specified serum levels of 25OHD (ng/mL). For conversion to SI units (nmol/L), multiply values in ng/mL by 2.5.

Association of 25OHD with PL

A positive pregnancy test was followed by PL in 29% (42 of144) of the cohort. Serum 25OHD level ≥38 ng/mL (≥95 nmol/L) was associated with an 82% reduced likelihood of PL compared to lesser levels (OR, 0.18; 95% CI, 0.02, 0.90; P = .020). There was no evidence of interaction between VitD status and OI treatment categories or between VitD and BMI for PL.

Association of serum 25OHD with hormonal and metabolic features of PCOS

Serum 25OHD levels demonstrated inverse associations with BMI (r = −0.21; P < .01) (Supplemental Figure 3), fasting insulin (r = −0.15; P < .01), and HOMA-IR (r = −0.11; P < .01) and positive associations with SHBG (r = 0.15; P < .01) and fasting GIR (r = 0.18; P < .01). VitD status did not exhibit any relationship with TT (P = .93), FAI (r = −0.05; P = .27), fasting glucose (P = .59), serum creatinine, or hepatic transaminases (data not shown).

Predictive value of VitD status for outcome of OI

Sensitivity, specificity, PPV, and NPV of serum 25OHD < 20 ng/mL (<50 nmol/L) for failed OV for the entire cohort were 48, 65, 32, and 78%, respectively (Supplemental Table 1). When analyses were restricted to the CC-treated population (CC alone or CC+M), respective values were marginally improved: sensitivity, 53%; specificity, 62%; PPV, 23%; and NPV, 86%.

A serum 25OHD level ≤45 ng/mL (≤112.5 nmol/L) demonstrated 99% sensitivity and 82% PPV for no LB; specificity and NPV were 5 and 50%, respectively (Supplemental Table 1).

A serum 25OHD level of <39 ng/mL (<97.5 nmol/L) demonstrated 100% sensitivity and 100% NPV for PL; values for specificity and PPV were 10 and 31%, respectively (Supplemental Table 1). Not a single case of PL was observed in women with serum 25OHD level ≥39 ng/mL (≥97.5 nmol/L).

Sensitivity analyses substituting BMI-adjusted 25OHD as independent variable of interest

Compared to 25OHD, BMI^aD demonstrated more robust associations with fasting hormonal and metabolic variables including insulin (r = −0.26; P < .01), glucose (r = −0.11; P = .01), GIR (r = 0.39; P < .01), HOMA (r = −0.21; P < .01), and SHBG (r = 0.41; P < .01). Unlike 25OHD, BMI^aD demonstrated an inverse correlation with hepatic alanine aminotransferase (r = −0.18; P < .001).

Supplemental Figure 2 presents a visual representation of the relationship between LB with BMI^aD. Association of BMI^aD with LB is a more uniform slope than that seen for 25OHD (Supplemental Figure 1), and this relationship was more robust when analyses were restricted to CC users (CC alone plus CC+M), wherein each unit increase in BMI^aD was associated with a 3-fold increase in the likelihood of LB (OR, 3.15; 95% CI, 1.80–5.54).

Association of VitD with specified outcomes was reassessed by substituting BMI^aD for 25OHD in the previously discussed multivariable logistic regression analyses (Supplemental Tables 2 and 3). For five of the 540 participants with BMI^aD > 2, ie, their serum 25OHD value was ≥ twice that of their BMI (mean serum 25OHD, 47 ± 4.45 ng/mL; BMI, 22.3 ± 2.39 kg/m²), all (100%) achieved OV, and four (80%) achieved pregnancy, with all pregnancies (100%) proceeding to LB.

Adjusted models exhibited sensitivities of 78 and 89%, respectively, for outcomes of OV and LB (essentially identical to those seen when 25OHD values were substituted for BMI^aD and BMI was included as an additional covariate) (Supplemental Tables 2 and 3, respectively).

Discussion

In a cohort of infertile women undergoing in vitro fertilization, we had previously identified facilitatory implications of replete VitD stores on in vitro fertilization success (7). Our current study reaffirms a relevance of adequate 25OHD for procreative success in women with PCOS undergoing OI. Beyond reaffirming a consistency in directionality of the previously observed associations, we have additionally noted that this association becomes apparent at serum 25OHD levels that are well beyond the threshold of 30 ng/mL (75 nmol/L), which is currently deemed a “normal” target level. The highest likelihood for LB was evident in women with a serum 25OHD level >45 ng/mL (>112.5 nmol/L); in contrast, 25OHD levels <20 ng/mL (<50 nmol/L) were predictive of a dampened OV response to OI strategies. These observations allow us to propose that a circulating 25OHD level of 45 ng/mL (112.5 nmol/L) or higher be considered as “optimal” for women attempting to conceive (Supplemental Figure 1).

Based on observations in this work, we propose concepts of two distinct “reproductive thresholds” of VitD below which OV to OI is blunted (lower reproductive threshold [LRT], <20 ng/mL [<50 nmol/L]) and beyond which (upper reproductive threshold [URT], >45 ng/mL [>112.5 nmol/L]) the likelihood of LB may be optimized. Although the study design does not allow us to comment on pathophysiological mechanisms, our data suggest that PL risk is mitigated at 25OHD levels of ≥39 ng/mL (≥97.5 nmol/L). These latter observations are in line with recent suggestions that VitD deficiency, by alterations in the status of cellular and autoimmunity, may be contributory to PL (31). Notably, the observed URT is higher than 25OHD levels of 20 ng/mL (50 nmol/L) and 30 ng/mL (75 nmol/L) that are identified by the Institute of Medicine (32) and The Endocrine Society (28), respectively, to reflect normal VitD status. Recapitulation of our earlier findings of higher VitD levels associating with a higher likelihood of fertility treatment success in the large sample size of the PPCOS I cohort and consistency with similar observations reported by others (8–12) reinforces the relevance of an individual woman's VitD status for fertility.

Racial disparities in fertility treatment success rates are well described, with lower pregnancy rates observed in infertile women of color after infertility treatments (33). The observed variability of VitD status across the spectrum of races represented in PPCOS I is in line with prior observations. The PPCOS I cohort was predominantly comprised of Caucasians; however, the racial and ethnic composition of enrollees still allowed exploration of the relationship between specified outcomes with VitD status and race. VitD deficiency has been suggested as a mechanism for lower treatment related fertility rates in Blacks (34). Our analyses reaffirm that OI-related LB rates are lower in Black compared to Caucasian women with PCOS; however, this relationship is independent of VitD status. Our study design does not allow postulation on mechanisms that could explain the observed racial differential in OI-related LBs.

Inverse correlations between serum 25OHD levels with BMI and with fasting insulin are recognized (5, 17) and are affirmed in our analyses. Associations of serum 25OHD with hyperandrogenemia and hyperandrogenism are described, and reductions in TT are achieved with VitD supplementation, albeit inconsistently (19, 20, 35, 36). We did not observe any association between serum 25OHD and TT, FAI, or hirsutism.

VitD signaling holds the potential of negating a spectrum of pathophysiological corollaries to obesity (4, 37), hence our rationale to adjust VitD status by BMI. We observed an increase in the magnitude of observed associations of VitD status with LB and OV (Supplemental Tables 2 and 3). Given a recognized inverse relationship between BMI and serum 25OHD and the higher inflammatory and metabolic burden of obesity, conceptually, one can make a case that the burden of obesity should be considered when defining norms for VitD status. Indeed, compared to 25OHD, BMI^aD demonstrated more robust associations with hormonal and metabolic variables. Given the improved magnitude of association of BMI^aD status for the various outcomes discussed, we propose that future efforts aimed at assessing a relationship between VitD status and disease states incorporate such a paradigm to better explain the relevance of VitD in health and disease.

A retrospective approach, lacking information on a number of variables that may modulate efficiency of VitD signaling (such as VitD binding protein levels that would allow a more precise calculation of bioavailable 25OHD) (38), population's VitD receptor genotype status, and VitD binding protein gene polymorphisms (39, 40), missing information on seasonality and on the intake of VitD supplements are obvious limitations of our work. Given the small number of participants with 25OHD levels above the specified thresholds for defined outcomes, ie, only 39 of 540 (7%) had a level >38 ng/mL (>95 nmol/L) associated with PL, and only 10 of 540 (2%) had levels >45 ng/mL (associated with LB), the possibility of α errors being reflected in the observed associations is plausible. However, the directionality of and consistency in the observed associations that held on analyses that utilized BMI^aD instead of 25OHD, wherein 118 subjects demonstrated a value >1 (ie, serum 25OHD level was greater than the value of an individual's BMI) is reassuring (Supplemental Figure 2). The robustness of PPCOS I, a randomized double-blind controlled trial, and assessment of VitD status in a large sample of women meeting the National Institutes of Health criteria for PCOS are all strengths of this work.

In a large sample representative of the PCOS population, we have systematically assessed and demonstrated a relevance of VitD for OV response and LB with commonly utilized OI strategies. Our interpretation of VitD status based on BMI consideration is in line with the concept of “individualized medicine.”

In summary, our data suggest that for infertile women with PCOS, VitD status as reflected by serum levels of 25OHD is relevant for procreative success. We hypothesize that a decline in circulating 25OHD below the LRT may be contributory to ovulatory dysfunction; conversely, 25OHD levels at and above the URT may confer improved endometrial receptivity, an effect that has been previously suggested for VitD (41, 42). Improved VitD status through supplementation can be theorized to hold potential for improving OI treatment related OV and LB rates and for reducing risk of PL in women with PCOS, a population that is already at an enhanced risk for pregnancy wastage (43). These functional observations provide support for the biological plausibility of our conclusions. The discrepancy between what may be an appropriate reproductive threshold and the currently defined norms for VitD status in adults merits further consideration. Future studies are needed to systematically assess these notions because the clinical, public health, and financial implications of such a simple, safe, and inexpensive strategy can be substantial.