Abnormal uterine bleeding patterns determined through menstrual tracking among participants in the Apple Women's Health Study

During the study period, there were 47,751 eligible enrollees. We excluded those aged >50 years (n=3804), and those reporting menopause (n=4439), pregnancy (n=3209), lactation (n=2260), or hormonal therapy (n= 15,989). Excluded participants could fall into multiple categories. There were 18,875 participant records included in this analysis. The demographic characteristics of included participants were similar to those of participants excluded because of reporting pregnancy, lactation, or hormonal therapy ( Supplemental Table 1 ). Table 1 presents demographics for the total cohort, those with and without confirmed tracking, and those with no AUB and with AUB (tracking accuracy not confirmed and confirmed). The mean age at study entry was 33 years, and the mean BMI was 29.3 (overweight); 68.9% of participants identified as White, non-Hispanic, 8.1% as Hispanic, 5.9% as Black, 4.0% as Asian, and 10.6% reported ≥2 race/ethnicities. The average number of cycles per participant was 16±9 cycles, and the average number of days of data per user was 260±62 days. Because participants may track bleeding with varying levels of fidelity, and incomplete tracking could introduce error into AUB detection, we excluded participants with AUB who did not confirm tracking accuracy (n=10,619). Before exclusion, we evaluated if those who confirmed tracking differed from those who did not ( Table 1 ). These groups were similar across demographics ( Table 1 ; Supplemental Tables 2 and 3 ). Although demographic characteristics of participants did not differ markedly by tracking confirmation status, rates of AUB differed: 43.2% of participants had AUB when not restricting to only those with tracking confirmation, whereas 16.4% had AUB after tracking accuracy was confirmed ( Table 2 ). Of those with confirmed tracking, 2.9% had irregular menses, 8.4% had infrequent menses, 2.3% had prolonged menses, and 6.1% had spotting. The remaining analysis includes data from participants who confirmed tracking accuracy. Table 3 presents the prevalence of AUB by race/ethnicity after controlling for age and BMI. Black participants had 33% (prevalence ratio [PR], 1.33; 95% CI; 1.09–1.61) higher prevalence of infrequent menses compared with White, non-Hispanic participants, whereas Asian participants had a higher prevalence of irregular menses (PR, 1.41; 95% CI, 0.92–2.14). Prevalence of prolonged menses and spotting did not vary by race/ethnicity. Participants with class 1 and class 3 obesity had 14% (PR, 1.14; 95% CI, 1.04–1.25) and 18% (PR, 1.18; 95% CI, 1.02–1.38) higher prevalence of AUB, respectively, after controlling for age and race/ethnicity ( Table 4 ). No differences in the prevalence of irregular menses or spotting were observed across weight categories. The prevalence of infrequent menses was higher in those with class 1, 2, and 3 obesity than in those with healthy weight. Those with class 3 obesity had 94% higher prevalence of prolonged menses (PR, 1.94; 95% CI, 1.29–2.90). Controlling for age and race/ethnicity, the association of self-reported medical conditions with AUB was evaluated. Participants with PCOS had 19% higher prevalence of AUB (PR, 1.19; 95% CI, 1.08–1.31) compared with participants without PCOS ( Table 5 ). Those self-reporting endometriosis had a 28% increased prevalence of AUB (PR, 1.28; 95% CI, 1.12–1.45). Participants with hypothyroidism and cervical dysplasia had higher prevalence of AUB ( Table 5 ; Supplemental Tables 4 and 5 ). Evaluation based on the individual abnormal bleeding patterns demonstrated an increased prevalence of irregular menses (PR, 1.69; 95% CI, 1.04–2.73) and spotting (PR, 1.54; 95% CI, 1.1–2.11) in those self-reporting hyperthyroidism ( Supplemental Tables 4 , 6 and 7 ). Endometriosis was more prevalent in participants reporting infrequent menses (PR, 1.58; 95% CI, 1.29–1.95) ( Supplemental Table 5 ). There was an increased prevalence of infrequent menses in those reporting PCOS (PR, 44; 95% CI, 1.23–1.68) ( Supplemental Table 5 ). We presented the prevalence of AUB in a large, digital cohort. AUB was reported by 16.4% of participants; 2.9% reported irregular menses, 8.4% reported infrequent menses, 2.3% reported prolonged menses, and 6.1% reported spotting. Use of tracking confirmation may provide a usable tool for limiting data for research analysis. AUB was more commonly identified in Black participants and those with elevated BMI, particularly class 3 obesity. Several conditions known to be associated with AUB, such as PCOS, thyroid disorders, fibroids, endometriosis, and cervical dysplasia, had expected associations with AUB in this study. Identification of AUB prevalence varies according to definitions, types of AUB included, populations evaluated, and study methodology, with ranges from 10% to 30%. 29 , 30 Our estimate of AUB (16.4%) falls in this range. We found 2.9% of participants reporting irregular menses. In a study analyzing 6375 person-years, irregular menses, defined as 3 to 5 periods in 90 days with 14 days, 20% of a cohort of 130 participants prospectively keeping menstrual diaries met the definition. 31 Irregular cycles (cycles 38 days) have been reported, ranging from 8.0% to 8.7% in a digital cohort of 18,076 users, evaluated in relationship to COVID-19 pandemic–related stress. 32 Reports relying on self-categorization demonstrate higher rates of perceived irregular menses. A Danish report, defining irregular as cycle variation >14 days, demonstrated 20.3% self-reporting irregularity in the previous year. 33 We found 8.4% reporting infrequent menses. The Tremin study reported infrequent menses rates ranging from 20% in adolescents to 5% in those in their late 30s, rising to 17% in participants aged >45 years. 8 When defined as >35 days without menses, a meta-analysis including 35 studies reported the mean proportion of those with infrequent bleeding as 13% across the reproductive lifespan. 34 Prolonged menses was reported by 2.3% of our cohort. When prolonged bleeding was defined as ≥14 days, it was almost never identified in the Tremin study. 8 With a definition of ≥10 days, reports range from 9% to 5.3%. 35 , 36 The Study of Women’s Health Across the Nation (SWAN) reported on >8 days of menses (10.2%) and >10 days of menses (6.6%) in a cohort of 1320 individuals aged 42 to 52 years. 37 The added burden of repeatedly tracking each bleeding day, as compared with surveys based on recall, may account for differences, and may have led to underreporting of prolonged menses in our cohort. Spotting was reported in 6.1% of our participants. This was lower than the 17% reported in an English cohort, which relied on surveys for retrospective reporting, but similar to results from a population-based, cross-sectional study reporting 4.3% of participants with spotting. 38 , 39 Previous work examined the relationship between BMI and AUB, with our results supporting this association. In the Nurses’ Health Study II, those who self-reported irregular or infrequent menses had a higher BMI than those with very regular menstruation. 6 In a digital cohort limited to cycles with evidence of ovulation, those with a BMI >35 had more cycle length variation and a longer follicular phase compared with those in the BMI range of 18 to 25, although the exclusion of anovulatory cycles may limit the variation described. 40 In our study, Black participants were more likely to report AUB. In the SWAN study, African-American participants were less likely to have prolonged menses, but more likely to have heavy menses compared with White participants. 37 In an evaluation of menorrhagia incidence in the United States Armed Forces, rates were highest in those of Black, non-Hispanic ethnicity. 41 In an online survey for participants with uterine fibroids, African-Americans were more likely to have heavy or prolonged menses. 42 Our findings support previous work demonstrating differing rates of AUB by reported race and ethnicity. Several existing relationships between AUB and medical conditions were reinforced with this study. For example, many patients with PCOS present with abnormal menses. 43 Here, those with PCOS had a 19% higher prevalence of AUB. In a study of 27,840 participants with endometriosis, 50.8% reported heavy/irregular periods, whereas we found a 28% increased prevalence of AUB in our cohort. 44 We found that AUB was more common in those with hyperthyroidism (34% higher prevalence) and hypothyroidism (17% higher prevalence), which supports previous work finding that 58% of patients with hyperthyroidism had oligomenorrhea or amenorrhea. 45 In a study of 79 patients with hypothyroidism, 21.5% had infrequent menses, as opposed to 6.7% of those who were euthyroid. 46 Several limitations should be noted, including reliance on self-report of conditions, the observation that our analysis does not account for those with >1 condition (thus the prevalence may be affected by comorbidities), and exclusion of participants using hormonal contraceptives to manage their AUB. Increasingly, digital tools are becoming the preferred method of tracking menstrual information, and self-collection of data may be done with sufficient granularity to detect changes or cycle events. 9 , 10 The AUB pattern identification presented herein is embedded in the Cycle Tracking feature available in the Health app for consumer use. Tracked data may limit artifacts related to retrospective recall. 47 Identification of AUB may provide opportunities for risk screening and prevention. AUB labeled in medical records or self-reports was a marker for higher risk of developing ovarian cancer, cardiovascular disease, or diabetes mellitus. 4 , 48 , 49 With these medical conditions linked to AUB, leveraging menstrual tracking to identify AUB may provide value. Digital tracking of menstrual data using smartphone apps provides the potential to extract meaningful research information, but there are challenges. 15 One consideration is how to determine the quality or completeness of tracking. Our reported rates of AUB would have been overreported without use of tracking accuracy confirmation. Other studies have used engagement dynamics to exclude cycles, leveraging tracked symptoms to distinguish true long cycles from missed tracking by requiring at least 5 interactions with the app in the month containing menstruation and the following month. 14 , 24 Another method is exclusion of cycles not confirmed as ovulatory, which allows for analysis of follicular and luteal phase characteristics, but may preclude identification of abnormal patterns. 40 Future investigations are needed to understand the temporal relationships between AUB and the identification of health conditions. In addition, application of definitions of AUB across populations may miss changes in an individual’s bleeding pattern. Further investigation into changes from one’s baseline and their relationship to disease development may benefit individual patient care and guide screening. Similarly, contextualization of bleeding patterns with logged symptoms may help identify which bleeding patterns and symptom combinations are most associated with disease. Lastly, future exploration of physiological data acquired from sensor-derived metrics, such as sleep, steps, or activity, may better identify AUB patterns of concern. Further validation of the data is provided by the AWHS measures of AUB from tracked data demonstrating expected associations with conditions. As the study continues and data accumulate, we will be able to both investigate additional risk factors for AUB and learn more about its longer-term health consequences. This was a prospective, digital study using tracked bleeding as opposed to reliance on recall or survey tools. The results expand our understanding of rates of AUB in a diverse and modern cohort. In addition, participants were not restricted on the basis of reproductive goals, medical histories, or particular clinical sites, providing an expanded evaluation of 4 types of AUB (irregular menses, infrequent menses, prolonged menses, and spotting). This study builds on previous cohort data by adding diversity, and extends modern datasets derived from menstrual tracking apps by collecting contextual information related to demographics and medical histories. 32 , 50 In addition, we addressed concerns about incomplete menstrual tracking related to poor engagement with tracking apps with the requirement that participants confirm their tracking monthly. This study has several limitations. When drawing comparisons with other studies, the generalizability should be considered in relationship to the use of iPhone as the research platform, and participant race/ethnicity (68.9% White, non-Hispanic; 8.1% Hispanic; 5.9% Black; 4% Asian in AWHS vs 60.1% White, non-Hispanic; 18.5% Hispanic; 13.4% Black; 5.9% Asian in the United States) and reported levels of education (53.3% in AWHS with at least a college degree vs 32.5% in the United States), which varied from the US census. 51 We also may not capture all those experiencing abnormalities with our definitions of AUB. In addition, we do not report on all possible abnormal bleeding patterns. Subjective information on heavy menstrual bleeding or menstrual flow level is not currently captured in our study, which will affect comparisons with studies that include this in composite rates of AUB. 13 Our exclusion of participants who reported pregnancy, lactation, or hormone use during the study may also lead to misestimation of overall AUB rates, although the frequency of their demographic characteristics and reported medical conditions were similar. Medical conditions were only self-reported by a subset of participants, and specific management of those conditions that may alter their current bleeding is absent from our data. The overall small number of participants with any individual condition precluded controlling for effects of such conditions in our analysis of age, BMI, and race/ethnicity; thus, there may be residual confounding by medical condition. Similarly, if a participant has AUB but has not yet been evaluated or diagnosed with a condition, or has a condition not accounted for in our study, this would limit our ability to describe those associations.

The AWHS is a digital, longitudinal study of menstrual health conducted using the Apple Research app. 16 Enrollment began on November 2019 and is ongoing. Inclusion criteria include having ever menstruated, having a compatible iPhone and version of iOS, living in the United States, age of at least 18 years (at least 19 years in Alabama and Nebraska, at least 21 years in Puerto Rico), ability to communicate in written and spoken English, being an exclusive user of an iPhone and iCloud account, and consenting to participation. For this analysis, participants with at least 180 days of menstrual tracking data during the study time frame from November 2019 through July 2021 were included. Participants aged >50 years or reporting menopause, pregnancy, lactation, or hormonal contraception use immediately preceding study enrollment or at any point within the study time frame were excluded. The study was approved by the Institutional Review Board at Advarra (CIRB PRO00037562) and registered on ClinicalTrials.gov ( ClinicalTrials.gov Identifier: NCT04196595 ). At enrollment, participants provide year of birth, race/ethnicity, gender identity, sociodemographic information, weight, and height. 17 Report of race/ethnicity used a single “select all that apply” question to determine race and ethnicity from the National Institutes of Health–sponsored All of Us study. 18 Survey content related to medical and reproductive histories has been published previously. 19 Participants self-reported medical conditions (eg, fibroids, cervical dysplasia) using survey responses. HealthKit provides a central repository for health and fitness data on iPhones and Apple Watches, and is viewable to participants through the Apple Health app. AWHS participants contribute HealthKit data after granting permission to the Research app. HealthKit stores data merged from multiple sources and contains data such as tracked menstrual bleeding, heart rate, exercise, and sleep. Bleeding is manually tracked in Cycle Tracking in the Health app or in any third-party menstrual tracking application that the participant permits to write to HealthKit. Menses, identical to those in Cycle Tracking in the Health app, were defined as follows: the minimum number of flow days for a menses was 1, with no maximum. 20 Menstrual bleeding separated by 1 day of missed tracking or no flow was merged into 1 menses, and ≥2 days without tracking was not merged. 21 The first day of flow was labeled as the first day of a cycle. 22 Spotting was not included as menstrual flow. The minimum number of days in a cycle was set to 10, with no maximum. Four types of AUB patterns were defined: irregular, infrequent, and prolonged menses and intermenstrual bleeding. Definitions were derived from recommendations by the International Federation of Gynecology and Obstetrics and are identical to those embedded in Cycle Tracking in the Health app 12 , 13 ( Figure ). Irregular menses was defined as varying lengths of cycles of ≥17 days within each of 2 consecutive 90-day analysis windows. Infrequent menses was defined as ≤1 menses in each of 2 consecutive 90-day analysis windows. Prolonged menses was defined as ≥2 menses lasting ≥10 days in a 180-day window. Intermenstrual bleeding (spotting) was defined as spotting tracked between menses at least once in each of the 2 consecutive 90-day windows. 8 Spotting adjacent to menstrual bleeding was excluded to account for the spotting datatype being used to mean light menstrual flow at the start or end of menses. Using the Monthly Survey, Menstrual Update, participants confirm the accuracy of the previous month’s tracked data by responding to “Are all your period days during the previous calendar month accurately reflected in the Health app?” by selecting “Yes, they are accurate,” “No, they are not accurate,” or “I prefer not to answer.” An optional, direct link to review and update the tracked data was offered before responding. If AUB was detected, we then defined an analysis window as having tracking confirmed if every month included in analysis had a response of “Yes, they are accurate.” Analysis windows with ≥1 “No, they are not accurate” response or no response (missing response) were considered as tracking not confirmed. Descriptive statistics were calculated, including means and standard deviations for continuous variables (eg, age, body mass index [BMI], etc.), and proportions and 95% score confidence intervals (CIs) for categorical variables (eg, race/ethnicity, gender identity, exercise minutes, sleep hours, etc.). 23 Proportions and 95% CIs were also calculated when estimating the prevalence of the AUB categories To understand associations between AUB and other characteristics for each AUB condition (any, irregular, infrequent, or prolonged menses, and spotting), multiple log-binomial models were estimated, with each containing a single subject characteristic as the independent variable. 24 In the regression of AUB prevalence against race/ethnicity, the reference group was White, non-Hispanic, with age and BMI being categorical covariates to isolate the effect of race/ethnicity. In the regression of AUB prevalence against BMI category, the reference group was the healthy weight group (18.5–24.9), with age and race/ethnicity being categorical covariates to isolate the effect of BMI category. Age categories were <25, 25 to 34, 35 to 44, and 45 to 50 years. BMI categories were underweight (40). 25 – 28 BMI was reported as kg/m 2 using self-reported height and weight. A directed acyclic graph for covariate selection is provided in supplemental materials . Because of the low numbers of participants reporting any individual medical condition, controlling for all combinations of medical conditions was not possible in the analysis evaluating the effects of BMI and race/ethnicity ( Supplemental Figure ). To understand the differences in AUB between participants with confirmed tracking who did and did not have a self-reported condition, we used a log-binomial model to regress an AUB outcome on each medical condition. The reference group were those who did not report the condition. 24

Overall, these findings provide the rate of AUB in a large, digital dataset of confirmed menstrual tracking. In addition to expanding our understanding of AUB across a diverse population, our findings confirm existing literature on the associations between AUB and medical conditions.

The menstrual cycle is considered a vital sign because regular menstruation requires normal functioning of the hypothalamic-pituitary-ovarian axis. 1 , 2 Abnormal uterine bleeding (AUB) may be a symptom of an endocrinopathy such as polycystic ovary syndrome (PCOS),or anatomic pathology such as fibroids, malignancies, or infections. 3 – 6 People track menstrual bleeding for a variety of reasons, such as trying to conceive, avoiding pregnancy, period prediction, or monitoring a condition. 7 Tracking bleeding is one way for patients to inform conversations with clinicians and for researchers to evaluate AUB from large samples. 8 The options for tracking have evolved, and menstrual tracking is most often done using digital platforms such as smartphone apps. 9 – 11 There exist evolving guidelines related to AUB to guide clinical research and care. 12 , 13 Definitions derived from these guidelines can be used to estimate the rates of AUB. Analysis of digitally tracked menstrual bleeding can present its own challenges, with absent, incomplete, or intermittent tracking limiting interpretation. 14 However, if self-tracking can be confirmed as accurate, the potential for AUB identification using app-based tools may bridge the gap between the self-tracked data and early screening for disease. Furthermore, this may facilitate population-level AUB analysis. 15 This study aimed to identify AUB patterns and their prevalence, and confirm existing associations between AUB patterns, demographics, and medical conditions in the Apple Women’s Health Study (AWHS) after confirming the accuracy of tracked data.