Prospective Associations of Vitamin D Status With β-Cell Function, Insulin Sensitivity, and Glycemia: The Impact of Parathyroid Hormone Status

This study was performed in the setting of a prospective observational cohort consisting of women representing the full spectrum of glucose tolerance in a recent pregnancy (from normal to gestational diabetes mellitus [GDM]), who thereby have a broad range of risk for the future development of prediabetes and T2DM in the years after delivery (23–25). This cohort provided a model for studying the longitudinal relationship between vitamin D/PTH and glucose metabolism in the first year postpartum for two reasons: 1) lactating women in the first year postpartum in Toronto, Ontario, Canada (latitude 43°42′N) are a population at risk for vitamin D deficiency/insufficiency and 2) the range of future diabetic risk within this cohort has been shown to manifest in changes in β-cell function, insulin sensitivity, and glycemia between 3 and 12 months postpartum (23,26).

As previously described (23–25), the women comprising this cohort were recruited at the time of antepartum screening for GDM in the late 2nd trimester and underwent metabolic characterization at recruitment and at both 3 months and 12 months postpartum. At our institution, women are screened for GDM by a 50-g glucose challenge test (GCT) in the late 2nd trimester, followed by referral for a diagnostic oral glucose tolerance test (OGTT) if the GCT is abnormal. In this cohort study, women are recruited either before or after the GCT, and all participants undergo a 3-h 100-g OGTT for determination of GDM status (regardless of the GCT result) (27). The resultant cohort thus reflects the full spectrum of glucose tolerance in pregnancy from normal to GDM, which translates to a gradient of future risk for postpartum progression to prediabetes and T2DM. For this cohort study, participants return to the clinical investigation unit at both 3 and 12 months postpartum to undergo repeat metabolic characterization, including evaluation of glucose tolerance by 2-h 75-g OGTT. The protocol has been approved by the Mount Sinai Hospital Research Ethics Board, and all women have provided written informed consent for their participation. The current study was performed in 494 women who have completed their 12-month postpartum visit and had vitamin D and PTH measured at 3 months postpartum, thereby enabling assessment of the longitudinal relationships between vitamin D/PTH status and metabolic outcomes (β-cell function, insulin sensitivity, glycemia) 9 months later.

At the study visits at 3 and 12 months postpartum, interviewer-administered questionnaires were completed including assessment of current medications/supplements, duration of breast-feeding, and physical activity. As previously described (25), physical activity was assessed by the validated Baecke questionnaire. This instrument measures total physical activity and its component domains of sport-related physical activity (sport index), nonsport leisure-time activity (leisure-time index), and occupation-associated activity (work index). Work index is not measured at 3 months postpartum, as most women would not yet have returned to their usual occupation at that time. At each study visit, physical examination was performed including measurement of weight and waist circumference.

Vitamin D status was assessed with measurement of serum 25-OH-D by competitive electrochemiluminescent immunoassay on the Roche Modular E170 (cat. no. 05894913190; Roche Diagnostics, Laval, Canada). This assay has a lower reporting limit of 8 nmol/L. Serum PTH was measured using an electrochemiluminescence immunoassay on the Roche Modular E170 Analyzer (cat. no. 11972103122; Roche Diagnostics), which has a detection range from 0.6 to 530 pmol/L.

All OGTTs at 3 and 12 months postpartum were performed in the morning after an overnight fast. Venous blood samples were drawn for the measurement of glucose and specific insulin at fasting and at 30, 60, and 120 min after the ingestion of the glucose load, as previously described (23,24).

At each OGTT, current glucose tolerance status was determined according to Canadian Diabetes Association guidelines (28). Dysglycemia refers to prediabetes (impaired glucose tolerance, impaired fasting glucose, or both) or T2DM. Area under the insulin curve (AUC_ins) and area under the glucose curve (AUC_gluc) during the OGTT were calculated using the trapezoidal rule. Insulin sensitivity was measured using the Matsuda index, an established measure of whole-body insulin sensitivity that has been validated against the euglycemic-hyperinsulinemic clamp (29). β-Cell function was assessed on each OGTT with the Insulin Secretion-Sensitivity Index-2 (ISSI-2). ISSI-2 is a validated measure of β-cell function that is analogous to the disposition index obtained from the intravenous glucose tolerance test (30,31). ISSI-2 has been directly validated against the disposition index from the intravenous glucose tolerance test, with which it exhibits stronger correlation than do other OGTT-derived measures of β-cell function (31), and has been used to measure β-cell function in several previous studies, including both clinical trials and observational studies (10,23,32–35). ISSI-2 is defined as the product of 1) insulin secretion measured by the ratio of AUC_ins to AUC_gluc and 2) insulin sensitivity measured by Matsuda index (30,31).

All analyses were conducted using SAS 9.1 (SAS Institute, Cary, NC). Continuous variables were tested for normality of distribution, and natural log transformations of skewed variables were used, where necessary, in subsequent analyses.

Participants were initially stratified into groups according to 1) vitamin D status at 3 months postpartum and 2) tertiles of PTH at 3 months postpartum, respectively. Vitamin D status was classified as per Endocrine Society guidelines as vitamin D deficient (25-OH-D <50 nmol/L) (n = 161), vitamin D insufficient (25-OH-D ≥50 nmol/L and <75 nmol/L) (n = 178), or vitamin D sufficient (25-OH-D ≥75 nmol/L) (n = 155) (2,36). The tertiles of PTH were defined as 1st tertile (PTH ≤2.6 pmol/L) (n = 167), 2nd tertile (PTH >2.6 pmol/L and ≤3.8 pmol/L) (n = 174), and 3rd tertile (PTH >3.8 pmol/L) (n = 153). Univariate differences across both the vitamin D groups and the tertiles of PTH were assessed by ANOVA and Wilcoxon rank sum test for continuous variables and the χ² test for categorical variables (Table 1).

For evaluation of whether these vitamin D and PTH groups at 3 months postpartum predict insulin sensitivity, β-cell function, and glycemia at 12 months postpartum, multiple linear regression models were constructed with metabolic outcomes of insulin sensitivity (Matsuda index), β-cell function (ISSI-2), and glycemia (fasting glucose and 2-h glucose) at 12 months postpartum as dependent variables (Table 2). In each case, model 1 was adjusted for risk factors for diabetes mellitus (age, ethnicity, family history of diabetes mellitus, previous GDM, BMI at 3 months) and baseline levels of the outcome. The subsequent models were further adjusted for possible confounders that could impact the association of vitamin D/PTH and the metabolic outcomes: duration of breast-feeding (model 2) and total physical activity and season of blood collection (model 3). The models were tested for collinearity of covariates using the variance inflation factor, which confirmed no significant collinearity. Sensitivity analyses were performed with adjustment for use of calcium and vitamin D supplements, smoking status, and change in BMI between 3 and 12 months postpartum. In addition, all analyses were repeated after excluding the 19 participants with PTH above the upper limit of the laboratory normal range (6.5 pmol/L).

To evaluate the combined impact of vitamin D status and PTH tertile at 3 months on future glucose tolerance, we evaluated the prevalence of dysglycemia (prediabetes or diabetes mellitus) at 12 months postpartum in groups defined by both vitamin D status and PTH using the χ² test (Fig. 1). For further evaluation of the combined impact of vitamin D and PTH status on metabolic outcomes, participants were stratified into four groups as follows: 1) vitamin D sufficient and PTH in the 1st/2nd tertile (reference group,; 2) vitamin D sufficient and PTH in the 3rd tertile, 3) vitamin D deficient/insufficient and PTH in the 1st/2nd tertile, and 4) vitamin D deficient/insufficient and PTH in the 3rd tertile. We then tested for a biological interaction between vitamin D deficiency/insufficiency and PTH in the 3rd tertile. Using the equation described by Rothman (37) and others (38–40), we evaluated whether the combined impact of vitamin D deficiency/insufficiency with PTH in the 3rd tertile on the outcomes of dysglycemia, lowest tertile of Matsuda index, and lowest tertile of ISSI-2 at 12 months postpartum exceeded the sum of the individual effects of these conditions alone (Fig. 2). With this approach (38–40), the following three measures of biological interaction are calculated to quantify the amount of interaction on a multiplicative scale: the relative excess risk due to interaction (RERI), the attributable proportion due to interaction (AP), and the synergy index (S). RERI can be interpreted as the risk that is additional to that which is to be expected on the basis of addition of the odds ratios (OR) under exposure, calculated as the difference between the expected risk and the observed risk (RER1 = OR₁₂ – OR₁ – OR₂ + 1). AP can be interpreted as the proportion of disease that is due to interaction among persons with both exposures (AP = RER1/OR₁₂). S can be interpreted as the excess risk from both exposures in the setting of interaction, relative to the risk from exposure without interaction: S = [OR₁₂ − 1]/[(OR₁ − 1) + (OR₂ – 1)]. As previously described (37,38), RERI = 0, AP = 0, and S = 1 indicate the absence of biological interaction.

Using ANCOVA, we compared the percentage change from 3 to 12 months postpartum for insulin sensitivity (Matsuda index), β-cell function (ISSI-2), and glycemia (fasting glucose and 2-h glucose) between these four vitamin D/PTH groups, adjusted for age, ethnicity, family history of diabetes mellitus, previous GDM, and BMI at 3 months (Fig. 4). Finally, sequentially adjusted multiple linear regression models (Table 3) were constructed with Matsuda index, ISSI-2, fasting glucose, and 2-h glucose at 12 months postpartum as the outcomes using the same approach to covariate adjustment as described earlier for Table 2.

The study population consisted of 494 women aged 34.8 ± 4.3 years. There were no women with renal disease or other serious medical comorbidities. We first evaluated the respective metabolic implications of their vitamin D status and their PTH status, in turn, before considering both vitamin D and PTH together.

Table 1 shows baseline (3 months postpartum) characteristics of the participants stratified into the following three groups based on their vitamin D status: vitamin D deficient (n = 161; 33% of study population), insufficient (n = 178; 36%), and sufficient (n = 155; 31%). As anticipated, the groups differed in ethnicity (P < 0.001), with the vitamin D–deficient group having the greatest proportion of nonwhite ethnicity (39.8%). In addition, the vitamin D–deficient group had higher prevalence rates of family history of diabetes mellitus (P = 0.04) and current smoking (P < 0.008) and lower levels of physical activity at 3 months postpartum (all P < 0.03).

Metabolically, BMI (P < 0.001) and waist circumference (P = 0.03) differed across the groups, with both being highest in the deficient group. Consistent with these differences in adiposity, the vitamin D groups also differed with respect to insulin sensitivity, β-cell function, and glycemia at 3 months postpartum. Specifically, there was a stepwise decrease in Matsuda index (P < 0.001) and ISSI-2 (P = 0.04) from the sufficient to insufficient to deficient group, coupled with an analogous progressive increase in fasting glucose (P < 0.001) and 2-h glucose (P = 0.002) across these groups.

On multiple linear regression analyses (Table 2), vitamin D deficiency at 3 months postpartum independently predicted increased fasting glucose (P = 0.008) and 2-h glucose (P = 0.01) at 12 months postpartum compared with the vitamin D–sufficient (reference) group. However, by itself, vitamin D deficiency did not independently predict poorer insulin sensitivity (Matsuda index) or β-cell function (ISSI-2). Vitamin D insufficiency at 3 months predicted decreased β-cell function (P = 0.006) and higher 2-h glucose (P = 0.04) at 12 months postpartum compared with the reference group.

As shown in Table 1, the study population was also stratified into tertiles based on serum PTH at 3 months postpartum. In contrast to the vitamin D groups, these PTH groups did not differ in ethnicity, family history of diabetes mellitus, and current smoking. Leisure-time index was the only physical activity measure that differed between the PTH tertiles and was lowest in the 3rd tertile (P = 0.03). As expected, 25-OH-D differed across the groups (P < 0.001).

In the same way as observed across the vitamin D groups, BMI and waist differed across the PTH tertiles, being highest in the 3rd tertile (both P < 0.001). In addition, insulin sensitivity progressively decreased (P < 0.001) and fasting glucose increased (P = 0.004) from the 1st to 2nd to 3rd tertile. ISSI-2 was lowest in the 3rd tertile, but the overall comparison across the groups did not reach significance (P = 0.06).

On multiple linear regression analyses (Table 2), the 2nd tertile of PTH at 3 months postpartum did not independently predict insulin sensitivity, β-cell function, fasting glucose, or 2-h glucose at 12 months compared with the (reference) 1st tertile. However, the 3rd tertile of PTH at 3 months independently predicted lower insulin sensitivity (P = 0.008) and higher 2-h glucose (P = 0.007) 9 months later.

Recognizing that evaluation of vitamin D in conjunction with PTH might better reflect the status of the vitamin D/PTH axis than would consideration of either hormone in isolation, we next assessed the prevalence of dysglycemia (prediabetes or diabetes mellitus) at 12 months in relation to vitamin D status and PTH tertile combined. Indeed, as shown in Fig. 1, there was a striking difference in the prevalence of dysglycemia at 12 months postpartum across the nine groups defined by vitamin D status (sufficient, insufficient, and deficient) and PTH tertile at 3 months postpartum (P = 0.003), with the highest prevalence in the group with vitamin D deficiency and PTH in the 3rd tertile (30%). Overall, of the 79 participants with prediabetes/diabetes mellitus at 12 months postpartum, 32% were in the group with vitamin D deficiency and PTH in the 3rd tertile at 3 months.

To further evaluate the combined impact of vitamin D and PTH on metabolic parameters, we stratified participants into the following four groups based on vitamin D and PTH status at 3 months postpartum: 1) vitamin D sufficient and PTH in the 1st/2nd tertile (reference group) (n = 130), 2) vitamin D sufficient and PTH in the 3rd tertile (n = 25), 3) vitamin D deficient/insufficient and PTH in the 1st/2nd tertile (n = 212), and 4) vitamin D deficient/insufficient and PTH in the 3rd tertile (n = 127). To test for a biological interaction, we investigated whether the combined impact of vitamin D deficiency/insufficiency and PTH in the 3rd tertile on the outcomes of 1) dysglycemia, 2) lowest tertile of Matsuda index, and 3) lowest tertile of ISSI-2 at 12 months postpartum exceeded the sum of the individual effects of these conditions.

With respect to dysglycemia at 12 months postpartum, the presence of vitamin D deficiency/insufficiency and PTH in the 3rd tertile alone conferred increments in the risk for dysglycemia of 48.4% and 71.4%, respectively, compared with the reference group (vitamin D sufficiency and PTH in the 1st/2nd tertile). Interestingly, when both conditions were present, there was a 203.2% increased risk for dysglycemia, which resulted in RERI = 83.4%, AP = 27.5% and S = 1.69 (Fig. 2A). In other words, the combined effect of vitamin D deficiency/insufficiency and PTH in the 3rd tertile conferred an excess risk of 83.4% beyond the sum of the individual effects. Furthermore, as indicated by the AP parameter, the effect of the interaction accounts for 27.5% of the dysglycemia among individuals with both vitamin D deficiency/insufficiency and PTH in the 3rd tertile.

Similarly, for the outcome of being in the lowest tertile of Matsuda index at 12 months postpartum, the presence of vitamin D deficiency/insufficiency and PTH in the 3rd tertile alone conferred excess risks of 109.8% and 21.3%, respectively, while the combined effect conferred an increased risk of 360.9% (RERI = 229.8%, AP = 49.8%, and S = 2.75) (Fig. 2B). Lastly, for the lowest tertile of ISSI-2, the presence of vitamin D deficiency/insufficiency and PTH in the 3rd tertile alone conferred risk increments of 90.8% and −5%, respectively, while the combined effect conferred an increased risk of 152% (RERI = 66.2%, AP = 26.3%, and S = 1.77) (Fig. 2C).

Figure 3 shows the adjusted estimates at 3 and 12 months postpartum for insulin sensitivity, β-cell function, fasting glucose, and 2-h glucose in the four vitamin D/PTH groups after adjustment for the diabetes mellitus risk factors age, ethnicity, family history of diabetes mellitus, previous GDM, and BMI. Notably, vitamin D deficiency/insufficiency with PTH in the 3rd tertile was associated with lower Matsuda index and ISSI-2 and increased fasting glucose and 2-h glucose. In addition, this group exhibited a pronounced deterioration in ISSI-2 and 2-h glucose between 3 and 12 months postpartum that was not observed in the other three groups.

To further explore the differential changes in metabolic outcomes between the four groups defined by vitamin D and PTH together, we evaluated the adjusted percentage change in insulin sensitivity, β-cell function, and glycemia in each of these groups between 3 and 12 months postpartum. As shown in Fig. 4, only the group with vitamin D deficiency/insufficiency and PTH in the 3rd tertile had a significant decline in ISSI-2 (−4.7 ± 3.4% vs. 7.6 ± 4.4%, P = 0.04) and an increase in 2-h glucose (6.6 ± 2.6% vs. −0.4 ± 2.8%, P = 0.05) compared with the reference group. Of note, none of the other vitamin D/PTH groups showed changes in these metabolic outcomes compared with the reference group (all P > 0.15).

Finally, we performed multiple linear regression analyses to evaluate the longitudinal associations between combined vitamin D/PTH status at 3 months postpartum and metabolic outcomes after full covariate adjustment. As shown in Table 3, vitamin D deficiency/insufficiency accompanied by PTH in the 3rd tertile was the only group that independently predicted lower insulin sensitivity (P = 0.01) and β-cell function (P = 0.008) and increased fasting glucose (P = 0.03) and 2-h glucose (P = 0.002) at 12 months postpartum compared with the reference group (Table 3). Of note, vitamin D deficiency/insufficiency with PTH in the 1st/2nd tertile did not independently predict any of these metabolic outcomes. In sensitivity analyses, these results were unchanged with further adjustment for use of vitamin D/calcium supplements, smoking status, and change in BMI between 3 and 12 months postpartum (data not shown). In addition, none of the results changed after excluding the 19 participants with PTH above the laboratory normal range, with the exception of fasting glucose with which vitamin D deficiency/insufficiency and PTH in the 3rd tertile were now associated at borderline significance (P = 0.06) (data not shown).

In this study, we show that the combination of vitamin D deficiency/insufficiency and increased PTH is an independent predictor of deterioration in insulin sensitivity, β-cell function, and glycemia in a cohort of women in the 1st year postpartum. Notably, vitamin D deficiency/insufficiency with higher PTH conferred an excess risk for future dysglycemia, decreased insulin sensitivity, and poorer β-cell function that exceeded the sum of the individual risks associated with these conditions. While both vitamin D and PTH alone were independently associated with some of the outcomes, only the coupling of vitamin D deficiency/insufficiency with higher PTH was consistently associated with declining insulin sensitivity and β-cell function and rising glycemia over time.

Previous studies have suggested an association of integrated assessment of vitamin D/PTH with glucose metabolism (41,42). Specifically, in a cross-sectional analysis evaluating 15 obese girls and 15 matched control subjects, Stanley et al. (41) demonstrated that the ratio of PTH to vitamin D was associated with insulin sensitivity and hsCRP. These results were confirmed by another study of 133 obese adolescents that showed an association of the PTH–to–vitamin D ratio with presence of metabolic syndrome (42). Our study further extends this concept by demonstrating an interaction between vitamin D deficiency/insufficiency and PTH on longitudinal changes in glucose metabolism in a prospective cohort. This novel analysis has two key implications. First, it supports an effect of vitamin D status on glucose homeostasis by demonstrating an independent association of the PTH–vitamin D axis with the development of hyperglycemia through the worsening of insulin sensitivity and β-cell function. Second and most importantly, these data highlight the need for assessment of the entire PTH–vitamin D axis when studying the effect of vitamin D on glucose metabolism. Specifically, by showing that vitamin D deficiency/insufficiency has a differential association with glucose metabolism and its underlying physiology depending on the concurrent level of PTH, our study suggests a new perspective for future clinical trials of vitamin D supplementation.

Previous studies evaluating the association of vitamin D and glucose metabolism have yielded inconsistent results. Although lower 25-OH-D has been associated with incident T2DM (7–9,43) and decline in β-cell function over time (10,11) in several observational studies, other investigators have noted conflicting results (13,15,16,18). In addition, interventional studies aiming to evaluate the effect of vitamin D supplementation on glucose metabolism have yielded inconsistent conclusions (19–21,44,45). Physiologic studies have found either no impact of vitamin D supplementation on clamp-derived insulin secretion (21) or an effect restricted to first-phase insulin secretion (44). In the same way, the results of clinical trials have not been consistent or robust in supporting a role for vitamin D supplementation in the prevention of T2DM (46). In a randomized trial of 71 obese men, vitamin D supplementation improved postprandial insulin sensitivity but had no effect on insulin secretion or hepatic insulin resistance (20). Moreover, in a study of 92 adults at risk for diabetes mellitus, supplementation with vitamin D yielded a slight improvement in β-cell function and a marginal effect on glycemic control (45). There are several possible reasons for these inconsistencies such as differences in study populations, diverse methodologies in the assessment of β-cell function and insulin sensitivity, and the potential impact of confounders (such as obesity and outdoor physical activity). However, the results of the current study suggest that the lack of careful documentation of the PTH–vitamin D axis may in part explain the contradictory findings of these previous reports.

Our analyses demonstrate that vitamin D deficiency/insufficiency of sufficient biologic impact to result in an increase in PTH was independently associated with dysglycemia, declining insulin sensitivity, and β-cell dysfunction. These data raise the possibility that previous randomized controlled trials failed to detect a robust effect of vitamin D supplementation on glucose metabolism because they were not performed in the specific patient population that would most benefit from this intervention.

Our findings are supported by biological factors suggesting a link between vitamin D status and glucose homeostasis. First, vitamin D receptors are expressed in pancreatic β-cells and target tissues for insulin action such as skeletal muscle and adipose tissue (9). Second, vitamin D receptor polymorphisms impact insulin secretion and sensitivity in humans (47,48). Finally, the active form of vitamin D, calcitriol, has an effect in modulating calcium influx and gene expression in β-cells (49). Most notably, it has been reported that calcitriol does not impact insulin release when pancreatic islets are under normal conditions, instead requiring a stressed environment such as exposure to pathologic cytokines or vitamin D deficiency for the detection of its effect (49). Thus, it would appear that low 25-OH-D coupled with increased PTH may be a better indicator reflecting the state of vitamin D deficiency/insufficiency that leads to dysregulation of glucose homeostasis.

Our study is robust, as it was performed in a well-characterized cohort undergoing serial metabolic evaluation at a time (first year postpartum when women may be breast-feeding and often indoors to limit infant sun exposure) and geographic location (northern latitude) that may contribute to an increased likelihood of vitamin D deficiency/insufficiency. Accordingly, this cohort provides a unique opportunity to study the impact of PTH–vitamin D status on glucose metabolism. Furthermore, to our knowledge, this is the first study to prospectively evaluate both vitamin D and PTH together in relation to glucose homeostasis and its underlying physiology. A possible limitation of this study is that we did not measure 1,25-dihydroxyvitamin D and vitamin D–binding protein, which could provide a more comprehensive evaluation of this pathway. However, 25-OH-D and PTH are the standard clinical measures for assessment of vitamin D status in practice. In addition, because vitamin D and PTH are associated with adiposity, it is reasonable to consider the possibility that the impact of the vitamin D/PTH axis on glucose metabolism is due to its association with obesity. However, it should be noted that all models were adjusted for BMI and that change in BMI was further evaluated in sensitivity analyses, which supported an independent association between the vitamin D/PTH axis and glucose homeostasis. Another consideration is that the study population consisted of young women in the postpartum period, such that the results should be confirmed in large population-based studies outside this setting. Lastly, β-cell function and insulin sensitivity were assessed with surrogate indices, rather than more time-consuming and cumbersome clamp studies, which would have been difficult to implement in 494 new mothers on two occasions in the first year after delivery. Moreover, ISSI-2 and Matsuda index are validated measures that have been widely used in previous studies (10,11,23,32–35).

In conclusion, we demonstrate that PTH status needs to be considered when evaluating the associations of vitamin D status with glucose homeostasis and its underlying determinants. Specifically, vitamin D deficiency/insufficiency with increased PTH is independently associated with deterioration in insulin sensitivity, β-cell function, and glycemia over time, which was not observed in women with vitamin D deficiency/insufficiency in conjunction with lower PTH. This concept provides novel pathophysiologic insight relevant to the impact of vitamin D status on T2DM and warrants further evaluation in future clinical trials of vitamin D supplementation.

Funding. This study was supported by operating grants MOP-84206 and MHC-115442 from the Canadian Institutes of Health Research and OG-3-11-3300-RR from the Canadian Diabetes Association (CDA). C.K.K. holds a CDA Postdoctoral Fellowship Award. R.R. holds an Ontario Ministry of Research and Innovation Early Researcher Award.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. C.K.K. wrote the first draft of the manuscript, contributed to statistical analysis, contributed to analysis and interpretation of data and revision of the manuscript for intellectual content, and gave approval for submission. B.S. contributed to statistical analysis, contributed to analysis and interpretation of data and revision of the manuscript for intellectual content, and gave approval for submission. A.J.H., P.W.C., M.S., and B.Z. contributed to study conception and design, contributed to analysis and interpretation of data and revision of the manuscript for intellectual content, and gave approval for submission. R.R. contributed to study conception and design, contributed to statistical analysis, contributed to analysis and interpretation of data and revision of the manuscript for intellectual content, and gave approval for submission. R.R. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.