Study population

In this study, we used data from the “Snart-Gravid” (“Soon Pregnant”) study, which is a Danish internet-based prospective cohort study of pregnancy planners, designed to examine the influence of lifestyle and behavioral factors on fecundability. The study design and data collection have been described in detail elsewhere [39]. Briefly, participants were recruited and followed via the internet during 2007–2012. Eligible women were aged 18–40 years, in a stable relationship with a male partner, attempting to conceive, and not receiving fertility treatment. After giving informed consent, participants provided their Civil Personal Registration (CPR) number, a unique personal identifier assigned to all Danish citizens at birth. The CPR number permits unambiguous identification and linkage of persons in Danish administrative and medical registries [40]. At enrollment, participants completed a baseline questionnaire with items on demographics, lifestyle and behaviors, and medical and reproductive history, including months of trying to conceive. Participants subsequently completed bimonthly follow-up questionnaires until they reported pregnancy, discontinuation of pregnancy attempts, beginning of fertility treatment, or had been followed for 12 months (end of study observation), whichever came first. Follow-up questionnaires elicited information on changes in relevant exposures and whether pregnancy had occurred.

By August 2012, 6,033 women had enrolled in the study by responding to the baseline questionnaire. We excluded 579 women who did not complete a follow-up questionnaire, 113 repeated entries, 294 women with implausible or missing information on date of last menstrual period, 538 women who had attempted to conceive for more than 11 cycles at enrollment, and 226 women who had been adopted, born after a non-singleton gestation, or had missing data on multiplicity of gestation. In order to obtain uniformly recorded data on gestational age at birth from the Danish Medical Birth Registry (DMBR), we also excluded 1,510 women who were born before January 1, 1978. The remaining 2,773 women were included in the analyses.

Measures of weight at birth

We obtained data on the participants' weight at birth from the DMBR. This registry records over 99% of births in Denmark, reported prospectively by midwives attending the birth [41]. Data on birth weight were registered in categories of 250 grams in 1978, in categories of 10 grams during 1979–1990, and in exact grams after 1990 [42]. We categorized birth weight as <2,500, 2,500–2,999, 3,000–3,999, and ≥4,000 grams, and used 3,000–3,999 grams as the reference category. In additional analyses, we estimated z-scores for birth weight by each completed gestational week as (participant's birth weight – mean of birth weights for the gestational week of birth)/(the standard deviation of the mean of birth weights for the gestational week of birth) [43]. Estimation of mean birth weight and standard deviation in each gestational week was based on the birth weight distribution of Danish girls in the period 1978–1992 (i.e., the period of the participants' births), as registered in the DMBR. The z-scores were then grouped into 6 categories of ≤-2, -2-≤-1, -1-≤0, 0-≤1, 1-≤2, and >2, with 0-≤1 as the reference category.

Measures of time-to-pregnancy (TTP)

The event of interest was participants' report of any pregnancy regardless of outcome. More than 96% of participants used a home pregnancy test to confirm conception [44]. At each follow-up, participants reported the date of their last menstrual period (LMP), whether they were currently pregnant, and occurrence since the previous follow-up of spontaneous abortion, therapeutic abortion, or ectopic pregnancy. Total number of menstrual cycles at risk of pregnancy (i.e., TTP) was calculated as (days of attempt time at study entry/usual cycle length)+(((LMP date from most recent follow-up questionnaire – date of study entry)/usual cycle length)+1). Participants could contribute information until their 12^th cycle of attempted pregnancy to the analysis. Observed cycles at risk of pregnancy were defined as cycles contributed after study enrollment and were left-truncated. Thus, if a woman had already attempted to conceive for 8 cycles when she entered the study, she could contribute up to 4 more cycles after enrollment into the study, with her observed cycles starting at cycle 9 (delayed entry). The follow-up of women who started fertility treatment during follow-up was censored at the cycle in which they started the treatment.

Covariates

We obtained data on participants' gestational age at birth from the DMBR. Data on gestational age were based on the date of the pregnant woman's last menstrual period, corrected by ultrasound examination if performed, and registered in full weeks. Gestational ages of the participants were 28–44 completed weeks. We defined preterm to be a gestational age <37 weeks; full term to be 37–41 weeks; and post-term to be ≥42 weeks. From the DMBR, we also obtained information on participants' mothers' lifetime parity and participants' birth order by using the CPR number to identify mothers and siblings. Siblings born before establishment of the DMBR in 1973 were identified by the mothers' self-reported parity, which was also registered in the DMBR and has high validity [45]. Data on mothers' lifetime parity were divided into categories 1 (study participant was an only child), 2–3 children, and ≥4 children (reference category). Participants' birth order was categorized as first-born, second-born, or greater than second-born (reference category). Data on participants' mothers' history of difficulty conceiving (yes/no) and history of spontaneous abortion (yes/no) were reported in the baseline questionnaire, and we defined participants' mothers without such history as the reference category. Reference categories were defined on the assumption that they represented mothers with normal fecundability.

From the DMBR we obtained data on mother's age and marital status at the time the participant was born. From the Danish National Registry of Patients (DNRP), which includes data on all admissions to Danish non-psychiatric hospitals since 1977, we obtained data on hospital diagnoses of hypertension or pre-eclampsia during the mother's pregnancy with the participant. These diagnoses were coded according to ICD-8 during the period of interest. We used ICD-8 codes 400–404 and 637.00 (essential and gestational hypertension) and 637.03, 637.04, 637.09, 637.19, and 637.99 (pre-eclampsia, eclampsia, and toxemia). Prevalence of hospital admission due to maternal diabetes was below 1%, therefore maternal diabetes as measured by hospitalization was not a strong confounder in our analysis.

From the baseline questionnaire we obtained data on participants' own reproductive history, including age at menarche, cycle regularity, gravidity, parity, and history of unsuccessful pregnancy attempts ≥12 months. At baseline, participants also reported their weight (in kilograms) and height (in centimeters) and we calculated their body mass index (BMI) as (weight (kilograms)/height squared (m²)). Further, data on participants' age, number of cycles of pregnancy attempt at study entry, intercourse frequency, mother's and father's educational level, and mother's smoking during pregnancy were reported in the baseline questionnaire.

Ethics statement

The “Snart-Gravid” study was approved by the Danish Data Protection Board (journal no. 2013-41-1922) and the Institutional Review Board at Boston University. Consent was obtained from the participants before completion of the first questionnaire. Data from the DMBR and the DNRP were retrieved from Statens Serum Institut (http://www.ssi.dk/Sundhedsdataotit.aspx). Data from the ”Snart-Gravid” study are hosted by the Department of Clinical Epidemiology, Aarhus University Hospital; as this study is still in progress, access to the data is not yet freely available. All data were anonymized after retrieval and no CPR numbers were included in the dataset that was the basis of our analyses.

Missing values

The proportion of missing values for the variables birth weight, birth order, mother's lifetime parity, mother's age at delivery, mother's marital status, and mother's smoking during pregnancy ranged from 4.8% to 8.4%. For 17.2% of the participants, values were missing on gestational age at birth, which was partly attributable to procedural changes instituted in 1978 in reporting this variable to the DMBR [46]. For 17.2% and 20.4% of participants, there were missing observations on mother's history of difficulty conceiving and mother's history of spontaneous abortion, respectively, most likely due to participants not knowing this information. For 30.4% and 35.0% of participants, there were missing observations on mother's and father's educational level, respectively. These missing data resulted from random assignment of half of the early study participants to a short-form baseline questionnaire that did not include questions on parental educational level.

On the assumption that data were missing at random, we imputed missing values using multiple imputation by chained equations (MICE program in Stata version 12.0). We included 36 variables in the imputation, including all variables used in the substantive analyses, and imputed five data sets. Distributions of continuous variables were examined by histograms and box plots. Variables that diverged from the normal distribution were transformed to the log-scale before imputation.

Data analysis

We calculated Kaplan-Meier estimates to assess the cumulative probability of conception within 3, 6, and 12 menstrual cycles, accounting for delayed entry using left-truncation, and losses to follow-up and other reasons for censoring (e.g., no longer trying to conceive or initiation of fertility treatment). We described the distribution of participants' characteristics (for women lost to follow-up, women who completed the study, and for all of the 2,773 women in the study cohort) according to weight at birth. Using a proportional probabilities model, we then estimated fecundability ratios (FR) and 95% confidence intervals (CI) for categories of birth weight (<2,500, 2,500–2,999, and ≥4,000 grams, with 3,000–3,999 grams as the reference category), using TTP measured in cycles. The FR of any two groups was calculated as the ratio of their cycle-specific probabilities [47]. Participants contributed cycles at risk from entry into the study until report of pregnancy, receipt of fertility treatment, discontinuation of pregnancy attempt, loss to follow-up, or end of observation (maximum 12 cycles). Distinct intercept parameters were included for each of the 12 cycles of follow-up, to allow for decline in the baseline conception rate over follow-up time.

We examined potential interaction between weight and gestational age at birth by including product terms for gestational age as a continuous variable in the regression model, and found no evidence of interaction. Adjustments were made in three steps: first, we adjusted for year of birth and gestational age as a continuous variable with values 28–44 weeks only (model 1); second, we included parental socio-demographic and medical characteristics (mother's age, mother's marital status, mother's and father's educational level, mother's smoking during pregnancy, and mother's history of hypertension and pre-eclampsia) (model 2); and third, we included markers of the participant's mother's fecundability in the regression model (mother's lifetime parity, participant's birth order, mother's history of difficulty conceiving, and mother's history of spontaneous abortion) (model 3). Variables included in the three models were chosen a priori because they have previously been associated with offspring weight at birth [13], [15], [33]–[38], [48]–[51], and may influence the daughter's fecundability [13], [15], [27]–[32], [52], [53]. Not much is known about the potential influence of maternal reproductive health on the fecundability of daughters. Based on evidence of familial clustering of other reproductive health outcomes [21]–[32], it is plausible that proxy markers of the mother's fecundability, e.g., mother's history of difficulty conceiving, might be causally associated with daughters' fecundability. On this basis, we investigated the potential confounding effect of maternal socio-demographic, medical and reproductive characteristics. We repeated the analyses restricted to women born at term, i.e., at 37–41 weeks of gestation, to restrict the influence of gestational age at birth. To evaluate sensitivity of the study result to inclusion of women who had tried to conceive for up to 11 cycles at study entry, we repeated the analyses restricted to women with only up to 6 cycles of attempt time. Previous reports indicate that accelerated weight gain in infancy, which is often exhibited by infants with a low birth weight, is associated with overweight or obesity later in life [54], [55], and obesity has been linked with reduced fecundability [56]. Thus, we also considered the potential mediating influence of pre-pregnancy BMI on an association between weight at birth and fecundability.

In addition to considering gestational age at birth by adjustment and restriction to term births, we also examined the association between weight at birth and fecundability by z-scores of birth weight, to compare infants of differing relative weights by using weight estimates that were adjusted for gestational age at birth [43]. We estimated fecundability ratios by categories of z-score (≤-2, -2-≤-1, -1-≤0, 1-≤2, and >2, with 0-≤1 as the reference category), using the same proportional probabilities regression model as in the initial analyses.

Analyses were performed using Stata version 12.0 (StataCorp., TX, USA) and SAS version 9.2 (Cary, NC, USA).