This study was based on the Norwegian Mother and Child Cohort Study (MoBa), conducted by the Norwegian Institute of Public Health [9
]. A total of 235,577 women in Norway were invited to participate in MoBa. Women were enrolled from 1999–2008 and received an invitation to participate in MoBa through the mail with their notice of an appointment for a routine ultrasound, given to all women in Norway around week 17–18 of gestation. The participation rate was 38.5%, resulting in 90,697 women who participated in MoBa, and 106,977 pregnancies. Additional details can be found at www.fhi.no/morogbarn
. MoBa was approved by The Regional Committee for Medical Research Ethics and the Norwegian Data Inspectorate. Informed consent was obtained from each participant.
Mothers completed a questionnaire around 17 weeks of gestation. Maternal demographics (age, height, prepregnancy weight, and education), lifestyle factors (physical activity, smoking, and alcohol use three months before pregnancy; frequency of intercourse one month before pregnancy), and maternal diseases (hypo- or hyperthyroidism, fallopian tube infection, endometriosis, and sexually transmitted diseases) were obtained from this questionnaire. Maternal prepregnancy body mass index (BMI) was constructed using the self-reported values from this questionnaire. The questionnaire also included questions related to reproductive history (cycle length, irregular cycles in the last year, parity, history of fertility treatments, and time-to-pregnancy (TTP)). Regarding TTP, women were first asked whether they planned their pregnancy. If they answered positively, they were asked a series of questions regarding how long it took them to achieve pregnancy including “how many months did you have regular intercourse without contraception before you became pregnant?” Women chose ‘<1 month’, ‘1–2 months’, or ‘3 months or more’. If they answered ‘3 months or more’, they were further prompted to list the number of months.
Data on MoBa pregnancies were linked with data in the Medical Birth Registry of Norway (MBRN). The MBRN data were collected by the physician or midwife attending the childbirth, who had access to the pregnancy medical record [10
]. Data from the MBRN were used to ascertain women’s diabetes status (pregestational type 1 diabetes, pregestational type 2 diabetes, unspecified pregestational diabetes, or gestational diabetes).
Eligibility for the present analysis was restricted to a woman’s first pregnancy registered with MoBa (that is, a woman’s first study pregnancy which may not be her first lifetime pregnancy). A total of 77,820 pregnancies were eligible. Further exclusions were made as shown in . Thus, a total of 58,004 pregnancies, including 221 type 1 diabetic women and 88 type 2 diabetic women, were included in the analysis.
Flow chart for subject inclusion based on MoBa Database Version 4.301, restricted to MoBa first pregnancies.
The logistic analog of the proportional probability model, a Cox-like discrete-time model [11
], was used to estimate fecundability odds ratios (FOR) and 95% CIs for associations between type 1 diabetes or type 2 diabetes and fecundability (odds of conception). The equation for this model is given in the online appendix
. This is a standard statistical method used in studies of fecundability [12
] and was implemented using PROC LOGISTIC in SAS (SAS Institute Inc., Cary, NC). For each month that contributes to the analysis, the FOR measures the fecundability in the exposed (women with type 1 diabetes or type 2 diabetes) divided by the fecundability in the unexposed (women with no diabetes), conditional on not having conceived in any previous month. The FOR is the weighted average of the estimates across all contributing months. An FOR of less than one indicates reduced fecundability (increased TTP). The unit of analysis was the woman-month and each woman contributed up to 12 months to the analysis. Women’s reported TTP was used to construct an analysis dataset with multiple records per woman (one record for each month she was trying to conceive). As recommended for avoiding medical intervention bias, TTP was censored at 13 months because women are considered clinically infertile after 12 months of unsuccessful pregnancy attempts [13
]. For example, if a woman reported getting pregnant after 14 months, only her first 12 (unsuccessful) months contributed to the analysis. Further, women who received infertility treatment at or before 12 months (2.0%) had their TTP censored at TTP-1 month. Thus, if a woman reported getting pregnant after nine months, and reported receiving fertility treatments, only her first (unsuccessful) eight months contributed to the analysis.
Maternal age and prepregnancy BMI (kg/m2) were assumed to be important confounders and included in the model a priori. Maternal age was modeled as a continuous variable while prepregnancy BMI was modeled using its cubic form and all lower order terms (BMI, BMI2, and BMI3, all continuous). We further considered as potential confounders variables that may be associated with both diabetes and fecundability, conditional on maternal age and prepregnancy BMI. These variables include: maternal education, physical activity, smoking and alcohol use, and frequency of intercourse. A backward selection method was used whereby the additional covariates were dropped from the full model and the change in the effect estimate for diabetes and fecundability was assessed in a step-wise manner. Because the effect estimates changed by less than 10% with each deletion, none of these variables were deemed confounders. To further evaluate the potential confounding effects of maternal diseases (fallopian tube infection, endometriosis, sexually transmitted disease, and hypo- or hyperthyroidism) without sacrificing degrees of freedom in the model, we examined the association between diabetes and fecundability after excluding women with these conditions (one at a time). Because the FORs for both type 1 and type 2 diabetes did not change, we concluded that these variables were not confounders. Interaction terms were fit to test effect modification between parity (yes/no) and diabetes, and age (continuous) and diabetes. To help determine whether interaction was present, we used an alpha level equal to 0.20 as a screening criterion. The p-values for interaction between parity and diabetes (p=0.61 for type 1 diabetes and p=0.68 for type 2 diabetes) and age and diabetes (p=0.44 for type 1 diabetes and p=0.65 for type 2 diabetes) all fell well above this cutoff, providing no support for interaction. All analyses were completed using SAS version 9.2 (SAS Institute Inc., Cary, NC).
We completed two additional sub-analyses. In the first, we recalculated FORs after excluding women who reported irregular menstrual cycles; in the second, we recalculated FORs after excluding women who reported irregular menstrual cycles and we based the TTP calculation on the estimated number of cycles it took to get pregnant (instead of the number of months). For the analysis using cycle lengths, those who reported cycle length greater than 60 days (0.2%) were excluded which also likely excluded women with polycystic ovary syndrome (PCOS).
Additional sensitivity analyses were conducted and the results are presented in Figure 1 of the online appendix
. These sensitivity analyses assessed planning bias (bias resulting from excluding women who did not plan their pregnancy and who may have higher fecundability than planners [13
]), medical intervention bias (bias that may result if the exposure under study is associated with the probability that women will receive successful medical help for infertility [13
]), and wantedness bias (bias which occurs when women who did not plan their pregnancy respond that the pregnancy was in fact planned when asked about it in retrospect [13