Study population
This secondary analysis includes women participating in Project Viva, an ongoing prospective cohort of mother-child pairs recruited between 1999 and 2002 from Atrius Harvard Vanguard Medical Associates at around ten weeks gestation. Details on recruitment and eligibility are described elsewhere [15]. Our sample included 2276 women with singleton pregnancies who had a live newborn (n = 2100) or a pregnancy loss (n = 176). For 30 women who participated with two different pregnancies, we only considered the first pregnancy enrolled in the study. We further excluded women younger than 18 years at enrollment (n = 30) and those with preexistent chronic hypertension, type 1, or type 2 diabetes (n = 45). Our final analytical sample included 2201 women with data on the exposure and at least one pregnancy outcome. All participants provided written informed consent at enrollment. The institutional review board of Harvard Pilgrim Health Care approved all study protocols in line with ethical standards established by the Declaration of Helsinki.
Exposure: infertility for the index pregnancy
Our primary exposure was history of infertility for the index pregnancy (yes/no), assessed via three sources of information. First, we used questionnaire data from the first study visit inquiring on whether they were actively trying to become pregnant and, if so, the number of cycles it took them to become pregnant. We classified women reporting ≥12 cycles to become pregnant, or ≥ 6 cycles if ≥35 years of age, as infertile [2]. Second, we used information obtained from the women’s medical records on history of infertility – specifically, whether they had a diagnosis of infertility (International Classification of Diseases-9 code 628.9 entries before the last menstrual period [LMP] + 60 days), a claim for infertility consultation or services, or prescriptions for fertility medications (e.g., clomiphene citrate, gonadotropins, or gonadotropin-releasing hormone agonists before LMP + 14 days). Lastly, women completed a detailed reproductive history on a follow-up questionnaire administered ~18 years after delivery. In this questionnaire, participants were asked if it had taken them ≥12 months to become pregnant, or ≥ 6 cycles if ≥35 years of age, or if they had used medical treatment for this purpose in each of their past pregnancies. If they responded “yes” to either question for the index pregnancy, we classified them as having history of infertility. If the time to pregnancy reported in this questionnaire was different from the information reported at the first study visit, we only considered the latter.
To identify whether the associations differed between subgroups defined by the use of MAR, we further classified women with infertility as those who used vs. did not use MAR. The former category included women who had prescriptions for fertility medications abstracted from the medical records (e.g., clomiphene citrate, gonadotropins, or gonadotropin-releasing hormone agonists), or who, at the 18-year study visit, reported the use fertility medications or treatments for the index pregnancy (e.g., medications to induce ovulation such as clomiphene, gonadotropic injections, intrauterine insemination, or assisted reproductive technology [ART] including in-vitro fertilization [IVF], intracytoplasmic sperm injection, sperm donation, egg or embryo donation, other). The latter category included women who did not report any treatment to conceive and did not have prescriptions for fertility medications in the medical records.
Outcomes: pregnancy outcomes
We were interested in several maternal and newborn outcomes. Women underwent a clinical glycemic screening at 26–28 weeks of gestation with a non-fasting glucose challenge test (GCT) [16]. The GCT consisted of the administration of a 50-g oral glucose load with venous blood sampled 1 h after the load. If the blood glucose was >140 mg/dL, the clinician referred the woman for a fasting 3-h, 100-g oral glucose tolerance test (OGTT). Abnormal OGTT results were a blood glucose >95 mg/dL at baseline, >180 mg/dL at 1 h, >155 mg/dL at 2 h, or > 140 mg/dL at 3 h. We classified women into the following categories: normoglycemic, isolated hyperglycemia (e.g., an abnormal GCT but a normal OGTT), impaired glucose tolerance (IGT) (e.g., one abnormal value on the OGTT), and GDM (e.g., at least two abnormal values on the OGTT).
We identified women with gestational hypertension (GH) or preeclampsia from the outpatient and hospital medical records. GH included women who did not have chronic hypertension but developed elevated systolic blood pressure (SBP, >140 mmHg) or diastolic blood pressure (DBP, >90 mmHg) on ≥2 occasions after 20 weeks gestation. We classified women as having preeclampsia if they did not have chronic hypertension but developed high SBP (>140 mmHg) or DBP (>90 mmHg) in addition to proteinuria or if they had chronic hypertension and developed proteinuria after 20 weeks gestation [17]. We combined GH and preeclampsia for analysis. We extracted clinical measures of blood pressure from the medical records and calculated average SBP and DBP within each trimester of pregnancy. We defined the first trimester as LMP to 91 days, the second trimester as 92 to 182 days, and the third trimester as 183 days to delivery. For analyses of average blood pressure, we present results for both SBP and DBP but focus the interpretation of results on SBP because it is a stronger predictor of long-term cardiovascular outcomes than DBP [18,19,20].
Using clinical prenatal weights, we calculated total gestational weight gain (GWG) as the difference between the last pregnancy weight, measured <4 weeks from delivery, and self-reported pre-pregnancy weight. Then, we categorized GWG as inadequate, adequate, or excessive based on the Institute of Medicine guidelines [21].
We obtained information on the newborn’s sex, birthweight, and delivery date from the medical delivery records. We calculated birthweight-for-gestational age and sex z-scores (BWZ) based on US national reference data [22]. We categorized it in tertiles due to the relatively small sizes in some cells when using conventional categories of birth size (small-, appropriate-, and large-for-gestational-age). We calculated gestational age at birth in weeks by subtracting the date of the LMP from the date of delivery; we used the 2nd-trimester ultrasound in cases where the estimated delivery date by LMP differed by >10 days [17]. If infants were born <37 weeks of gestation, we classified them as preterm. We categorized birth outcome as a live birth or pregnancy loss (stillbirth or miscarriage) based on information from the outpatient and hospital medical records and study disenrollment data.
Covariates
At enrollment, women reported their age, race/ethnicity, education level, marital status, annual household income, parity, and prenatal smoking habits via a self-administered questionnaire. We calculated pre-pregnancy body mass index (BMI, kg/m2) from self-reported pre-pregnancy weight and height. At a study visit conducted ~13 years after enrollment, the participants provided information on age at their first menstrual period.
Statistical analysis
Prior to the formal analysis, we assessed the distribution of maternal characteristics across categories of infertility and compared them using mean (standard deviation [SD]) for numerical variables or frequencies and proportions for categorical variables.
In our primary analysis, we examined the associations between infertility (yes vs. no [reference]) and pregnancy outcomes using logistic regression models for HDP, preterm birth and birth outcome, multinomial logistic regression for gestational glucose tolerance status, GWG, tertiles of BWZ, and linear mixed regression models for SBP and DBP across the three trimesters of pregnancy. We selected potential confounders and precision covariates for model adjustment based on a priori knowledge and a literature review. The variables under study are depicted in a Directed Acyclic Graph included as a supplemental figure (Fig. S1). The final inclusion of covariates in the models was based on bivariate associations with the exposure/outcomes, and their impact on the effect estimates. For each of the outcomes, we constructed a series of multivariable models. Model 1 included age at enrollment (18–29, 30–34, ≥35 years), race/ethnicity (white, Black, Asian, Hispanic, other), and age at menarche (<12, 12–14, ≥15 years). Model 2 was additionally adjusted for pre-pregnancy BMI (continuous) and prenatal smoking habits (former smoker, smoker during pregnancy, never smoker). We evaluated the presence of an interaction between age at enrollment and infertility using an interaction term, but this was not significant for any of the outcomes (p > 0.05); therefore, we did not include the interaction term. For all the outcomes, additional adjustment for education, marital status, household income, family history of type 2 diabetes (for glucose tolerance status), and family history of hypertension (for HDP/blood pressure) did not influence the results substantially; therefore, we did not include these variables in the final models.
For SBP and DBP, we used linear mixed regression models to account for the repeated blood pressure measurements across pregnancy. We had up to 3 repeated measurements per woman (e.g., trimester-specific averages). These models included a random intercept and slope, with unstructured covariance to account for within-woman correlations. The interaction between infertility and the trimester of pregnancy at blood pressure assessment was not significant; hence, we did not include it in the final model. The estimates from these models may be interpreted as the mean difference in SBP and DBP across pregnancy with respect to baseline infertility status.
In a secondary analysis, we classified women as having history of infertility with and without MAR vs. those without infertility as the referent. We evaluated associations with those outcomes for which we detected significant associations in the primary analysis. We considered models adjusted for the same covariates previously described. In our last analysis, we subclassified women with MAR by the type of medication reported, which in most cases were medications to induce ovulation. We subclassified medications into three groups: 1) clomiphene citrate (CC) (alone or with gonadotropins or gonadotropin-releasing hormone [GnRH] agonists, 2) gonadotropins or GnRH agonists without clomiphene, 3) other (unspecified medications to induce ovulation, other treatment or no treatment specified). For this analysis, we compared the associations of each of the three groups vs. two different referent groups: women without history of infertility and women with history of infertility without MAR, to parse out the specific associations of each medication from the global association of infertility.
We conducted two sensitivity analyses. First, we excluded women classified as having history of infertility (n = 2) (primary analysis) and history of infertility with MAR (n = 10) (subgroup analyses) based solely on information reported ~18 years after delivery, hence more likely subject to recall bias and misclassification. The results were identical to those of the main analyses; therefore, only these are presented. Second, we excluded 39 women diagnosed with polycystic ovary syndrome (PCOS) before the index pregnancy to assess whether this condition could explain our main findings (diagnosis abstracted from medical records or self-reported at the 18-year study visit).
To deal with missing covariate data, we conducted chained equation multiple imputation to generate 50 imputed data sets using an imputation model that included the exposure, outcomes, and covariates under study. Missingness varied from <2% for race/ethnicity and pre-pregnancy BMI to 52% for age at menarche. The imputed data sets were combined and analyzed using MI ESTIMATE in Stata 16.
We conducted all the analyses in Stata 16 (StataCorp L.P., College Station, Texas).