The diagnostic accuracy of wearable digital technology in detecting fertility window and menstrual cycles: a systematic review and Bayesian network meta-analysis.

Eight electronic databases (PubMed, Web of Science, Embase, CINAHL, Scopus, Science Direct, IEEE Xplore, EI village) were systematically searched for articles published until January 1, 2025. Details of the search strategies are available in the Supplementary Note (Searching strategy) for all the databases searched. Studies included participants of women of reproductive age to track the fertility window by means of calendar estimation, self-measured BBT, WDT, or electronic hormone testing systems. The index involved the use of WDT and other ovulation-tracking methods, including ① calendar estimation: evaluating ovulation based on historical cycle length. The ovulation day should be the 14th day before the first day of the next menstrual bleeding. ② self-reported BBT: evaluating ovulation based on the progesterone-induced thermogenesis in the post-ovulatory phase. The participants take body temperature immediately upon waking in a ‌consistent daily measurement time, and record in a BBT chart. The ovulation day should be the day with a temperature shift. ③ electronic hormone testing system: evaluating ovulation based on the urinary luteinizing hormone levels, which are tested by a strip and then analyzed by a mobile-mounted, app-connected home-based device, reporting a binary digital result or quantified hormone levels. The targeted conditions were the performance metrics (accuracy, sensitivity, specificity, PLR, NLR, DOR, and SROC) of the methods aforementioned in detecting fertility window, compared with a reference standard of ovulation. The gold standard for ovulation documentation is direct ultrasound visualization of an egg extruded from the ovary. Indirect validation included urinary progesterone excretion (PdG) 74 , the midcycle LH peak 75 , and serum progesterone levels 76 . Eligible studies included cross-sectional, case-control or cohort studies that were published exclusively in the English language were considered for inclusion. We excluded the studies with participants undertaking oral contraceptive pills since the BBT and hormone testing methods for ovulation determination are based on natural hormone fluctuations. The studies without full text or crucial demographic information were also excluded if the authors could not be contacted for additional information. This study was registered at PROSPERO (CRD42024601664) and subsequently amended to include a NMA to compare the diagnostic accuracy of different WDT designs, parameters, and algorithms, in addition to the original protocol. The results were reported following the PRISMA NMA guidelines 77 . WDT-detected fertility windows varied across different studies. There were 2 studies reported on the detection of exact ovulation day 41 , 47 , 2 studies reported one day before to after ovulation (±1 day) 47 , 63 , 4 studies reported 2 days before to after ovulation (± 2 days) 38 , 44 , 47 , 5 studies reported 3 days before to after ovulation (± 3 days) 37 , 40 , 42 , 47 , 63 , and 5 studies reports 5 days in advance to ovulation until the ovulation day (-5 days to ovulation) 36 , 39 , 43 , 45 (Supplementary Table 6 ). For each outcome, we only synthesized the highest accuracy to avoid dependency in effect sizes if the study reported multiple detection intervals. The primary outcomes were ① the accuracy of WDT, self-reported BBT, electronic hormone testing system and calendar estimation in detecting fertility window; ② the specificity, sensitivity, PLR, NLR, DOR, and SROC of WDT and self-reported BBT; Secondary outcomes were ① the accuracy of WDT based on distal BBT and proximal BBT in detecting fertility window; ② the accuracy of WDT with BT-based algorithm and multi-parameter-based algorithm in detecting fertility window; ③ the accuracy of WDT with algorithm of random forest model, linear mixed model and other AI models in detecting fertility window; ④ the accuracy of different modality of WDT (bands and rings) in detecting fertility window; ⑤ the accuracy of different interval of WDT in detecting fertility window. Accuracy was defined as the percent of cycles correctly classified by a method (the number of cycles correctly classified/total number of cycles). Sensitivity was defined as the ability to detect a true ovulation (the number of cycles correctly classified as ovulatory/total true number of ovulatory cycles. Specificity was defined as the ability to detect a true anovulation (the number of cycles correctly classified as anovulatory/total true number of anovulatory cycles). PLR was defined as the ratio of the probability of true ovulation to the probability of a false ovulation (the number of cycles correctly classified as ovulatory/total true number of ovulatory cycles divided by the number of cycles incorrectly classified as ovulatory/total true number of anovulatory cycles). NLR was defined as the ratio of the probability of a false anovulatory to the probability of a true anovulatory (the number of cycles incorrectly classified as anovulatory/total true number of ovulatory cycles divided by the number of cycles correctly classified as anovulatory/total true number of anovulatory cycles). DOR was defined as the ratio of the PLR to the NLR, representing the overall discriminative power of the diagnostic method. Two independent authors (Y.S. and C.C.W.) screened each article (title, abstract, and full text) eligible for the review and extracted data comprising the following components: characteristics of the studies (first author, year, region), characteristics of the subjects (age, BMI, health status, menstruation regularity), characteristics of the index test (parameter, algorithm, window range, measured time), characteristics of the standard (method for ovulation determination), number of participants analyzed, number of menstrual cycles analyzed, account of each outcome, and mean and/or standard deviation (SD), or the range of the continuous data. A third reviewer (Y.W.) resolved discrepancies between the 2 reviewers. We contacted the authors of the studies to request clarification for incomplete data. Risk of bias was assessed using the Quality Assessment of Diagnostic Accuracy Studies–Revised (QUADAS-2) 78 in Review Manager 5.4. Two review authors (Y.S. and C.C.W.) assessed the risk of bias for each trial via the risk of bias tool from 4 aspects: patient selection, index test, reference standard, and flow and timing. We ranked each domain as either a ‘low’, ‘unclear’ or ‘high’ risk of bias, and the first three in terms of concerns regarding applicability. We discussed with a third review author (Y.W.) to resolve discrepancies. The certainty of the evidence was assessed using the GRADE approach for the outcomes 79 . Data from independent cohorts within a study were synthesized into an outcome. Given the likelihood of increased inter-observation variance, a random-effects model was used to assess the pooled accuracy, sensitivity, specificity, PLR, NLR, and DOR. All statistical analyses were performed by using R version 4.1.0 (R Statistical Computing). Forest plots were created by using R package metafor and meta . Between-study heterogeneity was assessed using the Cochran Q statistic ( p  < 0.05 indicated heterogeneity), the between-study variance was assessed using I 2 , and the magnitude of between-study variation due to true differences in effect sizes rather than chance will be assessed using I 2 . I 2  ≥ 50% was considered substantial heterogeneity 80 . For any observed substantial heterogeneity, the possible reasons for this were examined by subgroup analyses to explore whether the diagnosis performance was moderated by the reference standards, health status, and menstrual cycle regularity. Chi-square test was applied to compare the detection accuracy among subgroups, and p  < 0.05 indicated a significant difference. The NMA was performed by using R package gemtc for the accuracy, sensitivity and specificity of the different methods, and different designs and intervals of WDT for fertility window detection, comparing with reference standard. We first checked the underlying assumption of NMA. In principle, all participants were able to observe fertility windows and applied similar methods or designs to monitor the fertility window across the studies within each group. We therefore determined that the assumption of transitivity and homogeneity held 81 . For the consistency test, we performed node-splitting assessments to determine the association between the direct and indirect evidence. All I 2 values = 0%, indicating consistency between the direct comparison, indirect comparison, and network pooled results. For outcomes without closed loops in the network, heterogeneity was assessed using I 2 , and all I 2  = 0% indicated low heterogeneity. Therefore, the consistency model was applied to the studies. Based on the held assumptions, we summarized the geometry of each evidence network using network plots for each outcome. Markov chain Monte Carlo (MCMC) simulation and potential scale reduction factor (PSRF) were used to evaluate the convergence of iteration. The rare variations and stability of various plots, and the PSRF value of all estimated outcomes of approximately 1.00 indicated complete convergence, good iterative effect, and stable results of the model. The ranking probabilities of each detective method were generated according to the Bayesian approach. The cumulative ranking was evaluated by SUCRA using R package meta4diag and INLA . The larger the SUCRA the higher its rank among all available methods 82 . Publication bias was assessed with the symmetry of Deeks’ funnel plot using R package meta and ggplot2 83 , and an asymmetric funnel plot indicated the presence of publication bias. Sensitivity analyses were conducted to assess differences in the pooled effect estimates using a random-effect model, after removing each study 84 .

We retrieved 1653 records from 8 databases, and 140 potential full texts were retrieved and assessed for their eligibility. Among them, 76 studies were conference abstracts or patent abstracts, and the full text or crucial demographic information and data for synthesis were unavailable. After the full-text screening, 27 studies were included, involving 6244 participants and 14288 menstrual cycles in this systematic review and NMA (Fig. 1 ). The characteristics of the included studies using the WDT and conventional methods were stated in Table 1 and Table 2 , respectively. Fig. 1 PRISMA flowchart of study search and selection strategy. We retrieved 1653 records from 8 databases, and 140 potential full texts were retrieved and assessed for their eligibility. Among them, 76 studies are conference abstracts or patent abstracts, and the full text or crucial demographic information and data for synthesis were unavailable. After the full-text screening, 27 studies were included for our Bayesian network meta-analysis. The figure was created by Microsoft Office PowerPoint 2021. Table 1 Characteristics of the included studies using wearable digital technology for detecting the menstrual cycle and fertility window Author Region Study design Participant/Cycle Age BMI Health status Menstruation Standard Time Analytical parameter Algorithm Outcome Niggli A.,2023 Switzerland Cross-sectional 61/205 26.5 ± 4.2 22.1 ± 2.9 healthy regular LH urine test night WST, HR, HRV, RR, SP random foreset model ①②③④⑤⑥⑦ 3268/6081 32.5 ± 4.3 25.5 ± 6.27 healthy regular WST, HR, HRV, RR, SP Yu J.L.,2022 China Cross-sectional 89/305 32.00 (27.00, 35.00) 20.82 (19.33, 22.59) healthy regular ultrasound, serum PdG levels night WST, (WST + HR) linear mixed model ①③④⑤⑥⑦ 25/77 28.00 (25.00, 31.00) 20.70 (19.29, 23.51) healthy irregular WST, HR Zhu T.Y.,2021 Switzerland Cross-sectional 57/193 26.7 ± 4.2 22.5 ± 3.6 healthy regular+irregular LH urine test night WST linear mixed model ①②③④⑤⑥⑦ Goodale B.M.,2019 Switzerland Cross-sectional 193/708 33.02 ± 3.68 22.70 ± 3.40 healthy regular LH urine test night WST, HR, RR random forest model ①②③④⑤⑥⑦ B.S.H.,2022 US, UK Cross-sectional 80/205 32 NA healthy+PCOS+hypothyroid regular+irregular overnight vaginal temperature night WST “spring- loaded beam” method ①②③④⑤⑥⑦ Shilaih M.,2018 Switzerland Cross-sectional 194/793 33.66 ± 3.86 22.97 ± 3.68 healthy regular LH urine test night WST linear mixed model ①③④⑤⑥⑦ Li Q.,2024 China Cross-sectional 93/93 NA NA healthy regular LH urine test night FST NA ①③④⑤⑥⑦ Gombert-Labedens M., 2024 US Cross-sectional 111/111 18-52 NA healthy regular LH urine test night FST cosinor curve ①③④⑤⑥⑦ Luo L.,2020 US Cross-sectional 22/39 NA NA NA NA LH urine test night BT hidden Markov model ①③④⑦ Prasannan R.,2020 India Cross-sectional 30/30 30-40 NA NA NA NA NA BT support vector machine ①③④⑦ Regidor P.A.,2018 Germany Cross-sectional 158/470 18-45 18.2-36.9 healthy+PCOS regular+irregular LH, FSH, E2, PdG urine test all day BT NA ①③④⑦ Weiss G.,2022 Austria Cross-sectional 74/74 20-40 18.5-30 primary/secondary infertility+frozen embryo transfer+healthy regular ultrasound, serum progesterone levels, urine LH test all day BT NA ①③④⑦ Sato D., 2024 Japan Cross-sectional 26/74 NA NA healthy NA LH urine test night BT linear mixed model ①②③④⑦ Abbreviations: BMI body mass index, BT body temperature, E2 estradial, FSH follicle stimulating hormone, HR heart rate, HRV heart rate variability, LH luteinizing hormone, NA not appliable, RR respiratory rate, SP skin perfusion, PdG pregnanediol, US United States, UK United Kingdom, WST wrist skin temperature, FST finger skin temperature, PCOS polycystic ovary syndrome. ① primary outcome 1: the accuracy of WDT, self-reported BBT, electronic hormone testing system and calendar estimation in detecting fertility window; ② primary outcome 2: the specificity, sensitivity, positive likelihood ratio, negative likelihood ratio, diagnostic odds ratio and SROC of WDT and self-reported BBT; ③ secondary outcome 1: the accuracy of WDT based on distal BT and proximal BT in detecting fertility window; ④ secondary outcome 2: the accuracy of WDT with BT-based algorithm and multi-parameter-based algorithm in detecting fertility window; ⑤ secondary outcome 3: the accuracy of WDT with algorithm of random forest model, linear mixed model and other AI models in detecting fertility window; ⑥ secondary outcome 4: the accuracy of different modality of WDT (bands and rings) in detecting fertility window; ⑦ secondary outcome 5: the accuracy of different interval of WDT in detecting fertility window. Table 2 Characteristics of the included studies using conventional methods for detecting the menstrual cycle and fertility window Author Region Study design Participant/Cycle Age BMI Health status Menstruation Standard Time Analytical parameter Outcome Zhu T.Y.,2021 Switzerland Cross-sectional 57/193 26.7 ± 4.2 22.5 ± 3.6 healthy regular+ irregular LH urine test night BBT ①② Luo L.,2020 US Cross-sectional 22/39 NA NA NA NA LH urine test night BBT ① Martinez A.,1992 Netherlands Cross-sectional 88/210 NA NA NA regular ultrasound night BBT ① Tabbaa S., 2024 US, Canada Cross-sectional 9/12 NA NA healthy regular ultrasound, serum LH levels night BBT ① Sato D., 2024 Japan Cross-sectional 26/74 NA NA healthy NA LH urine test night BBT ①② Guermandi E.,2001 Italy Cross-sectional 101/101 31.8 ± 3.4 NA NA NA ultrasound night BBT ① Behre H.M.,2000 Germany Cross-sectional 53/150 26 NA healthy regular ultrasound morning urine LH + E3G ① Pattnaik S.,2023 US Cross-sectional 52/52 27.6 ± 4.1 NA healthy regular PdG urine test 7 days after LH peak morning urine LH + E3G ① Mu, Q.,2023 US Cross-sectional 15/43 33.07 ± 4.50 23.57 ± 1.00 healthy regular CBFM morning urine LH, urine LH + E3G ① Bouchard T. P.,2019 US Cross-sectional 13/34 33.6 ± 6.4 NA healthy regular LH urine test morning urine PdG ① Wegrzynowicz, A. K.,2022 US Cross-sectional 40/40 35 NA NA NA PdG urine test morning urine FSH, E1G, LH, PdG ① MacGregor E. A.,2005 UK Cross-sectional 27/1122 43 NA menstrual related migraine NA PdG, LH, FSH, E1G urine levels morning urine LH + E3G ① Thakur R., 2020 India Cross-sectional 360/360 NA NA NA NA NA morning urine LH ① Barron M. L.,2018 US Cross-sectional 42/219 NA NA NA NA LH urine test morning & evening urine LH + E3G ① Mouriki. E,2019 Switzerland Cross-sectional 34/34 NA NA healthy regular LH urine test morning Standard Days method ① Johnson S.,2018 UK Cross-sectional 768/768 32 26.67 healthy+PCOS+endometriosis regular+ irregular LH urine test morning 73 calendar apps- cycle length,cycle variability,last menstrual period ① Suman S., 2023 India Cross-sectional 161/1685 21-43 NA NA NA NA NA artificial neural network and multiple linear regression based on age, luteal phase, previous cycle of ovulation, previous length of menses ① Abbreviations: BBT basal body temperature, BMI body mass index, CBFM Clearblue Fertility Monitor, E1G oestrone-3-glucuronide, E3G PdG pregnanediol, FSH follicle stimulating hormone, LH luteinizing hormone, PCOS polycystic ovary syndrome, NA not appliable, US United States, UK United Kingdom. ① primary outcome 1: the accuracy of WDT, self-reported BBT, electronic hormone testing system and calendar estimation in detecting fertility window; ② primary outcome 2: the specificity, sensitivity, positive likelihood ratio, negative likelihood ratio, diagnostic odds ratio and SROC of WDT and self-reported BBT. We retrieved 1653 records from 8 databases, and 140 potential full texts were retrieved and assessed for their eligibility. Among them, 76 studies are conference abstracts or patent abstracts, and the full text or crucial demographic information and data for synthesis were unavailable. After the full-text screening, 27 studies were included for our Bayesian network meta-analysis. The figure was created by Microsoft Office PowerPoint 2021. Characteristics of the included studies using wearable digital technology for detecting the menstrual cycle and fertility window “spring- loaded beam” method Abbreviations: BMI body mass index, BT body temperature, E2 estradial, FSH follicle stimulating hormone, HR heart rate, HRV heart rate variability, LH luteinizing hormone, NA not appliable, RR respiratory rate, SP skin perfusion, PdG pregnanediol, US United States, UK United Kingdom, WST wrist skin temperature, FST finger skin temperature, PCOS polycystic ovary syndrome. ① primary outcome 1: the accuracy of WDT, self-reported BBT, electronic hormone testing system and calendar estimation in detecting fertility window; ② primary outcome 2: the specificity, sensitivity, positive likelihood ratio, negative likelihood ratio, diagnostic odds ratio and SROC of WDT and self-reported BBT; ③ secondary outcome 1: the accuracy of WDT based on distal BT and proximal BT in detecting fertility window; ④ secondary outcome 2: the accuracy of WDT with BT-based algorithm and multi-parameter-based algorithm in detecting fertility window; ⑤ secondary outcome 3: the accuracy of WDT with algorithm of random forest model, linear mixed model and other AI models in detecting fertility window; ⑥ secondary outcome 4: the accuracy of different modality of WDT (bands and rings) in detecting fertility window; ⑦ secondary outcome 5: the accuracy of different interval of WDT in detecting fertility window. Characteristics of the included studies using conventional methods for detecting the menstrual cycle and fertility window Abbreviations: BBT basal body temperature, BMI body mass index, CBFM Clearblue Fertility Monitor, E1G oestrone-3-glucuronide, E3G PdG pregnanediol, FSH follicle stimulating hormone, LH luteinizing hormone, PCOS polycystic ovary syndrome, NA not appliable, US United States, UK United Kingdom. ① primary outcome 1: the accuracy of WDT, self-reported BBT, electronic hormone testing system and calendar estimation in detecting fertility window; ② primary outcome 2: the specificity, sensitivity, positive likelihood ratio, negative likelihood ratio, diagnostic odds ratio and SROC of WDT and self-reported BBT. For the included studies, the diagnostic accuracy comparison included 13 studies on WDT 35 – 47 , 6 studies on self-reported BBT 37 , 44 , 46 , 48 – 50 , 8 studies on the electronic hormone testing system 51 – 58 , and 3 studies on the calendar method by self-calculation or calendar apps 59 – 61 . Among them, 5 WDT studies 35 , 36 , 38 , 44 , 46 and 2 self-reported BBT studies 44 , 46 reported true/ false positive/negative accounts. Ovulation was confirmed using various standards: 12 studies used the urine LH test 36 – 38 , 41 , 44 – 47 , 53 , 59 , 60 , 62 , 2 studies were based on the urine pregnanediol (PdG) test 56 , 58 , 3 studies employed the urine multi-hormone test 40 , 54 , 55 , 3 studies used the ultrasound method 48 , 49 , 52 , 3 studies applied ultrasound with serum hormone levels 42 , 43 , 50 , one study referred to overnight vaginal temperature 35 , and 2 studies did not specify the standard approach 57 , 61 . Among the 13 studies on WDT, in the aspect of hardware design, 6 studies used wrist-wear devices (band form) 35 , 36 , 38 , 41 , 43 , 44 and 2 studies used finger-wear devices (ring form) 45 , 47 , where the rest studies applied in-ear devices 37 , lower abdomen band 39 , vaginal ring 40 , axillary sensor 42 , and chest sensor 46 , respectively. In the aspect of bio-signals, 10 cohorts collected distal or proximal body temperatures 35 , 37 , 39 – 47 , and 5 cohorts acquired multi-parameters, like BBT, HR, HRV, respiratory rate (RR), and skin perfusion (SP) 36 , 38 , 43 , 47 . In the aspect of algorithms for prediction, all WDTs were based on artificial intelligence (AI), where 2 studies applied random forest (RF) models 36 , 38 , 4 studies applied linear mixed models (LMM) 41 , 43 , 44 , 46 , and 7 studies applied other AI model methods 37 , 39 , 40 , 42 , 45 , 47 , 63 . In quality assessment, most studies included in our analysis presented a low risk of bias in patient selection and clinical measurements (Supplementary Fig. 1a, b ), except for one study that lacked a clear definition of the study population and enrollment 61 . In the aspect of reference standard, most of the studies presented an unclear risk of bias since these studies referred to urine hormone tests, which may detect varied percentages of LH surges with urinary LH kits of different manufacturers 64 , and 3 studies did not report the reference standard 39 , 57 , 61 . 6 studies presented a low risk of bias for they applied transvaginal ultrasonography, or in combination with serum hormone levels as the gold standard to determine ovulation 42 , 43 , 48 – 50 , 52 . One study was ranked as high risk since it referred to overnight vaginal temperature, which may not be reliable as a BBT method 35 . In terms of flow and timing, all studies presented an unclear risk of bias since not all the participants or cycles were included in the analysis. Publication bias was evaluated using six studies reporting the diagnostic odds ratio (DOR) of WDT 36 , 38 , 44 , 46 , 63 . The funnel plot exhibited an asymmetric distribution, suggesting potential publication bias (Supplementary Fig. 1c ). Moreover, the quality of the evidence on WDT in detecting ovulation was graded as ‘low’, which was further downgraded to ‘very low’ due to serious risk of bias, inconsistency, and imprecision. (Supplementary Table 1 ) . Sensitivity analysis showed that omitting Yu J.L.’s study (cohort 2) 43 reduced the heterogeneity in a multi-parameter group ( I 2  = 0%) and increased the pooled accuracy to 0.91 (95% CI: 0.90–0.92) (Supplementary Fig. 1d ), which may be due to this study enrolling participants with irregular cycles. For other outcomes, no study included or excluded impacted the overall result estimation; thus, we retained the results as they were. Currently, there are several methods commonly used for detecting ovulation and menstrual cycles, including calendar estimation, self-measured BBT, and electronic hormone testing systems. Therefore, we first evaluated the performance of WDT in detecting female fertility windows, in comparison with the traditional methods (Supplementary Tables 2 and 3 ). WDT demonstrated a pooled accuracy of 0.88 (95% CI: 0.86–0.90) for fertility window detection. Notably, it outperformed the self-reported BBT and calendar estimation, which showed a pooled accuracy of 0.75 (95% CI: 0.63–0.86) and 0.72 (95% CI: 0.63–0.80), respectively (Fig. 2a ). Furthermore, in the comparison between WDT and self-reported BBT method, WDT yield pooled sensitivity and specificity of 0.79 (95% CI: 0.70–0.87) and 0.80 pooled (95% CI: 0.60–1.00), respectively, alongside a pooled positive likelihood ratio (PLR) of 5.87 (95% CI: 2.49–13.88), 0.25 negative likelihood ratio (NLR, 95% CI: 0.13-0.51), and 23.39 diagnostic odds ratio (DOR, 95% CI: 3.45-158.71), with the area under the summary receiver operating characteristic (SROC) curve of 0.752 (Supplementary Fig. 2a–c and Supplementary Fig. 3 ). By contrast, BBT-based studies exhibited a lower pooled sensitivity of 0.45 (95% Cl 0.01–0.90) and a pooled specificity of 0.73 (95% CI: 0.59–0.88), with PLR of 1.45 (95% CI: 0.35–5.93), NLR of 0.69 (95% CI: 0.25–1.86), DOR of 2.20 (95% CI: 0.20–24.29), achieving only a SROC of 0.368 (Supplementary Fig. 2a–c and Supplementary Fig. 3 ). Similarly, the network ranking and surface under the cumulative ranking curve (SUCRA) also indicated that WDT was superior to the BBT method (Supplementary Fig. 2d, e ). Moreover, compared to the electronic hormone testing system, with a pooled accuracy of 0.88 (95% CI: 0.85–0.91), WDT presented a similarly pooled accuracy in detecting the fertility window. The data were further aligned with the network ranking and SUCRA analysis, where the electronic hormone testing displayed the highest detection performance, followed by WDT, self-reported BBT, and calendar methods (Fig. 2c ). Fig. 2 Comparison of wearable digital technology (WDT) and other methods in detecting the fertility window. a Forest plot and b Network plot of the pooled accuracy of WDT and other methods in detecting the fertility window. Edge thickness was proportional to the number of direct comparisons. Node sizes were proportional to the sample size. c Network meta-analysis ranking of the pooled accuracy of WDT and other methods for fertility window detection. Bars represented the ranking probability. Deeper blue represented a higher ranking. Among the bars in the same color, the length of the bar was proportional to its possibility in this ranking. Nodes represented the rank of surface under the cumulative ranking curve (SUCRA). A higher position of the node represented a higher ranking. Abbreviation: BBT basal body temperature. The figures were created by R software 4.1.0 (R Statistical Computing) using R package metafor and gemtc . a Forest plot and b Network plot of the pooled accuracy of WDT and other methods in detecting the fertility window. Edge thickness was proportional to the number of direct comparisons. Node sizes were proportional to the sample size. c Network meta-analysis ranking of the pooled accuracy of WDT and other methods for fertility window detection. Bars represented the ranking probability. Deeper blue represented a higher ranking. Among the bars in the same color, the length of the bar was proportional to its possibility in this ranking. Nodes represented the rank of surface under the cumulative ranking curve (SUCRA). A higher position of the node represented a higher ranking. Abbreviation: BBT basal body temperature. The figures were created by R software 4.1.0 (R Statistical Computing) using R package metafor and gemtc . Moreover, to identify potential heterogeneity of our results, we further performed subgroup analyses according to three categories, including 1) the menstrual regularity in participants (regular, irregular, mixed, or unspecified cycles), 2) health status of the participants (mixed populations including participants with comorbidities, or studies with unspecified health status), and 3) the ovulation detection method (ultrasound with serum hormone levels, serum LH alone, or overnight vaginal temperature). There was a significant difference in WDT diagnostic accuracy between menstrual regularity groups (Chi-square test, p  = 0.02), indicating that irregular cycles may diminish the performance of WDT in ovulation detection (Supplementary Fig. 4a ). In contrast, WDT exhibited stable and comparable performance across either different ovulation reference standards or participant health status (Chi-square test, p  = 0.13 and 0.74, respectively, Supplementary Fig. 4b, c ). Collectively, these findings suggest that although menstrual regularity remains a critical determinant of predictive reliability, WDT demonstrates consistent performance regardless of health background or ovulation validation method. Instead of capturing the exact date of ovulation, WDT generally estimates a time interval surrounding the ovulation date, with a different duration depending on device design and computational method. Our analysis showed that the detection performance of WDT for the precise date of ovulation was relatively low, presenting a pooled accuracy of 0.56 (95% CI: 0.00–1.00) for the specific ovulation day, and 0.61 (95% CI: 0.53–0.70) for ±1 day surrounding ovulation. Moreover, WDT exhibited comparable performance for detecting ±2 days with accuracy of 0.90 (95% CI: 0.88–0.93) and ± 3 days with accuracy of 0.88 (95% CI: 0.84–0.92) surrounding ovulation, whereas a slightly lower pooled accuracy of 0.85 (95% CI: 0.80–0.91) was detected to detect ovulation 5 days in advance. Consistently, network ranking and SUCRA analysis revealed that the detection accuracy 2 to 3 days surrounding ovulation exceeded detecting ±1 day or the exact day of ovulation. Moreover, the interval from 5 days in advance to ovulation day was also suboptimal for WDT detection (Supplementary Fig. 5 and Supplementary Table 4 ). Wearable devices are equipped with multiple biosensors to capture diverse physiological signals. We next examined the contribution of different physiological inputs in determining ovulation detection (Supplementary Table 5 ) . Body temperature rhythmically fluctuates due to the thermoregulatory effects of progesterone throughout the menstrual cycle 22 . We found that WDT used distal BBT as a major index for detection, including wrist skin temperature (WST) and finger skin temperature (FST), showed a pooled accuracy of 0.88 (95% CI: 0.86–0.91). Similarly, studies applied proximal BBT for detection, including BBT on sites of acoustic meatus, low abdomen, vagina, axillar, and chest, achieved a pooled accuracy of 0.87 (95% CI: 0.83–0.90) (Fig. 3a ). Consistently, network ranking and SUCRA analysis suggested both distal BBT and proximal BBT exhibited a comparable effect in fertility window detection (Fig. 3b ). Fig. 3 Comparison in the accuracy of the designs of wearable digital technology (WDT) in detecting the fertility window. a , b a Forest plot and b Network meta-analysis (NMA) ranking of the pooled diagnostic accuracy of WDT using different body temperature measurement in detecting the fertility window. c , d c Forest plot and d NMA ranking of the pooled diagnostic accuracy of WDT employing different parameters for fertility window detection. e , f e Forest plot and f NMA ranking of pooled accuracy of WDT based on different algorithms in detecting the fertility window. g , h g Forest plot and h NMA ranking of pooled accuracy of digital bands and rings in fertility window detection. In the NMA ranking plots, bars represented the ranking probability. Deeper blue represented a higher ranking. Among the bars in the same color, the length of the bar was proportional to its possibility in this ranking. Nodes represented the rank of surface under the cumulative ranking curve (SUCRA). A higher position of the node represented a higher ranking. Abbreviation: artificial intelligence (AI), body temperature (BT). The figures were created by R software 4.1.0 (R Statistical Computing) using R package metafor and gemtc . a , b a Forest plot and b Network meta-analysis (NMA) ranking of the pooled diagnostic accuracy of WDT using different body temperature measurement in detecting the fertility window. c , d c Forest plot and d NMA ranking of the pooled diagnostic accuracy of WDT employing different parameters for fertility window detection. e , f e Forest plot and f NMA ranking of pooled accuracy of WDT based on different algorithms in detecting the fertility window. g , h g Forest plot and h NMA ranking of pooled accuracy of digital bands and rings in fertility window detection. In the NMA ranking plots, bars represented the ranking probability. Deeper blue represented a higher ranking. Among the bars in the same color, the length of the bar was proportional to its possibility in this ranking. Nodes represented the rank of surface under the cumulative ranking curve (SUCRA). A higher position of the node represented a higher ranking. Abbreviation: artificial intelligence (AI), body temperature (BT). The figures were created by R software 4.1.0 (R Statistical Computing) using R package metafor and gemtc . Apart from BT, WDTs enable simultaneous measurement of multiple physical signals, including WST, HR, HRV, RR, and SP. Compared with the single-BBT method with the pooled accuracy of 0.87 (95% CI: 0.85-0.88), multi-parameter-based detection yielded an improved pooled accuracy of 0.89 (95% CI: 0.84–0.95) (Fig. 3c ). Consistently, network ranking and SUCRA ranked the multi-parameter design outperforming the single BBT design in the detection accuracy (Fig. 3d ), indicating the clinical value of multidimensional bio-signal tracking for monitoring the fertility window. To process multidimensional datasets, WDT incorporates sophisticated digital platforms with advanced algorithms, including conventional approaches, such as linear mixed models (LMM), to cutting-edge AI techniques like random forests (RF). Therefore, we subsequently interrogate the optimal computational method (Supplementary Table 5 ) . Our analysis suggested WDT based on LMM achieved a pooled accuracy of 0.86 (95% CI: 0.84-0.88). Furthermore, AI-based models, such as support vector machine 36 , 39 , 43 , 45 and hidden Markov model 37 , exhibited positive effects to improve the pooled accuracy (0.88, 95% CI: 0.84–0.91). Of note, WDT with RF model achieved the highest accuracy of 0.91 (95% CI: 0.90–0.92) (Fig. 3e ). Consistently, our network ranking and SUCRA analysis showed the order of probability was RF model > other AI model > LMM (Fig. 3f ). Taken together, our data suggested that applying WDTs integrating with advanced AI algorithms may enhance the precision and robustness of fertility window detection. WDTs are engineered in diverse forms, including wrist-worn bands, finger rings, and skin-adhered patches. Therefore, we subsequently evaluated how hardware design influences the ability of WDTs in fertility window monitoring. Among existing studies, only band and ring-type devices were utilized, with no alternative WDT forms reported. Our comparative analysis revealed band- and ring-type devices showed equivalent detecting performance, presenting pooled accuracy of 0.88 (95% CI: 0.86–0.91) and 0.88 (95% CI: 0.77–0.99), respectively (Fig. 3g ). Interestingly, network ranking and SUCRA analysis positioned the ring-type devices as slightly superior to the band in detective performance (Fig. 3h ), which is possibly due to their stable positioning and consistent skin contact.

Our study presents the first systematic review and Bayesian network meta-analysis to evaluate the effects of WDT on the detection of the menstrual cycle and fertility window in women. Despite current studies exhibiting considerable heterogeneity in physiological parameters acquisition, algorithm method, and device form, WDTs demonstrated reliable accuracy to monitor the fertility window compared to self-reported BBT and calendar-calculation methods. Moreover, our findings suggest that compact wearable formats, integrated with multiparameter monitoring and advanced AI-based algorithms, may further enhance predictive accuracy. The menstrual cycle and fertility window constitute a core physiological axis shaping women’s lifelong health. Real-time and personalized fertility window tracking may not only support conception and contraception but also serve as a potential early signal for detecting reproductive disorders. However, traditional methods for ovulation prediction and menstrual tracking are limited by their accuracy, accessibility, and usability. For instance, though transvaginal ultrasound is the gold standard for ovulation tracking, it requires clinical visit 65 . The calendar estimation method relies on subjective self-reported cycle history and thus is vulnerable to recall bias 66 , 67 . The urinary hormone test typically needs repetitive testing, limiting its cost-effectiveness and user adherence. Our results highlighted that WDTs emerge as promising tools for detecting fertility windows in a personalized and real-time manner. WDTs outperformances over ordinary calendar methods with noticeably higher sensitivity (0.79 vs. 0.45) and SROC (0.75 vs. 0.37) and present comparable accuracy with hormone-based methods. In terms of detection intervals, WDT generally presented a high predictive value when identifying a fertility window within ± 3 days surrounding ovulation. However, given the viability of sperm (up to 5 days) and oocytes (about 24 h after ovulation) 40 , 68 , the current predictive interval of WDT cannot perfectly align with the biological fertility window. Further WDT development should aim to optimize the prediction time window from 5 days before to 24 h post-ovulation, which is essential to minimize the unintended conception and enhance pregnancy planning in clinical practice. Alternations in BBT during the menstrual cycle are a well-established physiological phenomenon driven by the thermogenic effects of progesterone during the luteal phase. In our study, BBT still serves as the core parameter detected by WDT to provide the most reliable signal for improving predictive accuracy than the self-reported BBT method (0.87 (95% CI: 0.85–0.88) vs. 0.75 (95% CI: 0.63–0.86)). Although WDTs typically measure skin temperature, which may be affected by environmental factors or physical activity 69 , several studies have shown that continuously monitored skin temperature can linearly reflect core body temperature 70 . Of note, all the current WDTs employed the nocturnal BBT for the prediction, which can largely minimize environmental disturbance. Interestingly, we found FST and WST presented similarly pooled detection accuracy (0.88 (95% CI: 0.77–0.99) vs. 0.88 (95% CI: 0.86–0.91)), whereas FST presented higher network and SUCRA ranking, which may be attributed to the improved comfort and stability of ring-type devices for continuous BBT measurement. Except for BBT, growing evidence suggests the alternations RR and HR, and decreases in HRV after ovulation 71 , 72 . These physiological shifts are mainly driven by either vague nerve-mediated or hormone-induced sympathetic changes and increased metabolic rate. Our data found inclusion of those parameters into algorithms could slightly improve the accuracy of WDT in detecting fertility windows. As shown, WDT integrating BBT and other physical parameters (solo or in combination of HR, HRV, RR, SP) outperformed those only based on BBT in detecting accuracy (0.89 (95% CI: 0.84–0.95) vs. 0.87 (95% CI: 0.85–0.89), which was confirmed by higher ranking of WDT with multiple parameters in network and SUCRA ranking. Taken together, although these variations are subtler than BBT changes, they may still enhance fertility prediction accuracy when integrated into multi-parameter algorithms. It further underscores the advantages of WDT in fertility window management, as they can trace a spectrum of physiological signals. Interestingly, it has been reported that a WDT-equipped chemical nano-biosensor is available for testing estradiol levels in body fluid 73 . This advancement suggests a promising future in which WDT may integrate microfluidics technology for detecting trace-level hormones, thereby offering more precise fertility tracking and personalized care. Our study has several limitations. First, heterogeneity in the definition of ovulation varied across enrolled studies. Although the subgroup analysis on the accuracy of WDT with different reference standards showed comparable results, most studies estimated ovulation using serum LH levels and lacked validations of the gold standard method of ultrasound, which may impair the reliability and comparability. Second, although no significant effects were observed for ovulation detection standards, methodological inconsistencies and small sample sizes across studies may limit the generalizability of our findings. In our results, menstrual regularity was a key determinant of heterogeneity, and WDT was reported to exhibit lower detection accuracy in the irregular menstruation populations, implying that irregular cycles remain a challenge for WDT-based ovulation prediction. Further studies with rigorous design are needed to validate the capability of WDT in precisely predicting ovulation, especially in those with irregular cycles. Moreover, some cohorts we included in this analysis containing both healthy women and patients with reproductive disorders, such as Polycystic ovary syndrome (PCOS) 35 , 40 , hypothyroidism 35 , and infertility 42 , which may lead to a confounding factor. Further clinical study is urgent to assess the performance of WDT in women with irregular menstrual cycles. Nevertheless, there was comparable diagnostic accuracy of WDT between the two groups in our subgroup analysis, the impact of comorbidities on the diagnostic performance of WDT should be further investigated in future studies to provide personalized ovulation prediction. Furthermore, the absence of participant demographics in several studies limited our analysis to evaluate the impact of other potential confounders (e.g., age and body mass index (BMI) 35 , 37 , 39 , 40 , 42 , 45 . Third, although the probabilities and cumulative ranking suggested relatively optimal WDT designs, physiological parameters, and algorithmic approaches, these findings should be interpreted with caution. Most of pairwise comparisons revealed statistically significant differences only when compared to the reference standard, whereas direct comparisons between individual designs were largely non-significant. Therefore, the relative performance of different WDT configurations requires more rigorous head-to-head evaluation in future studies. Fourth, given the small number of included studies, the assumptions required for NMA may be unstable, leading to limited robustness of the findings. Moreover, publication bias and studies with a risk of bias may compromise confidence in the evidence, which may skew the overall distribution of observed values and overstate the actual diagnostic performance of WDT in the clinical practice. Together with the very low Grading of Recommendations, Assessment, Development and Evaluations (GRADE) rating of the certainty of the evidence, the accuracy of WDTs in menstrual tracking and fertility management should be interpreted with prudence. Importantly, large-scale cohorts with robust designs are warranted to rigorously validate the results. Importantly, the WDT detection interval with high accuracy does not optimally cover the biological fertility window. Future iterations should prioritize matching the prediction with the physiological fertility window, thereby improving clinical utility for both pregnancy planning and conception avoidance. In summary, our systematic review and NMA demonstrate that WDTs may serve as promising tools for detecting ovulation within 3 days before and after ovulation. The integration of BBT with additional physiological signals into advanced AI-based, multi-parameter algorithms may further enhance detection accuracy. With the development of digital health and biosensing technologies, WDTs hold the potential as next-generation, non-invasive approaches for personalized care in women’s reproductive health.

Pregnancy management remains a global health priority, reforming the trajectory of individuals and families, as well as a human population issue. A well-planned pregnancy with appropriate preconception preparation 1 (e.g., folic acid supplementation 2 , preconception counseling 3 ) and evidence-based pregnancy care 4 significantly reduces the risk of gestational complications. However, the annual number of unintended pregnancies increased from 80 million in 1994 to 121 million in 2019. Among them, over 61% of unintended pregnancies end in abortion 5 , 6 , which is associated with long-term infertility risks in females 7 . In addition, unplanned pregnancies are significantly linked with adverse maternal and neonatal outcomes 8 , including low birth weight 9 , prematurity 9 , and developmental disadvantages 10 , 11 . It also imposes a costly burden on the health care system 12 , 13 . In the United States, direct medical costs due to unplanned pregnancies are projected to be 5 to 12.6 billion 12 , 14 , whereas the prevention of unintended pregnancies will save over 5.6 billion annually 14 . Therefore, it is critical to develop effective strategies to promote the planned conception and pregnancy. The fertility window spans from 5 days before ovulation to 24 h after, based on the sperm and ovum viability 15 – 17 . Clinical studies showed that the probability of conception increases progressively during the fertile window and peaks on the day preceding ovulation 18 , 19 , and then rapidly declines thereafter 18 , 19 . A prospective longitudinal study suggested that avoiding unprotected intercourse during the fertile period only leads to 0.4 unintended pregnancies occurring per 100 women 20 . Therefore, accurately monitoring the timing of ovulation is essential for either successful fertilization or contraception. However, the fertile window is highly dynamic and personalized. Although ovulation typically occurs between days 10 and 17 of the menstrual cycle, only 30% of women ovulate entirely within this period 15 , underscoring the need for precision strategies to personalize ovulation detection. Interestingly, the fertility window is intricately fine-tuned by cyclic alternations of estradiol, progesterone, luteinizing hormone (LH), and follicle-stimulating hormone (FSH) 21 . These hormonal dynamics contribute to biphasic rhythms and physiological shifts in females. For instance, basal body temperature (BBT) elevates by 0.3 °C to 0.7 °C in the post-ovulatory luteal phase owing to the progesterone-derived thermoregulatory effects 22 . Hormonal rhythms also increase resting heart rate (HR) by approximately 2.5% and decrease approximately 2.5% vagally mediated heart rate variability (HRV) in the luteal phase 23 , 24 . Meanwhile, lifestyle factors, like night shift 25 , psychological pressure 26 , and intensive exercise 27 , may disrupt endogenous hormone homeostasis and thereby affect the timing and regularity of the fertility window. Those nuanced but substantial fluctuations during the menstrual cycle raise the possibility of real-time tracking and analysis of physiological trajectories for fertility window detection 28 , 29 . Wearable digital technology (WDT) integrates wearable hardware devices with digital components, including data processing algorithms and mobile apps for health assessment, alarming, and clinical care 30 . They typically incorporate one or multiple biological sensors to monitor real-time physiological signals, such as HR, body temperature (BT), blood pressure (BP), oxygen saturation (SpO 2 ), and track physical activities and sleep patterns 31 . With its advancement for continuous non-invasive monitoring of physiological parameters, WDT has been widely adopted in medical practice, including for detecting epilepsy during seizures 32 , asymptomatic cardiovascular diseases 33 , and rapid identification of SARS-CoV-2 infection 34 . However, the current application of WDT in female health is still in the early stages, especially for the menstrual cycle and fertility window. In this systematic review and network meta-analysis (NMA), we comprehensively synthesize current studies applying WDT to monitor the fertility window and menstrual cycles. Our primary outcome is to evaluate their accuracy, sensitivity, and specificity. We further employed Bayes network analysis to compare the performance of various device configurations, physiological parameters, algorithm methods, and detection intervals. Our study highlights the emerging clinical value of WDT in women’s reproductive health, providing a non-invasive and data-driven approach to personalized pregnancy plans and birth control.

Supplementary Information Supplementary Information