Antibiotics and Fecundability Among Female Pregnancy Planners

A Prospective Cohort Study

Holly Michelle Crowe; Amelia Kent Wesselink; Lauren Anne Wise; Tanran R. Wang; Charles Robert Horsburgh Jr.; Ellen Margrethe Mikkelsen; Elizabeth Elliott Hatch


Hum Reprod. 2021;36(10):2761-2768. 

In This Article

Materials and Methods

Pregnancy Study Online (PRESTO) is an ongoing, web-based preconception cohort study. The study methods have been described in detail elsewhere (Wise et al., 2015). Briefly, women are eligible for participation if they are aged 21–45 years, residing in the USA or Canada, and not using contraception or fertility treatment. Participants complete an online baseline questionnaire with items on demographics, lifestyle, reproductive history, and medical history (including medication use). They complete follow-up questionnaires every 8 weeks for up to 12 months to ascertain pregnancy status. This study was approved by the Institutional Review Board at the Boston University Medical Campus, and online informed consent was obtained from all participants.

From June 2013 through September 2020, 11 970 eligible women completed the baseline questionnaire. We excluded 138 women whose date of last menstrual period (LMP) at baseline was more than 6 months in the past, and 31 women with missing/implausible LMP data. We then excluded 2500 women who had been trying to conceive for more than six cycles at baseline, to reduce the potential for reverse causation (e.g. subfertility causing changes in medication use). The final data set thus included 9524 women. The questions pertaining to antibiotic type and indication were added to the PRESTO questionnaires in March 2016; therefore, the antibiotic type and indication analyses are restricted to the 7111 participants completing the baseline questionnaire in March 2016 or later. This complete case analysis included 59% of the overall sample and 68% of antibiotic users.

Assessment of Exposure

Women were asked at baseline and follow-up if they had taken antibiotics in the past 4 weeks. If they reported antibiotic use in March 2016 or later, they were asked to provide the name of the antibiotic in free text boxes. Free text antibiotic names were classified by active ingredient. Women were also asked the reason for their antibiotic prescription. Free text antibiotic indication responses were classified by type of infection (respiratory, urinary tract, vaginal, skin, pelvic, surgical, or other). Antibiotic indication categories were not mutually exclusive, and some women reported more than one indication for a single antibiotic prescription.

Assessment of Outcome

At baseline, participants reported their LMP date, usual cycle length, and number of cycles they had attempted to conceive prior to enrollment. On each follow-up questionnaire, participants were asked for their most recent LMP date and whether they had become pregnant since the previous questionnaire. Total discrete cycles at risk were calculated as follows: cycles of attempt at study entry + [(LMP from most recent follow-up questionnaire−date of baseline questionnaire completion)/usual cycle length] +1. Participants contributed cycles of observation from baseline until they conceived or experienced a censoring event, defined as loss to follow-up, withdrawal, initiation of fertility treatment, no longer trying to conceive, or 12 cycles, whichever came first. Notably, 49 women reported taking an antibiotic for infection prophylaxis following or in preparation for a hysterosalpingogram, a procedure to assess whether the fallopian tubes are blocked (García-Velasco et al., 2017). These women were censored at the cycle when the hysterosalpingogram was reported to avoid bias due to reverse causation.

Assessment of Covariates

We selected potential confounders a priori based on available literature. Information collected at baseline included age, education, income, BMI, multivitamin use, current smoking, perceived stress score (PSS-10) (Cohen et al., 1983), major depressive inventory (MDI) score (Bech et al., 2001), efforts to improve the chances of conception (e.g. charting menstrual cycles, ovulation testing), intercourse frequency, recent irregular menstrual cycles when not using birth control, history of spontaneous abortion, history of infertility (attempting to conceive for ≥12 months without becoming pregnant), and history of sexually transmitted infections (STIs), including genital warts, herpes, chlamydia or pelvic inflammatory disease.

Data Analysis

We used life-table methods to compute the proportion of participants who conceived during follow-up, accounting for censoring. We used proportional probabilities regression to estimate fecundability ratios (FR), the per-cycle probability of conception comparing exposed with unexposed individuals, and 95% confidence intervals (CI), adjusting for potential confounders. FR below 1.00 indicate reduced probability of conception (Weinberg et al., 1989). We used the Anderson–Gill data structure to account for left truncation (Cox, 1972), and we accounted for the decline in baseline fecundability with increasing attempt time by including indicators for cycle at risk in regression models. Results were adjusted for age (<25, 25–29, 30–34, ≥35 years), education (≤12, 13–15, 16, ≥17 years), household income (<50 000, 50 000–99 000, 100 000–149 000, ≥$150 000 US dollars), BMI (<25, 25–30, 31–34, ≥35 kg/m2), preconception multivitamin use (yes, no), PSS score (continuous), MDI score (continuous), any efforts to improve the chances of conception, (e.g. checking basal body temperature, charting menstrual cycles) (yes, no), intercourse frequency (<1, 1, 2–3, ≥4 times per week), current smoking (yes, no), and a history of STI (yes, no).

Antibiotic use was analyzed at baseline and as a time-varying exposure. As the potential effect of a single instance of antibiotic use on fecundability is likely to be transient, analyses with antibiotic use as a time-varying exposure are likely to be the most biologically relevant and were therefore considered to be the main analyses of interest. Analyses of antibiotic class and indication were similarly analyzed as time-varying exposures. To further isolate potential transient effects of antibiotic use on fecundability, we examined the effect of antibiotic use overall on fecundability, restricting to the first cycle of follow-up after baseline as a sensitivity analysis.

We conducted several stratified analyses to assess effect measure modification of the relation between antibiotic use and fecundability. We stratified by age (<30 years of age versus ≥30 years of age), to determine if any association of antibiotics with time to pregnancy is compounded by increasing age, a strong determinant of fertility (Wesselink et al., 2017). We also stratified by pregnancy attempt time at study entry (0–2 versus 3–6 cycles at study entry), to assess reverse causation, wherein participants who have been trying to conceive for longer are more likely to have altered their behavior in response to their difficulty conceiving (Wise et al., 2020). A stronger effect in the 0–2 cycles group may be more likely to be due to an effect of antibiotics rather than other behavioral changes undergone while trying to conceive. We also stratified by current smoking status, as smoking leads to increased oxidative stress and may compound the effect of antibiotic use or underlying infection on fecundability (Kamceva et al., 2016; Wesselink et al., 2019). We considered BMI (<30 vs ≥30 kg/m2) as a potential effect measure modifier due to the association between obesity and the vaginal microbiome (Brookheart et al., 2019). Finally, results were stratified by a composite variable representing history of reproductive health conditions (no history of STI, endometriosis, bacterial vaginosis, infertility, or spontaneous abortion versus a history of any of these conditions) to examine the effect of antibiotics among women for whom they may improve fecundability by treating an underlying pelvic infection.

We used multiple imputation to generate values for missing baseline and follow-up data on covariates and pregnancy status. Covariate information was missing for <1% of participants for all covariates except for income, which was missing for 3% of participants. Missingness for antibiotic use overall was also <1%. We assigned one cycle of observation to the 17% of women who did not complete any follow-up questionnaires and imputed the outcome of that cycle (pregnant vs not). We created five imputed datasets and statistically combined coefficient and standard error estimates from the five datasets.

In analyses of antibiotic class and indication for use, we excluded 2413 (25%) of the 9524 participants owing to their completion of the questionnaire prior to March 2016. Antibiotic class was unknown or could not be determined for 15% of included participants. Antibiotic class for these participants was imputed to the most common class reported for the indication they provided. Antibiotic indications were unknown or could not be classified for 24 participants (2%). These participants were excluded from antibiotic indication analyses and imputed to penicillins (the most commonly used antibiotic) for antibiotic class analyses.