Decoding menstrual health across the lifespan: a scoping review of digital health tools in research.

Overall, we found scant passively recorded biometric or behavioural data during perimenopause and limited data regarding changes in menstrual cycle characteristics. This is in part because many of the digital health studies we reviewed set inclusion criteria for menstrual cycle regularity that removed the irregular cycles that occur around perimenopause. The studies we found showed that cycle length increased and cycle variability dramatically increased with age 51 , 53 . Li et al found that compared to 35–39 year olds, cycle variability increased by 46% in a 45–49 year-old-age group and by 200% at age 51–55 51 . Furthermore, Cunningham reported that over half of women reported irregular menstrual cycles by age 51–55 53 . Markovic et al., who included participants aged 19–90 without exclusions based on cycle length or age, found that perimenopausal and menstruating participants had a higher skin temperature than the non-menstruating group (which included 577 postmenopausal women, 2 pregnant, and 223 non-menstruating females) 48 . The menstruating participants also had a higher respiratory rate than the non-menstruating participants. Finally, Costeira et al. developed an app to investigate the role of oestrogen, which is considered to be immune stimulatory, in COVID protection and found that premenopausal women, especially those using contraceptive pills, exhibited higher COVID protection compared to postmenopausal women 110 . However, the same protective effect of oestrogen was not seen in postmenopausal women on hormone replacement therapy 110 . Thus, the menopausal transition is ripe with opportunity to better understand how ovarian hormones relate to changing biometrics. Our literature review revealed good agreement between digital health and lab-based studies, supporting the use of wearables and apps to study menstrual cycle physiology, but continued validation against “gold-standard” techniques is needed. To guide researchers in selecting devices with sufficient accuracy to capture changes in biometrics associated with the menstrual cycle, the following sections summarise studies that directly compare wearables to gold-standard measurement approaches, where available (Table 3 ). We searched PubMed, Web of Science, Google Scholar and company websites for studies and public data regarding the accuracy and precision of the commercial wearables used in the digital health studies we found in our scoping review. This included the Oura Ring, WHOOP wristband, Ava bracelet, Garmin watch, Apple Watch, Huawei Band 5, and Fitbit wristband. We provide the reported accuracy of skin temperature, resting heart rate, respiratory rate, HRV and sleep metrics. To understand the accuracy of wearable devices in capturing physical activity throughout the day, we refer readers to prior reviews 111 – 113 . Table 3 Guide to wearables, measured metrics, and associated accuracy Biometric recorded Device Method, sensor type and sampling/timing of measurement, if known Accuracy Reference, study type, N of study Skin temp Oura ring Negative temperature coefficient sensor; measures every minute Versus iButton sensor: Accuracy (undefined) = 0.36 °C; Precision = 0.13 °C; r ² > 0.92 (real world), r ² > 0.99 (lab) 116 In-house, n  = 16 (company site) Temperature immediately after wake versus oral temperature: r  = 0.563 79 External, devices provided, n  = 22 WHOOP 4.0 Temperature sensor during sleep n/a 148 (company site) Ava bracelet Temperature sensor n/a 149 (company site) Apple Watch S8 onwards, all Apple Ultra Two temperature sensors: one skin-facing and one outside-facing to remove environmental bias n/a 150 (company site) Garmin (select models only) Temperature sensor n/a 151 (company site) Fitbit Fitbit Sense 1/2 uses a dedicated temperature sensor. Fitbit Charge 4/5/6, Google Pixel Watch 2, Fitbit Inspire 2/3, Fitbit Luxe and Fitbit Versa use a multi-purpose sensor n/a 152 (company site) Resting heart rate Oura ring and oura ring 2.0 Blood pulse volume changes are detected through infrared PPG sensors in the ring; measured nocturnally every 10 min 153 (company website) Versus ECG (bpm): Spearman’s r 2  = 0.996; Bias = −0.63 122 Sponsored, n  = 49 Versus ECG (bpm): r 2  = 0.998; Bias = −0.53; Range: 48-66 123 In-house, n  = 10 Versus ECG (bpm): Absolute Bias = 1.8 ± 4.1; Bias = 0.1 ± 4.5; Intraclass correlation = 0.85 121 External, n  = 53 WHOOP 3.0 PPG sensor in non-wake in the primary sleep episode, every 30 s 154 Versus ECG (bpm): Absolute Bias = -0.7 ± 0.8; Bias = −0.3 ± 0.1; Intraclass correlation = 0.99 121 External, n  = 53 Ava bracelet PPG sensor Versus Polar chest strap, two electrode ECG (bpm): Bias = 7.8 ± 3.9; r  = 0.92; Absolute Bias percent = 11.4% 134 External, n  = 37 Garmin Forerunner 245 PPG sensor during a 3-min period as close as practicable to the start of each sleep period, i.e. ~30–60 min prior to lights out 121 Versus ECG (bpm): Absolute Bias = 5.4 ± 12.6; Bias = 5.0 ± 12.8; Intraclass correlation = 0.41 121 External, n  = 53 Garmin Forerunner 45 PPG sensor Versus ECG (bpm): Bias = 0.67 at rest. During two sequential ramps from rest to exercise, the bias was 10.82 bpm and 3.49 bpm. During steady state, exercise bias was 0.2 bpm. No difference with skin tone except during the second ramp (underestimated and >71% had <−5 bpm difference) 126 External, n  = 33 Apple Watch S6 PPG sensor during sleep 121 Versus ECG (bpm): Absolute Bias = 1.5 ± 1.5; Bias = 0.5 ± 2.1; Intraclass correlation = 0.96 121 External, n  = 53 Fitbit Charge 2 PPG sensor during rest Versus ECG (bpm): Non-significant difference=0.09, p = 0.43; Bias = 0.51 bpm when heart rate 80 bpm 155 External, n  = 17 Huawei Band 5 PPG sensor Limited published accuracy. Heart rate versus Polar chest strap, two electrode ECG (bpm) during elliptical exercise: r  = 0.981 156 External, Huawei provided devices, n  = 10 Heart rate error range ± 10 bpm 157 Company site Respiratory rate Oura Ring 2.0 Calculates rate via increase/decrease in heart rate with breath in/out Versus ECG-calculation (respiratory rate in breaths/min): r  = 0.96; Absolute Bias = 0.71 158 Internal, n  = 43 WHOOP 2.0 Calculates median rate during sleep via increase/decrease in heart rate with breath in/out 159 (company site) Versus respiratory inductive plethysmography (respiratory rate in breaths/min): Bias = 0.1; Percent bias = 1.8%; Precision = 1.0 129 External, n  = 32 (21 female) Garmin Forerunner 245 Calculates rate via increase/decrease in heart rate with breath in/out n/a n/a Ava bracelet Calculates rate via increase/decrease in heart rate with breath in/out Versus chest strap (respiratory rate in breaths/min): Absolute Bias = 0.9; Absolute Bias percent = 6.7% 160 External, n  = 7 Apple watch Accelerometer-based estimate of nocturnal breaths per minute n/a n/a Fitbit (charge 2–4, versa 1–2, Inspire HR) Calculates rate via increase/decrease in heart rate with breath in/out 161 Versus polysomnography (respiratory rate in breaths/min): root mean squared error = 0.648; Absolute Bias = 0.46; Absolute Bias percent = 3%; Bias = −0.24 162 Internal, n  = 28 HRV Oura Ring and Oura Ring 2.0 PPG sensor every 5 min throughout the night—reports RMSSD Versus ECG (RMSSD in ms): r 2  = 0.980; Bias = 1.2 122 Sponsored, n  = 49 Versus ECG (RMSSD in ms): r 2  = 0.967; Bias = −0.47; range = 21–78 123 In-house, n  = 10 Versus ECG (RMSSD in ms): Absolute Accuracy = 18.9 ± 35.9; Bias = −10.2 ± 39.4; Intraclass correlation = 0.63 121 External, n  = 53 Versus ECG (RMSSD in ms): r  = 0.915; Bias = −14.97; 95% confidence interval = [−44.07, 14.3] 163 External, n  = 35 WHOOP 3.0 PPG sensor in non-wake in primary sleep episode every 30 s—reports RMSSD 154 Versus ECG (RMSSD in ms): Absolute Bias = 4.7 ± 3.6; Bias = -4.5 ± 3.9; Intraclass correlation = 0.99 121 External, n  = 53 Ava bracelet PPG sensor 164 (company site)—reports LF/HF ratio Versus actigraph with Polar chest strap, 2 electrode ECG (LF/HF ratio): r  = −0.28 134 External, n  = 33 Garmin Forerunner 245 PPG sensor—reports RMSSD Versus ECG (RMSSD in ms): Absolute Bias = 33.1 ± 39.9; Bias = −22.4 ± 46.9; Intraclass correlation = 0.24 121 External, n  = 53 Apple Watch S6 PPG sensor—reports RMSSD Versus ECG (RMSSD in ms): Absolute Bias = 22.5 ± 19.2; Bias = −9.6 ± 28.1; Intraclass correlation = 0.67 121 External, n  = 53 Fitbit Sense PPG sensor—reports RMSSD n/a n/a Huawei Band 5 PPG sensor—type of HRV measurement unknown n/a n/a Sleep Oura Ring 2.0 Accelerometer + temperature sensor + PPG sensor for 4-stage categorisation (wake, light, deep, REM) Versus polysomnography (sleep vs wake): Agreement = 88 ± 5%; Sensitivity = 93 ± 5%; Specificity = 49 ± 22% 165 External, n  = 29 Versus polysomnography: (sleep vs wake) Agreement = 89%; κ = moderate; (sleep duration in mins): Absolute Bias = 29 ± 28.6; Bias = 1.5 ± 40.9 121 External, n  = 53 WHOOP 3.0 or 4.0 PPG sensor + accelerometer for 4-stage categorisation (wake, light, deep, REM) Versus polysomnography: (sleep vs wake) Agreement = 86%; κ = moderate; (sleep duration in mins) Absolute Bias = 30.3 ± 23; Bias = −12.2 ± 36.3 121 External, n  = 53 Garmin Forerunner 245 Accelerometer + PPG sensor + SpO2 sensor for 4-stage categorisation (wake, light, deep, REM) Versus polysomnography: (sleep vs wake) Agreement = 89%; κ = fair; (sleep duration in mins) Absolute Bias = 45.3 ± 36.3; Bias = 43.8 ± 38.0 121 External, n  = 53 Ava bracelet Accelerometer Versus actigraphy: (sleep duration in hours) Absolute Bias percent = 8.5%; r = 0.73; Bias = 0.18 ± 0.88 134 External, n  = 33 Apple Watch S6 PPG sensor in 5 min epochs; 3-stage categorisation (wake, light, deep) Versus polysomnography: (sleep vs wake) Agreement = 88%; κ = Fair; (sleep duration in mins) Absolute Bias = 48.1 ± 30.4; Bias = 39.5 ± 41.5 121 External, n  = 53 Fitbit (Alta 133 ; Charge 165 ; Non-sleep-staging models = Charge HR, Flex, Alta, One, Ultra, Classic; Sleep-staging models = Surge, Alta HR, Charge 2, Versa) PPG sensor + accelerometer for 4-stage categorisation (wake, light, deep, REM) 166 Versus actigraphy: (sleep duration in minutes) Mean Absolute Error = 56; Intraclass correlation = 0.84 133 External, n  = 30 Versus polysomnography: (sleep vs wake) Agreement = 84 ± 7%; Sensitivity = 89 ± 9%; Specificity = 49 ± 18% 165 External, n  = 27 Older sleep-staging models using movement only: Versus polysomnography: (sleep duration in minutes) Bias = 7–67 min; (wake duration in minutes) Bias = 6–44; (sleep efficiency) bias = 2–15%; no difference in sleep onset latency. (sleep vs wake) Sensitivity = 87–99%; Specificity = 10–52% Newer sleep-staging models using HRV and movement; Versus polysomnography: Newer models show no difference for sleep duration, wake after sleep onset, but underestimate sleep onset latency. (sleep vs wake) Sensitivity = 95–96%; Specificity = 58–69% 167 Systematic review with 8 papers with quantitative data, n  = 7–63 Huawei Band GT Uses accelerometry, HR, HRV for 4-stage categorisation (wake, light, deep, REM) 168 (company site) Versus actigraphy (sleep time assessment): Mean Absolute Percentage Error = 23.29%; Mean Percent Error = 20.0%; Intraclass correlation = 0.25; 169 External, n  = 102 Where possible, accuracy is reported using the following metrics: absolute bias (magnitude of error), bias (directional error) and correlation coefficients (relationship strength). All measurement terms are defined below the table PPG photoplethysmography, ECG electrocardiogram, RMSSD root mean square of successive differences, LF low frequency, HF high frequency Correlation measures: r: Pearson’s correlation coefficient, r 2 : coefficient of determination Accuracy measures: Bias: the difference between device measurement and gold standard measurement, including direction of difference (positive = overestimation, negative = underestimation); Absolute bias: the absolute difference between the device measurement and the gold standard measurement; Mean absolute error: the absolute difference between predictions of a model combining multiple sensor measurements and observed values. In cases where papers report MAE when comparing a device measure to a proxy gold standard measure, we rename MAE to Absolute Bias for consistency; Agreement: the percentage of device classifications that match gold standard classifications (e.g. for sleep versus wake period classification). For imbalanced data (long sleep versus short wake periods), the following measures are reported, when available: ● κ (kappa): agreement corrected for chance, accounting for class imbalance ● Sensitivity: percent actual sleep correctly identified ● Specificity: percent actual wake correctly identified Precision and reliability measures: Precision: assumed standard deviation unless specified (e.g. coefficient of variation = standard deviation/mean × 100%); Intraclass correlation: the degree of correlation between measures from the same group Guide to wearables, measured metrics, and associated accuracy 158 Internal, n  = 43 Fitbit (Alta 133 ; Charge 165 ; Non-sleep-staging models = Charge HR, Flex, Alta, One, Ultra, Classic; Sleep-staging models = Surge, Alta HR, Charge 2, Versa) Older sleep-staging models using movement only: Versus polysomnography: (sleep duration in minutes) Bias = 7–67 min; (wake duration in minutes) Bias = 6–44; (sleep efficiency) bias = 2–15%; no difference in sleep onset latency. (sleep vs wake) Sensitivity = 87–99%; Specificity = 10–52% Newer sleep-staging models using HRV and movement; Versus polysomnography: Newer models show no difference for sleep duration, wake after sleep onset, but underestimate sleep onset latency. (sleep vs wake) Sensitivity = 95–96%; Specificity = 58–69% Where possible, accuracy is reported using the following metrics: absolute bias (magnitude of error), bias (directional error) and correlation coefficients (relationship strength). All measurement terms are defined below the table PPG photoplethysmography, ECG electrocardiogram, RMSSD root mean square of successive differences, LF low frequency, HF high frequency Correlation measures: r: Pearson’s correlation coefficient, r 2 : coefficient of determination Accuracy measures: Bias: the difference between device measurement and gold standard measurement, including direction of difference (positive = overestimation, negative = underestimation); Absolute bias: the absolute difference between the device measurement and the gold standard measurement; Mean absolute error: the absolute difference between predictions of a model combining multiple sensor measurements and observed values. In cases where papers report MAE when comparing a device measure to a proxy gold standard measure, we rename MAE to Absolute Bias for consistency; Agreement: the percentage of device classifications that match gold standard classifications (e.g. for sleep versus wake period classification). For imbalanced data (long sleep versus short wake periods), the following measures are reported, when available: ● κ (kappa): agreement corrected for chance, accounting for class imbalance ● Sensitivity: percent actual sleep correctly identified ● Specificity: percent actual wake correctly identified Precision and reliability measures: Precision: assumed standard deviation unless specified (e.g. coefficient of variation = standard deviation/mean × 100%); Intraclass correlation: the degree of correlation between measures from the same group Many wearables now passively monitor skin temperature using negative temperature coefficient sensors that contain semiconductive materials that decrease their resistance with increasing temperature. Lab-based studies show a body temperature change of 0.3–0.7 °C across the menstrual cycle 21 , 77 and 1.1°C during perimenopausal hot flushes, based on the gold-standard rectal thermometer 114 , 115 . While reported accuracy compared to core body temperature across all wearables is lacking (Table 3), an in-house study of the Oura Ring demonstrated a high correlation ( r ² > 0.92), including an accuracy of 0.36 °C and precision of 0.13 °C when compared to skin temperature measured with i-Button monitors 116 . The i-Button is a research-grade wireless semiconductor skin-temperature sensor 117 also used as a standard in some validation studies. Comparison of skin temperature immediately after waking, measured with an Oura Ring, versus oral temperature showed a correlation of r  = 0.56 79 . These results suggest that the Oura Ring is likely sufficiently accurate to capture changes in temperature across the menstrual cycle and in response to hot flushes. But researchers must keep in mind that skin temperature is known to systematically present around 2 °C lower than 118 , and out of phase 119 with core body temperature changes. BMI can also impact the absolute value of recordings due to insulation with visceral fat, although the magnitude of change is unaffected 80 . Wearables use photoplethysmography sensors that detect minor variations in the intensity of light transmitted through the skin to estimate changes in blood flow and cardiac activity. Studies using gold-standard electrocardiograms (ECGs) have shown that resting heart rate increases by 2.3–3 bpm between the follicular and the luteal phase of the menstrual cycle 88 , 120 while changes during menopause are unknown. In one externally conducted study, six devices (Apple Watch S6, Garmin Forerunner 245 Music, Polar Vantage V, Oura Ring 2.0, WHOOP 3.0, and Somfit) were tested overnight in 53 participants. Compared to ECG, the WHOOP 3.0 device demonstrated the highest intraclass correlation of 0.99 for nocturnal heart rate and the lowest average over-estimation bias of 0.7 ± 0.8 bpm 121 . Most devices with nocturnal recordings available reported a good intraclass correlation above 0.85 and low bias of less than 2.6 bpm (Table 3 ), with the exception being a 0.65 intraclass correlation reported with Somfit, which is a head-worn sensor and was not used in any of the studies we reported. The Oura Ring 2.0 showed an intraclass correlation of 0.85 and overestimation bias of 1.8 bpm, but an Oura-sponsored 122 and an in-house study 123 reported higher r 2 values of 0.99 and smaller mean bias of −0.63. Together, these results suggest that many of the commercially available ring and wrist-worn devices are sufficiently accurate to measure changes in resting heart rate across the menstrual cycle, but there are considerations to keep in mind. For example, discrepancies in data resolution and sampling windows (e.g. raw versus summary data) can hinder comparison between devices. Specifically, the Garmin devices provided pre-sleep instead of nocturnal cardiac data, which may have contributed to reported lower accuracy, as it may not have captured a true resting heart rate 121 . There is some evidence that photoplethysmography-measured heart rate can be influenced by BMI 124 and skin tone 125 , but the temporal location of the peaks used to detect heart rate remained stable. A study using Garmin devices during rest and activity found that ECG- and PPG-measured heart rates only significantly differed during a transition from rest to activity 126 . Taken together, the evidence to-date indicates that measures of resting heart rate may be fairly robust across these demographic characteristics with scope for further validation and improvement. Wearable devices estimate respiratory rate using small photoplethysmography sensors on the skin that detect the high-frequency variations in heart rate linked to respiratory cycles, a feature called respiratory sinus arrhythmia. In the laboratory, respiratory rate during sleep is measured with belts that sense chest wall motion or with respiratory inductive plethysmography, where coils of wire are placed around the chest and abdomen 127 . Studies using this approach show that respiratory rate changes across the menstrual cycle range from 0.63 to 2 breaths/min 97 , 128 . We found limited data on wearable accuracy for respiratory rate (Table 3 ). One external study funded by a grant from WHOOP reported no significant difference between recordings with WHOOP 2.0 and respiratory inductive plethysmography (15.7 ± 1.7 versus 15.6 ± 1.7 breaths/min, respectively), with a mean bias of 0.1 breaths/min (1.8%) and a precision bias of 1.0 breaths/min (6.7%) 129 , providing some confidence in using photoplethysmography on a wrist-worn device to measure changes in respiratory rate related to the menstrual cycle, with additional validation needed for larger and diverse cohorts and for other wearables. Wearables estimate HRV using heart rate recordings from the photoplethysmography sensor. In-lab studies use electrocardiograms to record individual heartbeats and then calculate the variability in heartbeat timing, most commonly using the root mean square of successive differences (RMSSD) between the interbeat intervals over a sleep or rest period, reported in milliseconds or log-transformed values. A meta-analysis of 37 in-lab studies indicated a significant overall decrease in cardiac vagal activity from the follicular to the luteal phase, but only 2 of 8 studies reported a significant difference in RMSSD in the range of 9-20 ms 98 , 130 , 131 . The comparative study of Miller et al. found that the WHOOP 3.0 device demonstrated the highest HRV estimation accuracy compared to ECG measures, with a bias of −4.5 ± 3.9 ms with a 0.99 intraclass correlation 121 . An external study funded by an Oura grant reported a 0.98 r 2 value and −1.2 ms bias when compared to ECG measurements 122 , which is an improved performance from the 0.85 intraclass correlation reported in the six-device comparative study 121 . These results indicate researchers should be aware of underestimation bias according to their device, as underestimation could be up to 50% of the variability seen across the menstrual cycle. Detection of sleep using commercial wearable devices typically uses a combination of cardiac metrics such as resting heart rate, HRV, respiratory rate and movement metrics in proprietary sleep-detection algorithms, while polysomnography is the gold standard for measuring sleep in a laboratory 127 . The comparative study of six wearables found that all could perform the binary classification of sleep vs. wake periods throughout the night with relatively high accuracy (above 85%) compared to polysomnography. However, the accuracy in estimating total sleep duration varied widely, with mean errors on the order of 1 min overestimation for the Oura Ring and Polar watch, and 40 min or more for the Apple Watch and Garmin watch. All six devices showed low accuracy in identifying the sleep stages (e.g. slow-wave sleep versus rapid eye movement (REM) sleep), reporting less than 65% agreement with polysomnography 121 . Several other external studies show that commercial wearable devices tend to overestimate sleep duration. Sleep duration was overestimated by 15 min with the Oura Ring 132 , by 56 min with the Fitbit 133 , and 11 min with the Ava bracelet) 134 . Taken together, sleep stage estimation is inaccurate across ring and wrist-worn devices, while researchers prioritising estimated total sleep accuracy should be aware of the variation in accuracy across devices and choose devices accordingly.

Our scoping review adhered to the recommendations outlined in the PRISMA guidelines ( https://www.prisma-statement.org/scoping ) (see Table S1 ). We searched for articles on Google Scholar, Web of Science Core Collection, and PubMed through December 18, 2025 with the terms (“digital health” OR “wearable technology” OR “wearable devices” OR “mobile health” OR “smartphone application” OR “app”) AND (“scientific” OR “research outcomes” OR “insights” OR “study results” OR “clinical research”). We combined these digital health terms first with (“menstrual cycle” OR “menstruation” OR “period tracking”), and then with (“ovulation” OR “ovulatory cycle” OR “cycle tracking”). We removed duplicates, then two authors (S.C.J. and J.O.) independently reviewed titles, abstracts, and keywords of all the retrieved articles (see Fig. 2 ). Any disagreements regarding study eligibility were resolved through discussion and consensus between the two reviewers. Articles were included if they (i) were written in English; (ii) involved human participants; (iii) used wearable devices or mobile health apps; (iv) analysed female health physiology or biometrics; and (v) provided quantitative data collected with digital health tools. We excluded dissertations, theses, conference proceedings, conference abstracts and reviews. We excluded articles that did not include menstrual tracking with a digital health tool and/or articles where longitudinal data concurrent with menstrual tracking were not reported (Fig. 2 ). We excluded reviews as primary articles, but reviewed reference lists to identify any additional papers that met our criteria. We repeated the process, replacing the first grouping of search terms with “menopaus* OR perimenopaus*”. Fig. 2 PRISMA flow diagram in the search of papers associated with the menstrual cycle. Adapted from the PRISMA flow diagram 146 . Adapted from the PRISMA flow diagram 146 . We extracted the following outcomes from the included articles: scientific insights on female health with regard to menstruation and menopause; how menstrual cycle phase was determined (e.g. self-report or hormonal testing); and study details, including number of participants, menstrual cycles, participant demographics (age, BMI and ethnicity), and the digital health tool(s) used in the study.

This scoping review reveals several promising contributions of digital health studies to the understanding of the interplay between physiology, behaviours, symptoms and hormonally-driven processes, including the menstrual cycle and menopausal transition. Studies using digital health tools have begun to establish detailed, normative biometric data across the menstrual cycle. For example, while skin temperature is often used to estimate fertile windows, we found that digital health studies indicate that the largest decrease in skin temperature, and in some cases, the minimum temperature, can occur as early as day five of the menstrual cycle, unlike the 1–2 days before ovulation reported by lab studies. This knowledge could improve the efficacy of wearable-based tools for contraception or conception. Digital health studies have also provided additional evidence to support fluctuations in resting heart rate and HRV across the menstrual cycle, and discovered a decrease in the magnitude of this fluctuation with age. Understanding these changes across the lifespan can help women set realistic training and recovery expectations. In addition, digital health tools have begun to address how biometric changes are connected to hormone-related symptoms. For instance, studies using data from wearables found an inverse association with HRV and symptoms (both affective and physiological) in women with self-reported premenstrual disorder symptoms. This information is needed for quantifying the effectiveness of symptom-reducing interventions. Despite new data from digital health studies to characterise normative trends, our review of studies to-date reveals a need to better understand deviations from the norm, both within and between individuals. In some cases, whole cohorts have been omitted from most research. For example, most of the research we reviewed was carried out in predominantly White cohorts and in the United States, the United Kingdom, and Scandinavian countries, which limits our understanding of the relationships between race, ethnicity, and menstrual cycle physiology. Perimenopausal women were often omitted from studies, so characterisation of normal hormonal fluctuations and physiological changes during perimenopause is lacking. This is reflected by the fact that fewer than 1 in 10 women in the United Kingdom feel that they have received sufficient advice on what to expect during perimenopause 135 , with similar trends in the United States, where 64% of adult women reported a need for more doctors specialising in menopause 136 . Furthermore, we found that many studies focused on regularly menstruating women, often only including analysis of cycles with estimated ovulation, and excluding women with conditions impacting menstruation or using hormonal contraceptives. Menstrual conditions are common; endometriosis affects 10% of women, and many women are undiagnosed for years 137 , 138 . Although some atypical menstrual cycles may present no health concerns, identifying and understanding the causes of irregularity and anovulation and their biometric connections becomes crucial for better diagnostics and monitoring intervention efficacy. The relative accessibility and lower cost of digital health tools compared to lab-based approaches offer promise to include these understudied cohorts in research. Future digital health studies should concurrently measure hormones, biometrics, and menstrual-associated symptoms. While we expected to find many studies exploring relationships between these measures, studies to date were few. For example, the physiological impact of hormonal contraception, including effects on cardiovascular biometrics associated with athletic performance, remains under-characterised. For female athletes striving for peak performance, even small physiological changes may be significant, yet women are often left to use trial-and-error approaches when selecting contraception. Furthermore, we need to understand the timing and characteristics of the transition to a natural hormone cycle after individuals discontinue using hormonal contraception 139 . Using wearables could help to characterise these effects and contribute to informed contraceptive choices, but studies have not yet explored these relationships. Further, the connections between symptoms, biometrics, and hormonal fluctuations in the menopausal transition are markedly understudied. Digital health tools can help quantify how behaviour and physical or social environment may influence hormones and physiology, enabling the development of guidelines to manage symptoms and optimise performance, but again, we found few studies that have explored these areas. For example, future digital health studies could examine whether altering sleep in the luteal phase could mitigate premenstrual symptoms, or quantify the relationships between physical activity and the menstrual cycle, bringing much-needed evidence to guide the popular trend of “cycle syncing”. By understanding how behaviours such as sleep and exercise can mitigate or exacerbate symptoms related to the menstrual cycle or perimenopause, women could be empowered to manage their health and performance more effectively. Wearables could also serve as intervention tools. For instance, in a pilot study, a mobile app providing HRV biofeedback training showed promise in reducing premenstrual anxiety, stress, and poor sleep 140 . We found that most commercial wearables used in the studies we reviewed had sufficient accuracy to capture changes in heart rate, respiratory rate and, to a lesser degree, HRV. While sensors that measure skin temperature appear sufficiently accurate to capture changes across the menstrual cycle, this is based on only one study. Results for the accuracy of sleep duration were device-dependent, and other sleep characteristics (like sleep staging), should be used with caution. The superior accuracy for cardiac biometrics likely reflects the technological maturity of sensors and extensive validation history compared to more recently integrated features like sleep and skin temperature, suggesting accuracy will likely improve as hardware and signal processing algorithms evolve. The reported accuracies mirror the consistency of the findings in the Guide to Wearables section, highlighting how improvements in sensor technology, combined with access to continuous monitoring in large cohorts enables higher resolution data and statistical power that can clarify mixed reports and reveal new insights. Additional research is still needed to validate the accuracy of wearables in cohorts with diverse characteristics (e.g. BMI, skin colour and presence of tattoos), given the limited research to-date in this area. Furthermore, retrospective studies using passively collected wearable data may contain self-selection bias, as individuals who purchase fitness tracking wearables may be more active, of a higher socioeconomic status, and more health-conscious than average. However, providing wearable devices to participants and intentionally recruiting diverse populations can help to mitigate this bias. Researchers selecting digital health tools must consider their cohort, research question, and the limitations of each tool. For instance, self-reports cannot confirm if a menstrual cycle is truly absent or due to a lapse in reporting of menses onset. While biometric data can be used to predict menstrual cycle and/or ovulation occurrence with good accuracy, ovulation can sometimes occur without the typical biometric pattern changes and vice versa. A study of endurance runners may require a device with long battery life that measures heart rate and distance with high accuracy and resolution during activity only, while a study focused on recovery may require capturing accurate nocturnal cardiac and sleep parameters with constant daily activity monitoring. As the wearables we reviewed are wrist or finger-worn, it is important to note that devices using 2-electrode chest straps 141 and upper-arm-photoplethysmography sensors can be more accurate than wrist-worn devices 142 for measuring heart rate during activity. As technology continues to improve, we hope for the development of new tools to monitor hormones, physiology, and behaviour alongside consistent protocols and safe practices. Continuous and minimally invasive monitoring of biomarkers, such as hormones via sweat or interstitial fluid hold promise. Such monitoring could provide insights into normal, optimal, and pathological states, ultimately guiding more effective management of symptoms and health. As research in this field grows, it is important for researchers to establish and use standardised methodologies (e.g. measuring HRV during consistent sleep periods—such as the final hour before wake—versus varying times throughout the day) and data reporting (e.g. consistent HRV metrics). With the increasing use and development of digital health tools in women’s health research, there is also a vital need to address privacy concerns, particularly given the legal complexities surrounding pregnancy termination. Careful consideration must be given to research planning, data handling, and privacy, as discussed in detail elsewhere 143 , including specific considerations for female health data and privacy analyses of existing apps 144 , 145 . Digital health research offers promise in improving our understanding of the interaction of physiology, behaviour, and symptoms during the menstrual cycle and perimenopause. The use of digital health tools in female health is often focused on pregnancy prevention and fertility. By facilitating characterisation of healthy menstrual cycle norms across a more diverse group of women, digital health tools can enable understanding of healthy norms and identify when medical evaluation is warranted. Furthermore, integration of biometric monitoring with underlying hormone fluctuations, symptoms, and health behaviours can help to identify targets to monitor interventions and develop healthy biometric ranges. Not only is this understanding a vital first step toward developing effective solutions, but it also empowers women to take ownership of their health and enhance their quality of life.

Women experience hormonal fluctuations throughout the menstrual cycle and across the lifespan. But there is a lack of knowledge about how to effectively manage associated symptoms and optimise health, athletic performance, and productivity. Between 74 and 90% of women report symptoms, both prior to and during menses, for example, mood swings and abdominal bloating or dysmenorrhoea (pain during the menstrual cycle) 1 – 3 , with 20–40% estimated to experience premenstrual syndrome (PMS) 2 . Furthermore, 3–8% of women report severe and disabling premenstrual symptoms that meet the criteria for premenstrual dysphoric disorder (PMDD) 2 , 4 . These symptoms associated with the menstrual cycle can affect quality of life. For example, in a study of over 32,000 women in the Netherlands, 81% reported impaired work performance in the past three months due to the menstrual cycle, and 14% reported absenteeism in the past six months 5 . Another study found 45% of surveyed women using the Flo app and living in the United States, reported absenteeism related to the menstrual cycle in the previous twelve months 6 , while in low and middle-income countries, there was pooled absenteeism of 15% 7 . In Japan, the estimated annual economic burden due to menstrual symptoms was estimated at 8.6 billion United States Dollars 1 . Across a range of sports, more than 50% of female athletes reported a negative impact of menstrual and premenstrual symptoms on training and competition 8 – 10 . Further, perimenopausal women have 40% higher odds of displaying depressive symptoms than premenopausal women 11 , and consistently show moderately impaired quality of life 12 . The economic burden due to menopausal transition (perimenopause) symptoms costs the United States more than $26 billion annually 13 . The cyclical changes in hormones are well characterised throughout the menstrual cycle: during the follicular phase, beginning with menses, increasing concentrations of follicle-stimulating hormone (FSH) stimulate ovarian follicles to secrete oestrogen. Once a critical oestrogen threshold is reached, a surge in luteinizing hormone (LH) and a smaller spike in FSH occur, triggering ovulation. In the subsequent luteal phase, progesterone is secreted from the corpus luteum, which releases the egg; LH and FSH concentrations decline, and oestrogen concentration remains high. Subsequent decline of progesterone and oestrogen triggers menses. The gold standard for ovulation confirmation is transvaginal ultrasound imaging of follicular release, with serum testing to confirm the LH surge and progesterone peak as the second preferred choice. However, LH testing alone (typically in urine) and/or tracking of basal body temperature increases at ovulation are commonly used as non-invasive measures. The use of hormonal contraception, as well as the menopausal transition, alters endogenous hormonal profiles in different ways. Many contraceptives involve the exogenous administration of hormones. For example, combined oral contraceptives release synthetic oestrogen and progesterone systemically, which suppresses endogenous ovarian hormone production 14 . The hormonal IUD secretes synthetic progestin, which acts locally to cause endometrial thinning, inhibition of sperm motility, and cervical mucus thickening, but may also suppress endogenous hormones, and inhibit ovulation in some women 15 . During perimenopause, rising FSH concentrations reflect a declining number of ovarian follicles and responsiveness, resulting in pronounced oestrogen fluctuations before permanent decline 16 , 17 . Findings from lab studies have identified some connections between these hormonal changes in the menstrual cycle and physiological measures and symptoms. Lab studies have shown that progesterone consistently elevates core body temperature across various female cohorts 18 , 19 . This effect, consistent with the observed high progesterone and higher core temperature in the luteal phase, appears to be reduced by oestrogen 20 . During the luteal phase, heart rate also increases, potentially due to the influence of progesterone and oestrogen on the autonomic nervous system or as a secondary effect of elevated body temperature 21 . Progesterone increases respiratory function by increasing carbon dioxide sensitivity, breathing rate, and potentially bronchodilation via smooth muscle relaxation, while evidence of oestrogen’s interaction on this effect is contradictory 22 , 23 . Understanding links between symptoms and physiology is limited, but several lab-based studies suggest that women with severe premenstrual symptoms or PMDD often exhibit lower heart rate variability (HRV), indicating reduced parasympathetic activity, at different times in the menstrual cycle 24 – 27 . While these findings are valuable, our knowledge of the relationships between hormones, physiology, symptoms, and health behaviours is still limited. Advances in digital health tools, including wearable technology and smartphone apps, provide a unique opportunity to bridge critical knowledge gaps in female health, wellbeing, and performance by enabling large-scale, continuous, and cost-effective collection of physiological and behavioural data. In-lab approaches, while valuable, have been limited by sample size and diversity, measurement frequency, and a lack of accompanying physical activity and sleep data. Wearables can continuously measure heart rate, HRV, respiratory rate, skin temperature, sleep metrics, and activity metrics. Independent or associated smartphone apps allow users to record menses dates, daily symptoms, mental state, as well as activity and behaviour details. Digital health tools also include apps with non-wearable devices, for instance, temporary smartphone camera add-ons can be used with an app to provide quantitative readings of hormonal urine test strips 28 . Given the proliferation of digital health tools, there is a need to evaluate how they are being used in women’s health research. It is important to assess their accuracy, the discoveries they have enabled, and opportunities for future research. Past reviews of digital tools addressing women’s health have focused on identifying unmet needs in software design and hardware development 29 – 31 , accuracy of predicting ovulation and menses onset for contraception and conception purposes 30 , 32 – 35 , and the utility and implementation of technology tailored to monitor and diagnose conditions prevalent in women 36 – 38 . We found only one review that summarised findings from passive monitoring with commercial wearables, but this review focused solely on trends in thermoregulation relevant to the menstrual cycle 21 . No existing reviews summarise findings from digital health tools that concurrently track female-specific, hormone-induced events such as the menstrual cycle or menopausal transition and report physiological, behavioural, and symptomatic outputs. Further, no prior review has systematically evaluated whether digital technology has led to significant new research insights. There is also a critical need to understand when women’s health researchers can trust digital health tools, which can be assessed by both direct validation compared to gold standard measures, as well as indirect validation to determine if tools reproduce known relationships from lab-based studies. While many studies and reviews have evaluated the accuracy of individual wearable sensors to predict characteristics related to the menstrual cycle (e.g. length, phase), none have characterised accuracy in the context of women’s health research or across multiple physiological biometrics and behaviours (i.e. resting heart rate, HRV, respiratory rate, skin temperature and sleep). We conducted a scoping review of studies that used digital health tools—including mobile applications, wearable devices, or integrated app-device systems—to characterise changes associated with female-specific, hormone-driven events such as the menstrual cycle and menopausal transition. Our review includes menstruating females, individuals using hormonal contraceptives and those navigating the menopausal transition. The review excludes findings regarding pregnancy and postpartum. While we did not exclude studies that included women with gynaecological conditions such as endometriosis or gynaecological cancer, these conditions were not the focus of the review. We first summarise the knowledge gained from digital health studies about the interplay between physiology, behaviours, symptoms, and hormonally driven processes. We discuss new insights gained, and compare findings from digital health studies to prior lab-based work. We then summarise the reported accuracy and resolution of digital health tools to assist researchers in making informed decisions about device selection and digital health study design. We close by discussing key unmet research needs in female health, wellbeing, and performance that could be addressed using digital health tools in order to inspire future research.

In total, 40 studies were considered eligible for this scoping review based on the Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for Scoping Reviews (PRISMA) process (see Methods). Our search terms related to digital health tools (“digital health” OR “wearable technology” OR “wearable devices” OR “mobile health” OR “smartphone application” OR “app”) and menstrual health (“ovulation” OR “ovulatory cycle” OR “cycle tracking”), (“menstrual cycle” OR “menstruation” OR “period tracking”) identified 168 original research articles. We screened all 168 abstracts and the full text of 51 articles. We ultimately included 23 of the identified articles. From the references of identified articles and websites, we extracted an additional 43 papers and included 16 additional articles to bring the total from 23 to 39. Our search for digital health studies of perimenopause and menopause (“menopaus*” OR “perimenopaus*”) resulted in 15 articles for screening, and only four for inclusion, of which only one was unique to the menopause search, bringing the total number of eligible articles from 39 to 40. These were predominantly retrospective and observational prospective studies with 1 randomised control trial. The study cohort size ranged from 10 to 19,000,000 female participants, predominantly aged 18 to 55, but as young as 13 or as old as 90 in a few studies. We summarised reported non-biometric menstrual characteristics (Table 1 ; 17 relevant studies) and biometric and behavioural insights (Table 2 ; 29 relevant studies) across the menstrual cycle. Table 1 Menstrual cycle characteristics recorded with digital health tools. In this table, we summarise the results of all studies that reported at minimum cycle length and age Ref Digital health tool Cohort size Age (yrs; mean ± SD, [range]) Ovulation estimation method Was naturally cycling status confirmed? Additional inclusion criteria Cycle or phase length (days; mean ± SD, unless noted) Bleed length (days; mean ± SD, unless noted) Inter-cycle variability (days; mean ± SD, unless noted) Cycle Follicular Luteal 42 Natural cycles app 18,076 33 ± 6.8 Algorithm with BBT on for at least 10+ days in 1 cycle Yes No medical condition that may influence cycle regularity 29.4 4.2; range [4.2-4.2] 43 Clearblue connected ovulation test system 32,595 Age not reported, but app intended for 18+ LH urine test Yes Cycle length is not 50% shorter or longer than baseline cycle length 28 mode 15 mode For women with a 28-day cycle: 62%: ±2; 26%: ≥4; 19%: ≥5 44 Flo app 1.5 M 40% were 18–24, [18–55] LH urine test No, contraceptive pill use included 3 cycles of data available 28 mode 13 mode 15 mode 25%: 0- 1.5; 69%: 50% + LH if available Yes Cycle length 10–90 days 29.3 ± 5 16.9 ± 5.3 12.4 ± 2.4 4.0 ± 1.5 2.6 ± 2.5 46 Kindara app 13,674 29 ± 6 Algorithm using BBT and FAM Yes No midcycle bleed that is 4 or more days longer than the bleed 28 16 median 12 median Sympto app 199,293 30 ± 6 Algorithm using BBT and FAM Yes No midcycle bleed, ovulation detected 28 16 median 13 median 47 Luna Luna app 7043 33, [20–45] 31% clinical diagnosis, 54% test kit Yes Cycle length 21–45 days 29.8 14.8; range [10–23] 13.91; range [10–19] 48 Ava bracelet 179 41 ± 8, [19–55] An algorithm using 5 biometrics, including skin temperature, heart rate No, any contraceptive users included Vaccinated with >1 cycle 28 16.92 11.02 4.8 49 Ovia app 98,903 70% 25–34, [18-45] LH urine test Yes, newly attempting conception Cycle length ≥19 and <60 days 29.6 ± 5.4 50 Oura Ring 1007 51% 27–34, [18+] LH urine test Yes, no hormone use or pregnancy Cycle length ≥12 days and ≤90 days 7 51 Clue app 378,694 25, [21–33] None Yes None 29.7 ± 6 4.1 ± 1.8 4.15 ± 4.94 52 WHOOP 3.0/4.0 11,629 35 ± 7 None No, contraceptive pill use included Cycle length 21–35 days, bleed duration ≤7 days 27.4 ± 2 4.7 ± 1.1 9968 35 ± 7 None Yes 27.4 ± 2 4.7 ± 1.0 1161 31 ± 7 None No, only contraceptive pill users 27.7 ± 1 4.4 ± 1.1 53 Flo app 19.2 M 53% 18.5–24.9, [18-55] None No, but excludes contraception reminder users Cycle length ≥10 days 28.5 ± 3.2 5.2 ± 0.9 4.01 ± 4.74 54 FitrWomen app 20 athletes 25 ± 2, [19–28] None No, 4 contraceptive users included At least 1 cycle available 34.9 ± 11.6 55 Ovia app 45,360 31 ± 5, [18–45] None Yes Women who conceived within 1 year of tracking 29.8 4.4 ± 4.7 86%: <9 14%: 9–25 56 Clue app 6897 29% 18–24, [18+] None Yes Cycle length ≤90 days, vaccinated group 29.9 ± 7.1, 2nd cycle post vaccination If vaccinated during menses: 4.5 ± 1.8; else: 4.2 ± 1.8 662 Cycle length ≤90 days, unvaccinated group 29.6 ± 7.1 4.3 ± 1.6 57 Clue app 584 Median 26 None Yes Cycle length 24–38 days and bleed duration <9 days 29 4 median 223 Median 24 None No, contraceptive users 28 5 median 40 with STI Median 27 None Yes 30 4 median 18 with STI Median 26 None No, contraceptive users 28 4 median 76 Proprietary sensor + vaginal sensor 80 32, [22–46] Vaginal temperature recordings Yes, attempting conception PCOS diagnosis 40.9 ± 30.1 Hypothyroid diagnosis 31.2 ± 4 PCOS + Hypothyroid diagnosis 76.2 ± 104.8 No known diagnosis 35.4 ± 16.6 FAM fertility awareness methods, LH luteinizing hormone, BBT basal body temperature, PCOS polycystic ovary syndrome, STI sexually transmitted infection. Table 2 Insights from monitoring of menstrual cycles with digital health tools (naturally cycling women, unless otherwise stated) Finding Ref. Measurement tool Phase determination Cohort (n participants; mean age ± SD; n cycles, when available) Notable additional details 2.2.1 Characterising and refining normative trends in menstrual cycle length, menstrual bleed length, and timing of ovulation  Bleed phase duration changed with very short and long cycles 45 Natural Cycles app Ovulation is estimated algorithmically from self-reported BBT and LH urine test when available n  = 124,648; age = 30.3 yrs; cycles = 612,613 ovulatory Very short cycles had shorter menstrual bleed lengths by 12%. Very long cycles had longer menstrual bleed lengths by 6%  Long ovulatory cycles had proportionally longer follicular phases 45 Natural Cycles app Ovulation is estimated algorithmically from self-reported BBT and LH urine test when available n  = 124,648; age = 30.3 yrs; cycles = 612,613 ovulatory Very short 15–20 day cycles had 34% shorter follicular and 35% shorter luteal phases. Long 36-50 day ovulatory cycles had 66% longer follicular phases and only 5% longer luteal phases. 43 Clearblue connected ovulation test system Ovulation is estimated with the LH urine test n  = 32,595; age not reported, but app intended for 18 + ; cycles = 75,981 App intended to aid conception. The follicular phase correlated with cycle length (Pearson r  = 0.77, p  < 0.001). Luteal phase correlated with cycle length, but not as strongly ( r  = 0.37, p  < 0.001) 2.2.1 Characterising and refining normative trends in menstrual cycle length, menstrual bleed length and timing of ovulation: cycle irregularities  Irregular cycles were associated with more negative symptoms and less change in heart rate 51 Clue app Ovulation not estimated n  = 378,694; age = 21–33 yrs; cycles = 4.9 M Users with consistently highly variable menstrual cycle length (median cycle length difference ≥9 days) are more likely to report headaches and tender breasts 92 Ava bracelet, PulseOn, Basis Peak Ovulation is estimated with the LH urine test n  = 91; age = 33.2 ± 4.7 yrs; cycles = 274 ovulatory Irregular cycles (outside 24–35 day-range) showed less resting heart rate change between ovulation and the luteal phase (1.8 vs 2.1 bpm) and luteal to menstrual phase  Irregular cycles were most common at 51–55 years 53 Flo app Ovulation not estimated n  = 19,266,573; age = 18–55 yrs; cycles = ≥3 each Irregularity (≥2 cycle lengths per year differing by >7 days) was highest in age 51–55 (44.7%) and lowest in age 36–40 (28.3%)  In healthy cohorts, 8% showed abnormal cycle length ( 38 days), 2.5% showed anovulation 42 Natural cycles app Ovulation is estimated algorithmically from self-reported BBT and LH urine test when available n  = 18,076; age = 33 yrs; cycles = 214,426 8.7–8% show abnormal cycle length and menses >8 days Anovulation in 2.9% to 2.5% of cycles  Ovulation was not indicative of ovulatory function 66 Proov app Urine tests for LH, FSH, oestrogen & progesterone metabolites n  = 40; age = 34 yrs; cycles = 40 38/40 cycles were ovulatory (exhibited LH and progesterone rise), but only 22/40 had a sustained progesterone rise indicative of ovulatory function  Abnormal uterine bleeding was observed in 17% of naturally cycling women and was associated with menstrual conditions 75 Apple Research app Ovulation not estimated n  = 18,875; age = 33 ± 8.2 yrs; cycles = 16 ± 9 each Of the 17%, 2.9% had irregular menses, 8.4% infrequent menses, 2.3% prolonged bleeding, and 6.1% intermenstrual bleeding Bleeding was increased in women with PCOS (+19%), hyperthyroidism (+34%), hypothyroidism (+17%), endometriosis (+28%), cervical dysplasia (+20%), and fibroids (+14%)  51% of women reported spotting or absence of bleeding after hormonal IUD insertion 58 MyIUS app Ovulation not estimated n  = 1734; age = 18+ (42% age 18–25) yrs; cycles = not reported During the 90–270 day post-insertion period, women reported experiencing mainly spotting (41%), bleeding (48%), or amenorrhoea (11%) 2.2.2 Differences in menstrual cycle characteristics across demographic groups and other subpopulations: age  The amplitude of resting heart rate and HRV change decreased with age 52 WHOOP 3.0 or 4.0 Ovulation not estimated n  = 9968 naturally cycling, 3322 combined pill; age = 18+ yrs; cycles = 45,811 The amplitude change in resting heart rate significantly decreased with age Amplitude change in HRV significantly decreased with age 96 Clue app Ovulation not estimated n  = 499,000; age = 21.2 yrs; cycles = not reported, ≥12 months data per person The premenstrual magnitude of increase in resting HR decreased with age  Premenstrual weight gain was less with age 96 Clue app Ovulation not estimated n  = 499,000; age = 21.2 yrs; cycles = not reported, ≥12 months data per person The increase in premenstrual weight gain decreased with age  Premenstrual negative mood rose with age Premenstrual negative mood effect increased with age from age 15-20 to 30-35  Highly variable, short and long cycles were greater in year 1 post-menarche 70 Clue app Ovulation not estimated n  = 6,486; age = 16, range [13-18]; cycles = 38,916 51.5% of people showed variable cycles (45 days or bleed phase >8 days). Females <1 year vs ≥6 years post-menarche had lighter flow but more highly variable or short cycles  Earlier menarche was related to cycle length and bleeding Menarche age ≤10 vs ≥14 was associated with shorter cycle length, increased odds of dysmenorrhoea and higher cycle variability  Cycles shortened with age, then became more variable 53 Flo app Ovulation not estimated n  = 19,266,573; age = 18–55 yrs; cycles = ≥ 3 each From age 18–24, cycles shortened from 28.5 days to 27 days, and cycle variability decreased until age 46–50, when menstrual cycles and menses became longer and more variable 69 Apple Research app Ovulation not estimated n  = 12,608; age = 33.8 yrs; cycles = 165,668 Menstrual cycle length shortened by 2.4 days up to age 50, then became longer. Cycle variability was lowest at the age of 35–39, and increased 200% for the age 50+ 45 Natural Cycles app Ovulation is estimated algorithmically from self-reported BBT and LH urine test when available n  = 124,648; age = 30.3 yrs; cycles = 612,613 ovulatory From age 18–45, cycles shortened from 30 to 27 days, and cycle variability decreased by 0.5 days. Only the follicular phase was shortened  Luteal phase length was stable with age 44 Flo app Ovulation is estimated with the LH urine test n  = 1,579,819; age = 18+ (40% aged 18–24) yrs; cycles = ≥1 each Naturally cycling women are targeted by excluding users with oral contraceptive reminders. The luteal phase remained 15 days long across ages 2.2.2 Differences in menstrual cycle characteristics across demographic groups and other subpopulations: BMI  High BMI was associated with altered menstrual characteristics 45 Natural Cycles app Ovulation is estimated algorithmically from self-reported BBT and LH urine test when available n  = 124,648; age = 30.3 yrs; cycles = 612,613 ovulatory In ovulatory cycles, those with BMI > 35 had 0.4 days greater cycle variability compared with those with BMI 18.5–25. 44 Flo app Ovulation is estimated with the LH urine test n  = 1,579,819; age = 18+ (40% aged 18–24) yrs; cycles = ≥1 each Naturally cycling women are targeted by excluding users with oral contraceptive reminders. Women with the lowest BMI tended to have the highest cycle length variability. More women with a BMI > 35 had a longer median cycle length of ≥36 days than those with a BMI ≤ 35. 69 Apple Research app Ovulation not estimated n  = 12,608; age = 33 ± 8 yrs; cycles = 165,668 Menstrual cycle length increased with BMI (compared to participants with BMI 18.5–25, participants with BMI 30–34, 35–40, and ≥40 had 0.5, 0.8, and 1.5 days longer, respectively). Those with BMI > 25.5 had more cycle variability and a higher likelihood of a long cycle (>38 days) compared with those with BMI ≤ 25.5. 74 Luna Luna app Ovulation or anovulation is estimated with BBT n  = 8745; age = 30.1 ± 7.2 yrs; cycles = 191,426 Compared to a BMI of 20, those with a BMI of 16 or 30 had 1 day longer cycles and 1.1 or 1.5 days more cycle length variability, respectively. Compared to BMI 18.5–22.5, those with BMI > 35 had a higher risk of absent menstrual bleeding (OR 1.94), and those with BMI of 22.5–35 and BMI > 35 had a higher risk of irregular menstrual bleeding (OR of 1.56 and 2.63, respectively). 75 Apple Health app Ovulation not estimated n  = 18,875; age = 33 ± 8.2 yrs; cycles = 16 ± 9 each Prevalence of infrequent menses increased in those with BMI > 30, > 35, and >40 by 31, 25, and 51%, respectively. Those with a BMI > 40 had an 18% higher prevalence of abnormal bleeding.  Low BMI was associated with altered menstrual characteristics 44 Flo app Ovulation is estimated with the LH urine test n  = 1,579,819; age = 18+ (40% aged 18–24) yrs; cycles = ≥1 each Naturally cycling women are targeted by excluding users with oral contraceptive reminders. Women with the lowest BMI showed the most cycle length variation 74 Luna Luna app Ovulation or anovulation is estimated with BBT n  = 8745; age = 31.1 ± 7.2 yrs; cycles = 191,426 Compared to a BMI of 18.5–22.9, a higher risk of absent menstrual bleeding was found with a BMI < 18.5 (OR 1.78). Compared to those with BMI = 20, those with BMI = 16 had a greater (+1.03 day) mean cycle length and +1.1 day more cycle variability 45 Natural Cycles app Ovulation is estimated algorithmically from self-reported BBT and LH urine test when available n  = 124,648; age = 30.3 yrs; cycles = 612,613 ovulatory 0.2 day (5%) longer mean menstrual bleed length in those with BMI 15–18.5 compared to those with BMI 18.5–25)  HRV change was related to BMI 52 WHOOP 3.0 or 4.0 Ovulation not estimated n  = 9,968; age = 35 yrs; cycles = 45,811 The amplitude of HRV change across the menstrual cycle increased with BMI 2.2.2 Differences in menstrual cycle characteristics across demographic groups and other subpopulations: race and ethnicity  Ethnicity was associated with differences in cycle length and infrequent menses prevalence 69 Apple Research app Ovulation not estimated n  = 12,608; age = 33 ± 8 yrs; cycles = 165,668 Menstrual cycles were 1.6 days longer for Asian and 0.7 days longer for Hispanic participants, with more cycle variability compared to White non-Hispanic participants. No difference was seen with Black participants 75 Apple Research app Ovulation not estimated n  = 18,875; age = 33 ± 8.2 yrs; cycles = 16 ± 9 cycles each Black participants had a 33% higher prevalence of infrequent menses compared to White non-Hispanics after controlling for age and BMI 2.2.3 Physiological changes throughout the menstrual cycle  Skin temperature was lowest in the menstrual or follicular phase, then increased to a peak in the late luteal phase 85 Ava bracelet Ovulation is estimated with the LH urine test n  = 193; age = 33 ± 4 yrs; cycles = 705 ovulatory Range of 0.45 °C with minimum of 33.87 °C in the follicular phase 84 Oura ring (v 2.43.1) Ovulation is estimated with the LH urine test n  = 26; age = 24 ± 1 yrs; cycles = 1 ovulatory each Range of 0.32 °C. Minimum in ovulation (only 0.01°C less than menses) and highest in the luteal phase 79 Oura ring Ovulation is estimated with the LH urine test n  = 22; age = 35 ± 9 yrs; cycles = 2 ± 1 ovulatory each Range of 0.30 ± 0.12 °C with a lower value in the follicular compared to the luteal phase 80 Ava bracelet Ovulation is estimated with the LH urine test n  = 136; age = 33.66 ± 3.86 yrs; cycles = 437 ovulatory Range of 0.81 °C with a minimum of 35.23 °C in the follicular phase (not menses), and a peak in the early luteal phase 81 Ava bracelet Ovulation is estimated with the LH urine test n  = 57; age = 27 ± 4 yrs; cycles = 193, 170 ovulatory The range of 0.29 ± 0.21 °C is lower in the follicular compared to the luteal phase 82 Abdominal sensor Ovulation is estimated with the LH urine test n  = 41; age 39 ± 6 yrs; cycles = 100 Detecting the transition to the peak temperature. Change of 0.34 °C from 36.3 ± 0.03 °C to 36.7 ± 0.03 °C, transition ~4 days from LH surge 83 Fitbit sense Ovulation estimated with LH and oestrogen urine test n  = 49; age 18–30 yrs; cycles = 123 Change of 0.3 °C with a minimum of 33.7 ± 0.8 °C in the late follicular and a peak in the luteal phase of 34.0 ± 1 °C  Resting heart rate increased across the menstrual cycle 95 WHOOP ≤ 3.0 Ovulation not estimated, cycle divided into 4 equal phases n  = 3870; age = 34 ± 7 yrs; cycles = 13,535 Mean range of 3 bpm from 55.5 to 58.5 bpm with a minimum in the early follicular phase 92 Basis Peak, Ava bracelet, PulseOn Ovulation day estimated with the LH urine test n  = 91; age = 33 ± 5 yrs; cycles = 274 ovulatory Mean range of 3.8 bpm with a minimum in the menstrual phase 85 Ava bracelet Ovulation is estimated with the LH urine test n  = 193; age = 33 ± 4 yrs; cycles = 705 with LH surge Mean range of 3.9 bpm with a minimum of 56.6 bpm in the follicular phase 84 Oura ring (v 2.43.1) Ovulation is estimated with the LH urine test N  = 26; age = 24 ± 1 yrs; cycles = 1 ovulatory each Mean range of 2.7 bpm from 63.3 to 66 bpm, with a minimum during menses 52 WHOOP 3.0 or WHOOP 4.0 Ovulation not estimated n  = 9968; age = 18+ yrs; cycles = 45,811 Mean of 60 ± 8 bpm with amplitude 2.73 bpm ±1.9. Lowest point at day 5 and peak at day 26 93 Fitbit Inspire 2 Ovulation is estimated as 14 days prior to menses n  = 33; age = 24 ± 3 yrs; cycles = 50 Mean range of 4 bpm from 64 in menses, mean 67 ± 7 bpm. The study included 4 contraceptive users 94 HRV4Training app Ovulation not estimated, cycle bisected into two phases n  = 5425; age (all gender) = 37 ± 1 yrs; cycles = ≥5 each Mean 61 ± 8 bpm. 1.6% increase between the follicular and luteal phase. The study included an unspecified number of contraceptive users 91 Huawei Band 5 Ovarian ultrasound and serum testing n  = 89 regular, 25 irregular; median age = 31 yrs; cycles = 77 Range 2.7 bpm, with a minimum in the menstrual phase and a maximum in the luteal phase  Respiratory rate decreased in the late follicular phase, then increased through the luteal phase 95 WHOOP ≤ 3.0 Ovulation not estimated, cycle divided into 4 equal phases n  = 3870; age = 34 ± 7 yrs; cycles = 13,535 Range of 0.3 bpm with a minimum of 15.9 bpm in the late follicular phase 85 Ava bracelet Ovulation is estimated with the LH urine test n  = 193; age = 33 ± 4 yrs; cycles = 705 ovulatory Range of 0.64 bpm with a minimum of 16.4 bpm in the fertile window. Higher in the late luteal phase compared to menses  HRV decreased across the menstrual cycle 94 HRV4Training app Ovulation not estimated, cycle bisected into two phases n  = 5425; age (all gender) = 37 ± 12 yrs; cycles = ≥5 each Daytime RMSSD decreased 3.2% from the follicular to the luteal phase. The study included an unspecified number of contraceptive users 95 WHOOP ≤ 3.0 Ovulation not estimated, cycle divided into 4 equal phases n  = 3870; age = 34 ± 7 yrs; cycles = 13,535 Nocturnal RMSSD decreased from 70.1 ms in the follicular phase to the luteal phase in participants not using hormonal birth control. Range was 6.6 ms 85 Ava bracelet Ovulation is estimated with the LH urine test n  = 193; age = 33 ± 4 yrs; cycles = 705 with LH surge HRV ratio range of 0.24. Nocturnal HRV ratio trended higher in the follicular and fertile phase compared to the luteal phase, but was not significant after Bonferroni correction. 84 Oura ring (v 2.43.1) Ovulation is estimated with the LH urine test n  = 26; age = 24 ± 1 yrs; cycles = 1 ovulatory each Nocturnal RMSSD was lower in the late luteal phase than in the menses phase 52 WHOOP 3.0 or 4.0 Ovulation not estimated n  = 9968; age = 18+ yrs; cycles = 45,811 Nocturnal RMMSD peaked at day 5, decreasing to a minimum at day 27 with a range of 4.65 ± 6.9 ms 91 Huawei Band 5 Ovarian ultrasound and serum testing n  = 89 regular, 25 irregular; median age = 31 yrs; cycles = 77 The natural logarithm of SDNN decreased while the low frequency to high frequency ratio increased from the follicular to the luteal phase  Sleep did not change across the menstrual cycle 84 Oura Ring (v 2.43.1) Ovulation is estimated with the LH urine test n  = 26; age = 24 ± 1 yrs; cycles = 1 ovulatory each No difference in total sleep time, sleep onset latency, wake after sleep onset, and sleep stages. Sleep efficiency was marginally lower in the mid-luteal phase compared with menses 93 Fitbit Inspire 2 Ovulation is estimated as 14 days prior to menses n  = 33; age = 24 ± 3 years; cycles = 50 ovulatory No change in sleep duration between phases. Included 4 contraceptive users 147 Luna-Luna app Ovulation is estimated with the LH urine test n  = 10; age = 22 ± 1 yrs; cycles = not reported No significant differences in sleep duration across phases. Sleep quality was higher in the early and late follicular phase compared to the menses or the early or late luteal phase 2.2.3 Physiological changes throughout the menstrual cycle: Effects of hormonal contraception on physiology  Combined pill users showed an overall higher heart rate with less change across the cycle; different patterns of HRV and respiratory rate change 95 WHOOP ≤ 3.0 Ovulation not estimated, cycle divided into 4 equal phases n  = 3870 naturally cycling, 455 combined pill; age = 34 ± 7 yrs; cycles = 13,535 Resting heart rate range of 1.5 bpm with a minimum in the early follicular phase, peak of 59 bpm in the late follicular phase, and a decrease into the late luteal phase; unlike the increase from the follicular to luteal phase in naturally cycling women. Resting heart rate was elevated by 2 bpm throughout the follicular phase compared with that of naturally cycling women. 52 WHOOP 3.0 or 4.0 N/A n  = 9968 naturally cycling, 3322 combined pill; age = 18 + yrs; cycles = 45,811 Resting heart rate showed less fluctuation in women on the combined pill compared to those naturally cycling (0.28 vs 2.73 bpm, respectively) 95 WHOOP ≤ 3.0 Ovulation not estimated, cycle divided into 4 equal phases n  = 3870 naturally cycling, 455 combined pill; age = 34 ± 7 yrs; cycles = 13,535 HRV range of 2.5 ms; HRV decreased across the follicular phase, then rose during the luteal phase, unlike naturally cycling women, whose HRV decreased across the cycle. Respiratory rate range of 0.2 rpm, with a minimum of 16 respirations per minute (rpm) in the early follicular phase and a peak in the late luteal phase; contrast with a range of 0.45 rpm in naturally cycling women, whose lower minimum of 15.75 rpm occurred in the late follicular phase, though also peaked in the late luteal phase.  Progestin-only pill users showed similar resting heart rates and HRV to naturally cycling women, but increased respiratory rate across the menstrual cycle 95 WHOOP ≤ 3.0 Ovulation not estimated, cycle divided into 4 equal phases n  = 3870 naturally cycling, 269 progestin-only (pill or IUD not specified); age = 33.5 ± 7.3 yrs; cycles = 13,535 Heart rate range of 2.2 bpm, rising from early follicular to late luteal phase. HRV range of 5.5 ms; HRV decreased from 71 ms in the early follicular to 65.5 ms in the early luteal phase, and rose to 66.5 ms in the late luteal phase. Respiratory rate range of 0.2 rpm, with a minimum of 16 rpm in the early follicular phase and a peak in the late luteal phase; no decrease in the late follicular phase as seen in naturally cycling women. 2.2.4 Linking physiology and symptoms across the menstrual cycle  HRV metrics were inversely associated with premenstrual disorder (PMD) symptoms and mood, and decreased more with a more severe diagnosis 103 Huawei Fitness Tracker 6 Pro N/A n  = 125 without PMDs, 68 with PMDs; age = 20 ± 2 yrs; cycles = 293 In women with PMDs, nocturnal HRV metrics pre and post-menses were inversely associated with symptoms, particularly the affective Women with PMDs showed more HRV change in the premenstrual week compared to those with only PMS 104 Juli app (HRV data integrated from other apps or devices) Ovulation not estimated n  = 352 with depression; age = median 28 yrs; cycles = ≥2 each Mood decreased 3–14 days pre menses and was associated with lower HRV 0–3 days prior in clinically diagnosed women with depression Menstrual cycle impacted mood but not sleep; mixed results for exercise 93 Fitbit Inspire 2 Ovulation is estimated as 14 days prior to menses n  = 4 on hormonal contraception, 29 not; age = 24 ± 3 yrs; cycles = 50 ovulatory Participants with severe PMS versus no or low PMS walked fewer steps during the luteal phase (10,283 ± 6277) vs menses (11,694 ± 6458) 96 Clue app Ovulation not estimated n = 499,000; age = 21.2; yrs; cycles = not reported, ≥12 months data per person Menstrual cycle impacted mood, behaviour, and vital signs, while exercise and sleep behaviour were more consistent LH luteinising hormone, FSH follicle-stimulating hormone, PMD premenstrual dysphoric disorder, combining PMDD and PMS, BBT basal body temperature, RMSSD root mean square of successive differences, SDNN standard deviation of normal-to-normal intervals, BMI body mass index Menstrual cycle characteristics recorded with digital health tools. In this table, we summarise the results of all studies that reported at minimum cycle length and age For women with a 28-day cycle: 62%: ±2; 26%: ≥4; 19%: ≥5 7 86%: <9 14%: 9–25 If vaccinated during menses: 4.5 ± 1.8; else: 4.2 ± 1.8 FAM fertility awareness methods, LH luteinizing hormone, BBT basal body temperature, PCOS polycystic ovary syndrome, STI sexually transmitted infection. Insights from monitoring of menstrual cycles with digital health tools (naturally cycling women, unless otherwise stated) n  = 32,595; age not reported, but app intended for 18 + ; cycles = 75,981 n  = 9968 naturally cycling, 3322 combined pill; age = 18+ yrs; cycles = 45,811 n  = 19,266,573; age = 18–55 yrs; cycles = ≥ 3 each n  = 1,579,819; age = 18+ (40% aged 18–24) yrs; cycles = ≥1 each n  = 8745; age = 30.1 ± 7.2 yrs; cycles = 191,426 Oura ring (v 2.43.1) n  = 26; age = 24 ± 1 yrs; cycles = 1 ovulatory each n  = 91; age = 33 ± 5 yrs; cycles = 274 ovulatory Oura ring (v 2.43.1) N  = 26; age = 24 ± 1 yrs; cycles = 1 ovulatory each n  = 9968; age = 18+ yrs; cycles = 45,811 n  = 5425; age (all gender) = 37 ± 1 yrs; cycles = ≥5 each n  = 89 regular, 25 irregular; median age = 31 yrs; cycles = 77 n  = 5425; age (all gender) = 37 ± 12 yrs; cycles = ≥5 each Oura ring (v 2.43.1) n  = 9968; age = 18+ yrs; cycles = 45,811 n  = 89 regular, 25 irregular; median age = 31 yrs; cycles = 77 Oura Ring (v 2.43.1) n  = 3870 naturally cycling, 455 combined pill; age = 34 ± 7 yrs; cycles = 13,535 n  = 9968 naturally cycling, 3322 combined pill; age = 18 + yrs; cycles = 45,811 n  = 3870 naturally cycling, 455 combined pill; age = 34 ± 7 yrs; cycles = 13,535 n  = 3870 naturally cycling, 269 progestin-only (pill or IUD not specified); age = 33.5 ± 7.3 yrs; cycles = 13,535 n  = 125 without PMDs, 68 with PMDs; age = 20 ± 2 yrs; cycles = 293 n  = 352 with depression; age = median 28 yrs; cycles = ≥2 each LH luteinising hormone, FSH follicle-stimulating hormone, PMD premenstrual dysphoric disorder, combining PMDD and PMS, BBT basal body temperature, RMSSD root mean square of successive differences, SDNN standard deviation of normal-to-normal intervals, BMI body mass index

S1 Supplementary Data 1. S1 Supplementary Data 1.