Study subjects
The ethics committee of the Research Institute for Endocrine Sciences approved the study and written informed consent was obtained from all subjects before initiation of the study. Data was obtained from the Tehran Lipid and Glucose Study (TLGS), an ongoing prospective study in Tehran, Iran, initiated in 1998 (17). To date TLGS has completed five phases at 3-year intervals (phase 1: 1999–2001, phase 2: 2002–2005, phase 3: 2005–2008, phase 4: 2008–2011 and phase 5: 2011–2014). Current data are available for five phases, including baseline and four follow-ups. TLGS involves 15,005 subjects, aged ≥ 3 years, who were selected from a geographically defined population using multi-stage cluster sampling. At the time of data collection (both base line and follow-ups) women were interviewed by trained personnel using pretested questionnaires including information on demographic and lifestyle variables, smoking habits, various risk factors for non-communicable diseases, family history of hypertension and medical and reproductive histories; all clinical, and anthropometric parameters included weight and Waist Circumference (WC) were measured by interviewers, as well. WC was measured with an unstretched tape measure at the level of the umbilicus, without any pressure to the body surface. Systolic Blood Pressure (SBP) and Diastolic Blood Pressure (DBP) were measured twice in a seated position after a 15-min rest period. At each visit, women were asked about having any experience of PE, using a validated self-reporting questionnaire (18, 19); when the women were not sure of the diagnosis their medical documents were referred to.
A blood sample was taken from all participants between 7:00 am and 9:00 am after a 12-h overnight fast. Biochemical assessments were performed at the TLGS research laboratory on the day of blood collection. Back up samples were stored at -80 degree of centigrade.
Triglyceride (TG) levels were assayed using glycerol phosphate. Total Cholesterol (TC) was assayed using the enzymatic colorimetric method with cholesterol esterase and cholesterol oxidase. The level of High-density Lipoprotein Cholesterol (HDL-C) was measured after precipitation of the apolipoprotein B (apo B)-containing lipoproteins with phosphotungstic acid. We used a modified Friedewald equation to calculate Low-density Lipoprotein Cholesterol (LDL-C). Fasting plasma glucose (FPG) and 2-hour post-challenge plasma glucose (2h-PCPG) were measured using an enzymatic colorimetric method with glucose oxidase; inter- and intra-assay Coefficients of Variations (CVs) at baseline and follow-up phases were both < 2.3%. All metabolic analyses were performed using related kits and a Selecta 2 autoanalyzer. Intra-assay and inter-assay CVs for TG, TC, HDL-C, and LDL-C were less than 2.1, 1.9, 3, and 3%, respectively.
Serum concentration of AMH was measured in stored samples at the time of recruitment by the two-site enzyme immunoassay (EIA) method using Gen II kit (intra- and inter-assay CVs were 1.9 and 2.0%, respectively) and sunrise ELISA reader. More details on measurements have been previously published elsewhere (10).
Our study population included all women who met the following eligibility criteria: (1) age between 20-50 years, (2) having regular cycles at the time of enrollment, (3) having history of normal fertility and delivery (at least one term pregnancy within one year of stopping contraception) without having a history of pregnancy after Assisted Reproductive Technologies (ART), and (5) having no history of endocrine problems, hysterectomy, oophorectomy, or any other surgeries on ovaries. There were 1015 subjects who met the inclusion criteria; after exclusion of those with uncertain data regarding history of PE, those with history of PE or chronic hypertension at initiation of the study (n=112), data of 781 women remained for inclusion in the present study (Figure 1).
Definitions
The normal-based methodology used to calculate age-specific AMH percentiles has been described previously in detail by Altman and Chitty (20) and Royston and Wright (21, 22). Age-specific AMH was estimated using the exponential–normal three-parameter model. Our previous study depicts cut-off values for women of each specific age for defining the age-specific AMH quartiles (23). PE was defined as SBP > 140 mm Hg or DBP ≥ 90 mm Hg and 24-hour proteinuria ≥ 0.3 g, first observed at > 20 weeks in a one of the pregnancies and was recorded as a self-reporting variable.
Statistical Analysis
All continuous variables were checked for normality using the one-sample Kolmogorov–Smirnoff test, and expressed as mean ± standard deviation, or median with inter-quartile range (IQ25-75). To normalize the distribution of AMH, log transformation was used. Characteristics of women at the time of recruitment were compared between those experienced PE during follow-ups and those who did not experience (non-PE) using either two independent sample t-test or the Mann-Whitney U-test. The categorical variables, expressed as percentages, were compared using Pearson’s test.
A Receiver Operating Characteristics (ROC) curve was used for assessing predictive power of AMH for event of PE, and the Area Under Curve (AUC) was calculated. To explore the association between PE status and serum AMH levels linear regression analysis with censoring, based on the Buckley-James method (on log-transformed AMH) was used and adjusted for the confounders such as smoking status (ever/never), BMI (kg/m2), SBP (mmHg) and family history of hypertension (yes/no). If the AMH level was undetectable (<.16 mg/L), AMH values were censored (14). To compare the potential nonlinear relationship between AMH and age/BMI between those who experienced and those who did not experience PE, fractional polynomial function was used; the relationship has been depicted graphically.
We also used pooled logistic regression to assess the association between dichotomous outcome variable PE and the time-dependent covariates as the data was interval censored and time to PE was not known, and to calculate odds ratios (24). This model treats every interval as a mini follow up study, pools the observations of all intervals together into one pooled sample and does a logistic regression on the pooled dataset; it has been adjusted for above mentioned confounders.
We used Stata Statistical Software (Release 14. College Station, TX: Stata Corp LP) and a P-value of <0.05 was considered statistically significant