Methods
We conducted a retrospective cohort analysis of prospectively collected menstrual cycle data. Menstrual cycle data ranged from 1 October 2020 to 7 November 2021; initial covid-19 vaccine doses were received between 2 January and 31 October 2021. Individuals who use the digital fertility awareness application Natural Cycles voluntarily choose to prospectively track physiological data related to their menstrual cycles for purposes of pregnancy prevention or planning without the use of hormonal methods and consent to the use of their de-identified data for research (consent can be removed, if desired). A detailed description of variables tracked by the app has been published elsewhere.16 To be eligible for study inclusion, individuals had to consent to use of their deidentified data for research purposes, report their covid-19 vaccination status, and record at least one cycle after 1 October 2020. Users were informed of the purpose of the research. We included individuals aged 18-45 years who were at least three cycles after pregnancy or after use of hormonal contraception and not menopausal, with normal prevaccination menstrual cycle lengths (average of 24-38 days),9 and known geographical location. Each individual contributed a minimum of four consecutive cycles of data. For those who received a covid-19 vaccination, we included three prevaccine cycles, at least the first covid-19 vaccine dose cycle and subsequent consecutive cycles recorded in the application through the cycle following the second vaccine dose. For individuals who were not vaccinated, we included four to six consecutive cycles from a similar time period, depending on the number recorded. We excluded individuals with no cycle data for time points after the first covid-19 vaccine dose cycle from analyses for later time points.
Our primary exposure was covid-19 vaccination status, as reported by individuals within the Natural Cycles application. Prompted by in-app messages, individuals recorded their vaccination date or dates or confirmed their unvaccinated status. For a sensitivity analysis focused on vaccine type, we categorised vaccine brands by mechanism of action: mRNA (Pfizer-BioNTech (BNT162b2) and Moderna (mRNA-1273)), adenovirus vector (Oxford-AstraZeneca (ChAdOx1 nCoV-19), Covishield (ChAdOx1-S), Johnson & Johnson (Ad26.COV2.S), and Sputnik V (Gam-COVID-Vac)), and inactivated virus (Covaxin (BBV152), Sinopharm (BBIBP-CorV), and Sinovac (CoronaVac)).
Our primary outcome was the mean change within individuals in cycle length (in days), from the three cycle prevaccination average to the first vaccine dose cycle, comparing vaccinated and unvaccinated groups. For vaccinated individuals, cycle four was the first vaccine dose cycle; the cycle of the second dose varied based on when this second vaccine dose occurred (cycles four through 13), if applicable. We designated the cycle after the second vaccine dose cycle as the postvaccine cycle in order to determine whether any menstrual cycle changes attenuate or disappear after vaccination. For the unvaccinated cohort, we designated cycle four as the notional first vaccine dose cycle, cycle five (the cycle when the largest proportion of vaccinated individuals received their second dose) as the notional second vaccine dose cycle, and cycle six as the postvaccine cycle; cycles one, two, and three were considered the equivalent of prevaccination cycles.
Secondary outcomes were the same mean change within individuals in cycle length for the second vaccine dose cycle as well as the postvaccine cycle, and corresponding changes in menses length for both doses. For individuals who were vaccinated during their menses, we used the menses length of the vaccine cycle, and for those who were vaccinated after completing their menses, we used the menses length from the subsequent cycle. We also examined the proportion of individuals who had a clinically significant change in cycle length (≥eight days).9
We included additional sociodemographic information collected within the Natural Cycles application via an in-application message; response was voluntary and some questions were only sent to a subset of users, resulting in a large amount of missing data (see online supplemental table 1 for distributions of missing data). Missingness was non-ignorable and was included as a category in our analyses. We categorised age at the start of the first cycle as 18-24, 25-29, 30-34, 35-39, or 40-45 years. Race and ethnic group were reported as Asian, Black, Hispanic, Middle Eastern or North African, Native Hawaiian or Pacific Islander, or white, which we collapsed into a binary variable for modelling. We classified geographical region as UK or Channel Islands, Europe, US or Canada, Australia or New Zealand, and other. Most individuals in the European region were from Sweden (56%) and in the other category, most were from Brazil (62%). We categorised body mass index as underweight (<18.5), normal weight (18.5-24.9), overweight (25.0-29.9), and obese (30.0 or above), combining underweight and normal weight for modelling due to the small sample size of underweight individuals. Additional characteristics included parity (nulliparous v parous), education (at least an undergraduate degree or not), and relationship status (in a relationship or not).
The Oregon Health and Science University Institutional Review Board approved the protocol (No 00023204), as did Natural Cycles’ research oversight committee, and the Reading Independent Ethics Committee, UK (No 230721). All participants consented to the use of their de-identified data for research, which were used under a data use agreement with Natural Cycles USA.
Statistical analysis
We described sociodemographic characteristics of our sample by vaccination status. We compared all mean changes within individuals in cycle and menses length, by vaccination status, using two sided t tests. We created histograms overlaying vaccination status to compare the distributions of changes in cycle and menses length, and compared the proportion of individuals who had a clinically significant change in cycle length (≥eight days) using Pearson’s χ2 tests. We compared sociodemographic and prevaccination menstrual characteristics of individuals who had a change of eight days or more change to those who did not using two sided t tests and Pearson’s χ2 tests, by vaccination group.
We developed longitudinal multivariable mixed effects models for all cycle and menses length outcomes, and plotted the adjusted lengths (in days) before and after vaccination. Models contained random intercepts and slopes at the individual level, and an interaction term between time (before and after vaccination) and vaccination status to determine the effect of vaccination. The effect was defined as the adjusted difference in the change in cycle and menses length between vaccination groups. All estimates were adjusted for age, body mass index, parity, race or ethnic group, education, relationship status, and global region.
Based on previous findings in the US cohort,15 we performed a subanalysis focused on the number of doses received in a single cycle (one v two). About 5% of the vaccinated sample received their second dose in the same cycle as the first. We stratified vaccinated individuals by the number of doses received in the first dose vaccine cycle and compared all outcomes for each group to the unvaccinated cohort.
We conducted multiple sensitivity analyses to confirm the robustness of our results. Firstly, although the data did not meet the missing at random assumption required for imputation techniques, we performed 500 iterations of imputation followed by weighting with covariate balancing propensity scores and bootstrapped standard errors. We did this analysis to confirm that our results were not biased by missing data or covariate imbalance between vaccination groups.17 We compared the changes in cycle length among vaccinated individuals by the vaccine’s mechanism of action (mRNA, adenovirus vector, or inactivated virus), adjusting for age group to account for age dependent differences in vaccine rollouts. Any individual who reported polycystic ovarian syndrome, thyroid disorder, or endometriosis was excluded (804 individuals). We excluded people who reported use of emergency contraception during at least one study cycle (715 individuals). Additionally, we excluded individuals with any cycle before vaccination whose absolute cycle length was outside of the 24-38 day range (3006 individuals). We also analysed changes during the first vaccine dose cycle, excluding individuals with no data for the second vaccine dose cycle (n=5599) and for the after vaccination cycle (n=6617). Finally, we stratified by global region to examine any potential regional differences.
We had more than 99% power to detect an unadjusted one day difference in cycle length change or 0.5 day difference in menses length change by vaccination status, at a Bonferroni-corrected significance level of 0.007 (99.3% confidence intervals). We accounted for multiple comparisons among the seven primary and secondary outcomes: cycle and menses length for the first and second vaccine dose cycles, cycle length for the cycle after the vaccine, and the proportion of individuals who had a clinically significant change in cycle length (eight days or more) for the first and second vaccine dose cycles. We adjusted all P values to reflect this reduced significance level of 0.007 (see online supplemental file 2 for prespecified analysis plan).
Patient and public involvement
No members of the public were directly involved in this study, although the research question was developed in response to public reports of menstrual changes after covid-19 vaccination. Natural Cycles informs users of study results through monthly newsletters, via their research library within the application, and through social media channels. The research team uses academic and public dissemination channels to inform the public of the results.