Table 3.

Summary of studies conducted by internal company investigators or independent investigators on the performance of commercial wearable devices or apps that track changes in body temperature across the menstrual cycle.

Wearable	Reference	Site of temperature measurement	Methods	Results	Comment
BodyMedia FIT	175	Band placed on upper arm, continuously measures skin T	Compared pre-wake T (device) with basal oral T (digital thermometer) over 17 days in 15 women	- Poor agreement between device and thermometer; wide range of Ts with device (29.7–36.7°C). - Poor agreement in detecting ovulation	BodyMedia was acquired by Jawbone; products were discontinued in 2016
Duofertility	179	- Under-arm sensor patch continuously measures nocturnal skin T & movement. - Proprietary algorithms identify date of ovulation	Pilot study compared device with ultrasound monitoring for accuracy of detecting ovulation in 8 infertile women [18 cycles].	Device detected ovulation in all cycles.	- Independent study. - High sensitivity in this small sample. - There were no annovulatory cycles recorded in this group, therefore, specificity in detecting ovulatory cycles could not be determined. - Sensor relies on skin T, which is not always a good reflection of core body T.
Ava bracelet	177	Wrist skin T, continuously measured across the night	Used device to track nocturnal T changes across MCs in 136 women (437 cycles], aged 20–40 y, with regular MCs, confirmed as ovulatory (LH surge).	- Showed an upward shift in wrist skin T in 82% of cycles, with higher average skin T in LP than early FP. - In only 41% of cycles, a T nadir (lowest T in a given cycle) was detected in a 5-day period pre-ovulation.	- Study done by Ava team and collaborators. - Even when considering potential confounders [e.g. age, exercise, evening meal), effect of menstrual cycle phase remained. - Prolonged measurement of nocturnal skin T may be more robust than a single T measurement, however, skin T has limitations due to environmental T influences.
Ava bracelet	178	Wrist skin T, continuously measured across the night	Evaluated effectiveness of a fertility algorithm using multiple sensors in device (skin T, HR, HR variability, breathing rate] to estimate the ‘fertile window’ (encompasses ovulation). Data from 708 cycles in ~193 women.	- Skin T, HR and respiratory rate increased in LP. - Algorithm showed good performance (90% of cycles) in detecting 6-day fertile window, with specificity of 0.93 & sensitivity of 0.81.	- Study done by Ava team and collaborators. - Suggests that multi-sensor data may be more sensitive than a T sensor alone in detecting ovulatory cycles, and possibly predicting a fertility period.
Ouraring	180	Placed on the finger, measures skin temperature, HR, and HR variability continuously	- Study compared skin T continuously measured with device and oral waking T. Also evaluated algorithms using skin T to predict menses and ovulation. - Data used from 22 women (21–49 y] across 120 to 150 days. Ovulation measured from LH surge.	- Skin T with device was 0.30°C higher in LP vs FP, and correlated with waking oral T. - Sensitivity of the algorithm in detecting menses was 81.4% [window of ±3 days). - Sensitivity of the algorithm in detecting ovulation was 83.3% within a 6-day fertile window.	- Some investigators are employed by Oura Health. - Shows a difference in skin T between menstrual cycle phases, and good sensitivity in detecting menses and ovulation. - Oura ring temperature sensors were not calibrated before measurements so absolute values could not be used. - Sensor relies on skin T, which has limitations due to environmental T influences.
Natural Cycles app	181	Web/mobile app. Algorithm uses predictive mathematical models applied to data entered daily by user: basal body T measured with external device; LH surge measured with external device; self-report info (e.g. age, cycle length, BMI, OC use)	- Retrospective study performed on 1501 cycles from 317 women aged 18 to 39 years. - A combination of information (waking oral T, ovulation test results in a subset, date of menses) was used to predict fertility window	- Mean cycle length was 28.8 ± 5 days. - Algorithm estimated that day of ovulation was 1.9 ± 1.4 days after LH surge based on T data alone.	- Study done by company founders. - Shows accuracy in assessing the fertile window. - Requires a high level of user compliance to measure basal body temperature (oral) every morning over multiple cycles to ensure the predictive models are accurate, and relies on user-entered information, subject to human error.
Natural Cycles app	39	Web/mobile app.	- Description of menstrual cycle data from 612,613 ovulatory cycles in 124,648 users worldwide. - Age range: 18–45 y BMI: 15–50 kg.m−2; not using hormonal contraception within past 12 months	- FP length: 16.9 days (95% CI: 10–30), LP length: 12.4 days (95% CI: 7–17). - Cycle length ↓ by 0.18 days (95% CI: 0.17–0.18) and FP length ↓ by 0.19 days (95% CI: 0.19–0.20) per year from 25 to 45 years. - Variation of cycle length was 14% higher in women with a BMI >35 relative to BMI of 18.5–25.	- Study funded and conducted by Natural Cycles Nordic AB. - Results are not fully representative since only cycles with ovulation detected were included in the study (ovulation was not detected in 48% of cycles, of which most did not have sufficient body T measurements to enable detection). - Same limitations as above.
Tempdrop	33	Placed under the arm; measures environmental and skin T, and activity continuously across the night		No data available
YONO	182	Sensor placed in the ear canal, measures tympanic T every 5 minutes across the night.	- Publication describes development of an adaptive statistical learning algorithm to predict the point of thermal shift using historical body T data. - Thermal shifts are used to forecast day of ovulation. - Data are from 125 cycles, from 34 users (22–42 y], data collection periods range from 28 to 222 days.	- Using 64 cycles with sufficient T data, ovulation [LH surge] (± 3 days) was detected with 92.3% sensitivity using T sensor. - Historical data was used to predict ovulation in 39 cycles from 22 users. Rate of correct prediction using T sensor data was 76.92%, which improved further with 2 historical cycles.	- Data collected by YONO labs. - Importantly, sensor was calibrated in a water bath and had low measurement error (±0.05° C). - Relies on user wearing the sensor; there were missing data as a result of users forgetting to wear sensor, or not taking it when traveling. - Relies on tympanic T, which can be contaminated by environmental T.
OvuSense fertility monitor [conference abstract)	48	Intra-vaginal T sensor that continuously measures core body T across the night.	- Conference abstract details comparison of the accuracy of vaginal T-based fertile period prediction (device) with that of LH surge, combination of LH/ultrasound folliculometry, or morning oral T. - 81 cycles from 21 women were analyzed.	- Device-predicted probability of conception per fertile day curve closely matched that of ultrasound/LH combination. - Device prediction of fertile period was superior to oral T.	- Some authors are paid consultants of Fertility Focus Ltd. - Uses a true core body T measurement site and tracks temperature continuously across the night. - Other data shows user acceptability [183], however, validity data are only briefly presented in abstract form. - More invasive T measurement than other body sites.
Ovularing	184	Intra-vaginal T sensor that measures core body T. Algorithms are applied to predict fertile window retrospectively and prospectively.	- Study of 470 cycles in 158 women using Ovularing. - Hormonal assessments of LH, follicle-stimulating hormone, estradiol and progesterone, and vaginal ultrasound were performed in a sub-set across days 9–36 of the cycle, although unclear how these data were used.	− 83.4% of the cycles were biphasic. – Ovulation day was determined 96.5% of the time in retrospective cycles. - A 7-day window of fertility could be predicted in 88.5% of prospective cycles, following measurement of 3 ovulatory cycles.	- Unclear what was the gold-standard against which detection of ovulation from device was compared. - Uses a true core body T measurement site and tracks vaginal T continuously during use. - More invasive T measurement than other body sites.

; BMI, body mass index; FP, follicular phase; HR, heart rate; LH, luteinizing hormone; LP, luteal phase; MC, menstrual cycle; OC, oral contraceptives; T, temperature