Study design

Data were derived from Pregnancy Study Online (PRESTO), an Internet-based preconception cohort study of pregnancy planners in the United States and Canada. Study methodology is described in detail elsewhere. In brief, female participants were recruited via banner advertisements on social media and health-related Web sites. Women were eligible for the study if they were aged 21–45 years, not using contraception or receiving fertility treatment, in a stable relationship with a male partner, and not pregnant. The study was approved by the institutional review board and all participants provided informed consent.

All questionnaires were completed online. The baseline questionnaire assessed demographics, medical and reproductive history, physical and mental health, current and past medication use, lifestyle habits (eg, alcohol use), and fertility-related factors. To assess pregnancy status and changes in exposures, participants completed shorter questionnaires every 8 weeks for up to 12 months or until reported conception, initiation of fertility treatment, or loss to follow-up. Over 80% of women completed at least 1 follow-up questionnaire.

Assessment of depression and anxiety

At baseline, participants reported depression symptoms during the past 2 weeks via the 12-item Major Depression Inventory (MDI). A total score for the MDI was calculated using standard scoring criteria and categories of depression symptomatology (eg, none/low [<20], mild [20–24], moderate [25–29], severe [≥30] were created based on established cut points corresponding to International Statistical Classification of Diseases, 10th Revision . The MDI total score can range from 0–50. Depression and anxiety diagnoses before study entry were self-reported in the baseline survey with 2 separate questions (ie, “Have you ever been diagnosed with depression?” and “Have you ever been diagnosed with anxiety/panic disorder?”).

Assessment of psychotropic medication use

In baseline and follow-up surveys, participants were asked about psychotropic medication use separately for depression and anxiety as well as use of any medications for other conditions (eg, sleep disorders). Information was collected on ever use, total years of use, the name of the medication taken for the longest and most recently (open text; coded for analysis by the first author), and medication use in the past 4 weeks. Given the large overlap between medications prescribed for depression, anxiety, and other conditions, we categorized psychotropic medication as never, former, and current use, regardless of diagnosis. Psychotropic medications were defined as: anxiolytics, anticonvulsants, antipsychotics, atypical antidepressants, benzodiazepines, beta-blockers, mood stabilizers, sedative-hypnotics, selective serotonin-norepinephrine reuptake inhibitors, selective serotonin reuptake inhibitors (SSRIs), stimulants, tetracyclic antidepressants, and tricyclic antidepressants. We also categorized former and current users separately by class of psychotropic medications for the most frequently used medications in our cohort (ie, SSRIs, benzodiazepines, and other). Finally, we categorized time since last psychotropic medication use among former users (cessation of use: 1–5, 6–11, ≥12 months ago) and compared them with never users.

Assessment of covariates

At baseline, women reported data on age, race/ethnicity, education, income, height, weight, physical activity, parity, smoking, alcohol intake, intercourse frequency, and last method of contraception. We calculated body mass index (BMI) as weight (kg)/height (m) ² . Total metabolic equivalents (METs) of physical activity were calculated by multiplying the average number of hours per week spent participating in moderate and vigorous activities by 3.5 and 7 METs, respectively. We updated information on frequency of intercourse over time using data from the follow-up questionnaires.

Assessment of pregnancy and cycles at risk

At baseline, women reported whether their menstrual cycles were regular (“in a way that you can usually predict when your next period will start”) and how long their cycles usually lasted (eg, 28 days). Cycle length for women with irregular cycles was estimated based on baseline last menstrual period (LMP) date, expected date of next period, and LMP dates recorded over follow-up. On follow-up questionnaires, participants reported their LMP date, whether they had conceived since their last follow-up, and if so, how they confirmed the pregnancy (eg, home pregnancy test). Total cycles at risk were calculated using the following formula: (cycles of attempt at study entry) + ([{LMP from most recent follow-up questionnaire – date of baseline questionnaire completion}/usual cycle length] + 1). Women contributed observed cycles to the analysis from enrollment until conception, initiation of fertility treatment, loss to follow-up, or 12 cycles, whichever came first.

Exclusions

During 33 months of enrollment, 3089 women completed the baseline questionnaire. From these, we excluded women who had been trying to conceive for >6 cycles at study entry (N = 341), did not complete any follow-up questionnaires (N = 513), had insufficient or implausible information on baseline LMP or first pregnancy attempt (N = 60), or were pregnant at baseline (N = 29), resulting in a study population of 2146 women.

Data analysis

We used a proportional probabilities regression model to estimate the FR, a measure of the per-cycle probability of conception in exposed compared with unexposed women, and its 95% CI. This model incorporates the decline in baseline fecundability over time using binary indicators of cycle number at risk. Left truncation was accounted for using the Anderson-Gill data structure. The model is fit using PROC GENMOD in SAS.

Confounders were selected a priori based on a literature review and the drawing of a causal diagram. Associations were adjusted for age (<25, 25–29, 30–34, ≥35 years), race/ethnicity (white/non-Hispanic, non-white or Hispanic), education (less than college degree, college degree, graduate school), annual household income (<$50,000, $50,000–$99,999, ≥$100,000), parity (parous, nulliparous), BMI (<25, 25–29, 30–34, ≥35 kg/m ² ), smoking history (never, former, current occasional, current regular), physical activity (<10, 10–19, 20–39, ≥40 MET h/wk), alcohol use (<1, 1–6, 7–13, ≥14 drinks/wk), intercourse frequency (<1, 1, 2–3, ≥4 times/wk), and last method of contraception (hormonal methods, barrier methods, withdrawal/rhythm methods). First, we examined the association between baseline exposures and fecundability. Then, we conducted secondary analyses where variables for psychotropic medication use were updated over time using data from bimonthly follow-up questionnaires. In addition, we fit restricted cubic spline models to account for the possibility of a nonlinear relationship between MDI score and fecundability.

Given that intercourse frequency and cycle irregularity are potential mediators of the associations of interest, models were run with and without controlling for these variables.

We used PROC MI in SAS to impute missing values for exposures and covariates by creating 5 imputed data sets. The prediction model included 137 demographic, lifestyle, behavioral, and reproductive variables to impute missing values. We used PROC MIANALYZE in SAS to combine coefficient and SE estimates from the 5 data sets. The prevalence of missing data ranged from 0% (age) to 3.7% (income). The prevalence of missing data on depression, anxiety, and psychotropic medication use was <1%.

Women who had been trying to conceive for many cycles before study entry may have experienced an increase in depressive symptoms as a result of delayed conception. To evaluate the potential for selection bias and reverse causation, we stratified the analysis by cycles of attempt at study entry (<3 vs ≥3 cycles).

Study design

Data were derived from Pregnancy Study Online (PRESTO), an Internet-based preconception cohort study of pregnancy planners in the United States and Canada. Study methodology is described in detail elsewhere. In brief, female participants were recruited via banner advertisements on social media and health-related Web sites. Women were eligible for the study if they were aged 21–45 years, not using contraception or receiving fertility treatment, in a stable relationship with a male partner, and not pregnant. The study was approved by the institutional review board and all participants provided informed consent.

All questionnaires were completed online. The baseline questionnaire assessed demographics, medical and reproductive history, physical and mental health, current and past medication use, lifestyle habits (eg, alcohol use), and fertility-related factors. To assess pregnancy status and changes in exposures, participants completed shorter questionnaires every 8 weeks for up to 12 months or until reported conception, initiation of fertility treatment, or loss to follow-up. Over 80% of women completed at least 1 follow-up questionnaire.

Assessment of depression and anxiety

At baseline, participants reported depression symptoms during the past 2 weeks via the 12-item Major Depression Inventory (MDI). A total score for the MDI was calculated using standard scoring criteria and categories of depression symptomatology (eg, none/low [<20], mild [20–24], moderate [25–29], severe [≥30] were created based on established cut points corresponding to International Statistical Classification of Diseases, 10th Revision . The MDI total score can range from 0–50. Depression and anxiety diagnoses before study entry were self-reported in the baseline survey with 2 separate questions (ie, “Have you ever been diagnosed with depression?” and “Have you ever been diagnosed with anxiety/panic disorder?”).

Assessment of psychotropic medication use

In baseline and follow-up surveys, participants were asked about psychotropic medication use separately for depression and anxiety as well as use of any medications for other conditions (eg, sleep disorders). Information was collected on ever use, total years of use, the name of the medication taken for the longest and most recently (open text; coded for analysis by the first author), and medication use in the past 4 weeks. Given the large overlap between medications prescribed for depression, anxiety, and other conditions, we categorized psychotropic medication as never, former, and current use, regardless of diagnosis. Psychotropic medications were defined as: anxiolytics, anticonvulsants, antipsychotics, atypical antidepressants, benzodiazepines, beta-blockers, mood stabilizers, sedative-hypnotics, selective serotonin-norepinephrine reuptake inhibitors, selective serotonin reuptake inhibitors (SSRIs), stimulants, tetracyclic antidepressants, and tricyclic antidepressants. We also categorized former and current users separately by class of psychotropic medications for the most frequently used medications in our cohort (ie, SSRIs, benzodiazepines, and other). Finally, we categorized time since last psychotropic medication use among former users (cessation of use: 1–5, 6–11, ≥12 months ago) and compared them with never users.

Assessment of covariates

At baseline, women reported data on age, race/ethnicity, education, income, height, weight, physical activity, parity, smoking, alcohol intake, intercourse frequency, and last method of contraception. We calculated body mass index (BMI) as weight (kg)/height (m) ² . Total metabolic equivalents (METs) of physical activity were calculated by multiplying the average number of hours per week spent participating in moderate and vigorous activities by 3.5 and 7 METs, respectively. We updated information on frequency of intercourse over time using data from the follow-up questionnaires.

Assessment of pregnancy and cycles at risk

At baseline, women reported whether their menstrual cycles were regular (“in a way that you can usually predict when your next period will start”) and how long their cycles usually lasted (eg, 28 days). Cycle length for women with irregular cycles was estimated based on baseline last menstrual period (LMP) date, expected date of next period, and LMP dates recorded over follow-up. On follow-up questionnaires, participants reported their LMP date, whether they had conceived since their last follow-up, and if so, how they confirmed the pregnancy (eg, home pregnancy test). Total cycles at risk were calculated using the following formula: (cycles of attempt at study entry) + ([{LMP from most recent follow-up questionnaire – date of baseline questionnaire completion}/usual cycle length] + 1). Women contributed observed cycles to the analysis from enrollment until conception, initiation of fertility treatment, loss to follow-up, or 12 cycles, whichever came first.

Exclusions

During 33 months of enrollment, 3089 women completed the baseline questionnaire. From these, we excluded women who had been trying to conceive for >6 cycles at study entry (N = 341), did not complete any follow-up questionnaires (N = 513), had insufficient or implausible information on baseline LMP or first pregnancy attempt (N = 60), or were pregnant at baseline (N = 29), resulting in a study population of 2146 women.

Data analysis

We used a proportional probabilities regression model to estimate the FR, a measure of the per-cycle probability of conception in exposed compared with unexposed women, and its 95% CI. This model incorporates the decline in baseline fecundability over time using binary indicators of cycle number at risk. Left truncation was accounted for using the Anderson-Gill data structure. The model is fit using PROC GENMOD in SAS.

Confounders were selected a priori based on a literature review and the drawing of a causal diagram. Associations were adjusted for age (<25, 25–29, 30–34, ≥35 years), race/ethnicity (white/non-Hispanic, non-white or Hispanic), education (less than college degree, college degree, graduate school), annual household income (<$50,000, $50,000–$99,999, ≥$100,000), parity (parous, nulliparous), BMI (<25, 25–29, 30–34, ≥35 kg/m ² ), smoking history (never, former, current occasional, current regular), physical activity (<10, 10–19, 20–39, ≥40 MET h/wk), alcohol use (<1, 1–6, 7–13, ≥14 drinks/wk), intercourse frequency (<1, 1, 2–3, ≥4 times/wk), and last method of contraception (hormonal methods, barrier methods, withdrawal/rhythm methods). First, we examined the association between baseline exposures and fecundability. Then, we conducted secondary analyses where variables for psychotropic medication use were updated over time using data from bimonthly follow-up questionnaires. In addition, we fit restricted cubic spline models to account for the possibility of a nonlinear relationship between MDI score and fecundability.

Given that intercourse frequency and cycle irregularity are potential mediators of the associations of interest, models were run with and without controlling for these variables.

We used PROC MI in SAS to impute missing values for exposures and covariates by creating 5 imputed data sets. The prediction model included 137 demographic, lifestyle, behavioral, and reproductive variables to impute missing values. We used PROC MIANALYZE in SAS to combine coefficient and SE estimates from the 5 data sets. The prevalence of missing data ranged from 0% (age) to 3.7% (income). The prevalence of missing data on depression, anxiety, and psychotropic medication use was <1%.

Women who had been trying to conceive for many cycles before study entry may have experienced an increase in depressive symptoms as a result of delayed conception. To evaluate the potential for selection bias and reverse causation, we stratified the analysis by cycles of attempt at study entry (<3 vs ≥3 cycles).