Detecting variations in ovulation and menstruation during the COVID-19 pandemic, using real-world mobile app data.

On March 11, 2020 the World Health Organization (WHO) declared the COVID-19 outbreak a global pandemic, calling upon all countries to take urgent and aggressive action to prevent further spread of the disease. The pandemic declaration led to increasingly strict and widespread stay-at-home orders, which contributed to population level concerns about the risk of not only infection and loss of life, but the potential irrecoverable loss of livelihood, as non-essential businesses were shut down and citizens deprived of income, while facing continued expenses. Persisting beyond a full year, the COVID-19 pandemic produced a global population-level stressor that continues to influence how people live. Historically, population level stressors have been linked to menstrual changes. Examples include World War II from 1939–1945 [ 1 ], the Dutch Famine from 1944–1955 [ 2 ], and the Desert Storm War in 1996 [ 3 ]. These examples represent a range of stressors, whether they be physical vs. psychological, direct vs. indirect, or short vs. long-term. At the extremes of stress, a case series of women studied prior to being executed noted that almost all became amenorrheic [ 4 ]. In an observational study of Lebanese women exposed to wartime bombing from April 11–27, 1996, 35% experienced menstrual aberrations for 3 months, as compared to an unexposed group that reported abnormalities in 2.6% [ 5 ]. Abnormal menstrual patterns are also commonly reported among desert-dwelling hunter–gatherer women who face difficult living conditions [ 6 ]. Yet mortal stresses of this magnitude are not required to induce menstrual aberrations; chronic job-related stress can be a contributor [ 7 , 8 ]. Further, in a study of Japanese college students, 15.8% reported a correlation between their school examinations and irregular menses [ 9 ]. The range of stressors that can influence one’s likelihood of menstrual irregularity is wide, suggesting that other mediating factors may play a role. The impact of stress on the reproductive system is grounded in biology whereby stress-related glucocorticoids can inhibit the release of gonadotropin releasing hormone (GnRH), luteinizing hormone (LH), and estradiol (E2) from the hypothalamic-pituitary-gonadal axis, which is most easily observed as secondary amenorrhea. More subtle manifestations would include delays in ovulation, anovulation, and changes in both cycle and menstruation lengths. The COVID-19 pandemic represents a unique stressor, independent of whether individuals were infected by the virus, because of its far-reaching psychologic, social, and economic consequences beyond physical alone. Consequently, reproductive age females may have experienced changes in their ovulatory and menstrual cycles during this time. No studies have yet examined the influence of stress linked to an ongoing global pandemic on ovulatory and menstrual changes among a reproductive age female population. The pandemic provides an opportunity to identify and characterize potential changes. Natural Cycles is the first Food and Drug Administration (FDA) cleared and CE-marked mobile app that uses women’s records of menstruation and basal body temperatures to identify their fertile window and estimate appropriate times when they can have unprotected sexual intercourse and be reasonably certain that they will avoid pregnancy. As of the beginning of the COVID-19 pandemic in 2020, Natural Cycles had more than 1.5 million members using the app for contraception, making its database one of the largest collections of menstrual cycle data ever compiled. We conducted analyses using aggregated, real-world menstrual cycle data from app users both before and during the COVID-19 pandemic with the primary objective of detecting potential changes in ovulation, cycle length, and menstrual duration during the pandemic. The secondary objective of this study was to examine the association of any menstrual changes with self-reported perceived stress related to the pandemic.

Our sampling frame included a total of 214,426 cycles of data from 18,076 individual users from over 60 countries worldwide ( Table 1 ). Users were primarily from Great Britain (29.3%), the United States (22.6%), and Sweden (17.8%). The average user was 32.5 ± 5.8 years of age at the time of analysis, and most held at least a university degree (79.9%). Most users reported being in a relationship, engaged, or married (85.1%); only 25.6% of users reported ever being pregnant and 16.9% of users reported having at least one child. *p<0.05 **p<0.01 ***p<0.001. † Sum different from total number of users due to missing data. The proportion of those reporting being very to extremely stressed rose from 46.2% pre-COVID to 61.1% during COVID. Nearly half of users (45.4%) reported more pandemic-related stress, with 33.2% reporting no change, and 21.4% reporting less stress compared to the pre-COVID period. Individuals between the ages of 25–34 years, from the United States, with a graduate level degree, who reported being engaged or married, and having a child were significantly more likely to report more stress during the pandemic than prior. Users individually contributed approximately 6 cycles to each of the pre-COVID and during COVID sampling frames. With respect to cycle characteristics ( Table 2 ), the average cycle length among users significantly decreased from 29.40 (95%CI 29.34–29.46) days pre-COVID to 29.16 (95%CI 29.10–29.22) days during COVID (p<0.001). The average menstrual duration significantly increased from 4.21 (95%CI 4.19–4.23) days to 4.32 (95%CI 4.30–4.34) days. The average incidence of anovulation and abnormal cycle length decreased significantly across cycles, from 2.9% (2.7%-3.0%) to 2.5% (2.3%-2.6%) and 8.7% (8.4%-8.9%) to 8.0% (7.8%-8.2%), respectively, while the average incidence of prolonged menstruations increased from 0.9% (0.8%-1.0%) to 1.0% (9.0%-1.1%). Abnormal cycle length: 38 days; Prolonged menses: >8 days. At the user level, the proportion of users experiencing any anovulatory cycles and any abnormal cycle lengths decreased from 11.1% to 9.1% (p<0.001) and 28.4% to 26.6% (p<0.001), respectively. The proportion experiencing increased menstrual duration increased from 3.5% to 3.9% (p = 0.04). We did not detect any statistically significant associations between changes in user ratings of stress before and during the pandemic and abnormal cycle parameters ( Fig 1 , Table 3 ). To explore factors associated with changes in the proportion of anovulatory cycles and abnormal cycle lengths after the pandemic, we conducted a multinomial logistic regression, setting no changes in the proportion of abnormalities as the reference point. We included those potential covariates significantly associated with changes in abnormal cycle parameters at the p<0.05 level. While changes in stress were not linked to changes in cycle parameters, we included this covariate given our interest in the potential role of subjective stress on menstrual changes ( Table 4 ). Via this regression, only age and relationship status remained independently associated with any of the cycle outcomes. Users above the age of 45 years were more likely to report more anovulatory cycles, abnormal cycle lengths, and prolonged menses. Individuals who were in a relationship or were engaged or married were less likely than their complicated and single counterparts to experience more anovulatory cycles. * p<0.05.

The COVD-19 pandemic did not induce population-level changes to ovulation and menstruation among women using a mobile app to track menstrual cycles and predict ovulation. While some women experienced abnormalities during the pandemic, this proportion was smaller than that observed prior to the pandemic. As most app users in this study were well-educated women over the age of 30 years, and from high-income countries, their experience of the COVID-19 pandemic may differ in ways that limit the generalizability of these findings.

Our data was comprised of real-world, sociodemographic, limited clinical, and menstrual cycle data submitted by users of the Natural Cycles pregnancy prevention and fertility tracking mobile application [ 10 ]. All users who contributed data agreed to make their data available for clinical investigation prior to starting application use. All data was de-identified by the Natural Cycles data management team (EBN) and stored within a closed database, which was transferred to the research team at the University of Southern California (USC; BTN, RDP, HR, ALN) for subsequent data analysis. This research plan was reviewed and approved by the USC Institutional Review Board, which classified the study as exempt, non-human subjects research (HS-20-00402). We included users who registered an account for contraception prior to September of 2019 and who consented to release their de-identified data for research purposes. We excluded cycle data from users who were breastfeeding, who reported a pregnancy in the 12 months prior to registration within the data collection period, and who became pregnant during use of the app. We additionally excluded any user on a hormonal contraceptive method, as well as reporting at the time of registration, a diagnosed medical condition that could influence cycle regularity, such as polycystic ovarian syndrome, endometriosis, thyroid disorders, and perimenopausal symptoms. To account for any undiagnosed secondary amenorrhea or ovulatory disorders, as well as to improve our ability to detect pandemic-induced changes in ovulation and menstruation, we excluded users who recorded any cycle length lasting more than 90 days during the pre-COVID analysis frame. We further excluded users who did not contribute at least two cycles of data prior to the COVID-19 pandemic and at least two cycles during the COVID-19 data collection period. With respect to cycle data, we included only complete cycles (i.e., cycles with a start and an end) collected from March through September 2019 (pre-COVID) and March through September 2020 (during COVID). All cycles at the beginning of the sampling frame started in March; all cycles included from the end of the sampling frame must have started in September even if the cycles did not end until later. We excluded cycle data starting from October 2019 through February 2020 due to ambiguity at the time about the scale and threat of infection among various populations of users. We did not include cycles in which the Natural Cycles algorithm did not have enough data to definitively determine whether an ovulation occurred within the cycle (i.e., not enough user data entered). To improve the precision of contributed cycle data, we further excluded cycles that were not validated by at least 10 basal body temperature entries. We were primarily interested in examining the proportion of users experiencing anovulation, abnormal cycle length, and prolonged menstruation among app users pre-COVID compared to during COVID. Anovulatory cycles were defined according to the Natural Cycles app’s proprietary algorithm. The algorithm identifies ovulation retrospectively based on the first day of menstruation and basal body temperatures, which may be supplemented by positive urinary LH tests. Basal body temperatures are recorded each morning using a thermometer sensitive to the hundredth place, and with measures excluded if the user reports any illness, alcohol intake, or changes in sleep that might influence basal temperatures. Users of Natural Cycles record basal temperatures for approximately 70% of the days; approximately 25% use LH tests. To reduce the risk of misidentifying ovulations, the algorithm reports ovulation by rising basal body temperature only if the average temperature from three consecutive calendar days is greater than the woman’s follicular phase average and her baseline average across all data entries, as well as consistent with her luteal phase average [ 11 ]. If no temperature rise is observed and the data quality and quantity is deemed sufficient, the cycle is flagged as anovulatory. Of note, users with stable measurements (e.g., small day-to-day variations in the same cycle phase), require fewer data points for the Natural Cycles algorithm to draw conclusions about changes in the basal body temperature. Cycles with low data quality (e.g., high day-to-day temperature variability) or data that is insufficient to detect or exclude an increase in the basal body temperature are excluded by the analysis. While missing data may affect the app’s ability to predict the exact day of ovulation, they are not expected to affect its determination of whether ovulation has occurred. Based on criteria from the International Federation of Gynecology and Obstetrics (FIGO), we defined abnormal cycle length as lasting less than 24 days or more than 38 days, and prolonged menstruation length as lasting greater than 8 days [ 12 , 13 ]. Natural Cycles implemented a query on May 11, 2020 of its users’ experience of pandemic-related stress. Users could provide a response to this item through June 11, 2020. We incorporated this measure to examine a stress-related pathway for menstrual abnormalities. The researchers asked users to rate two Likert-type items: (1) “Thinking about your stress level before the COVID-19 pandemic started, how stressed were you then?” (2) “How stressful is the COVID-19 pandemic to you now?” The Likert items were anchored from “Not at all stressed” (1) to “Extremely stressed” (5). We then calculated a COVID-related stress change score by subtracting the user’s report of stress before COVID from their stress rating during COVID. For ease of interpretation, we categorized perceived changes in stress using five categories: “much less (1),” “less (2), “unchanged (3),” “more (4),” and “much more (5).” In addition, we used sociodemographic data and reproductive histories provided by users at the time of app registration and updated during use of the app as covariates in our analysis. These covariates included: age, country of registration, education (from less than a high school degree to graduate degree), relationship status (e.g., in a relationship, engaged or married, it’s complicated, and single), history of pregnancy, and whether they have any children. At the population level, we examined differences in average cycle lengths and menstruation lengths before and during the COVID-19 pandemic via paired t-tests. We calculated the proportions of users experiencing an anovulatory cycle before and during the pandemic and analyzed for differences via Chi-square tests of association. Given that individual users did not contribute the same number of cycles to the study, we compared differences in the proportions of abnormal menstrual parameters recorded by calculating for each user their total number of cycles with anovulation, abnormal cycle length, and prolonged menses for each sampling frame and divided them by the number of cycles contributed. We examined for statistically significant changes in these abnormal cycle parameters, pre-COVID and during COVID, via paired t-test. Given our interest in COVID-related stress as a potential mediator of abnormal cycles, we conducted a subset analysis of data from individuals contributing stress ratings during the pandemic and examined its relationship categorically via Chi-square tests and continuously via paired t-tests. We examined the role of the user’s age, country of registration, education, relationship status, and history of pregnancy and/or children on any increase in mean anovulation, cycle length, and menstruation length among users via Chi-square tests. We included covariates associated with increased anovulatory cycles, abnormal cycle lengths, and abnormal menstruation length (at an alpha level of 0.05) in a separate multivariable logistic regression to determine each covariate’s adjusted odds of influencing the outcome of interest. All analyses were conducted in Stata (Version SE/14.2; College Station, TX).