Tracking of menstrual cycles and prediction of the fertile window via measurements of basal body temperature and heart rate as well as machine-learning algorithms

Abstract

Background: Fertility awareness and menses prediction are important for improving fecundability and health management. Previous studies have used physiological parameters, such as basal body temperature (BBT) and heart rate (HR), to predict the fertile window and menses. However, their accuracy is far from satisfactory. Additionally, few researchers have examined irregular menstruators. Thus, we aimed to develop fertile window and menstruation prediction algorithms for both regular and irregular menstruators.

Methods: This was a prospective observational cohort study conducted at the International Peace Maternity and Child Health Hospital in Shanghai, China. Participants were recruited from August 2020 to November 2020 and followed up for at least four menstrual cycles. Participants used an ear thermometer to assess BBT and wore the Huawei Band 5 to record HR. Ovarian ultrasound and serum hormone levels were used to determine the ovulation day. Menstruation was self-reported by women. We used linear mixed models to assess changes in physiological parameters and developed probability function estimation models to predict the fertile window and menses with machine learning.

Results: We included data from 305 and 77 qualified cycles with confirmed ovulations from 89 regular menstruators and 25 irregular menstruators, respectively. For regular menstruators, BBT and HR were significantly higher during fertile phase than follicular phase and peaked in the luteal phase (all P < 0.001). The physiological parameters of irregular menstruators followed a similar trend. Based on BBT and HR, we developed algorithms that predicted the fertile window with an accuracy of 87.46%, sensitivity of 69.30%, specificity of 92.00%, and AUC of 0.8993 and menses with an accuracy of 89.60%, sensitivity of 70.70%, and specificity of 94.30%, and AUC of 0.7849 among regular menstruators. For irregular menstruators, the accuracy, sensitivity, specificity and AUC were 72.51%, 21.00%, 82.90%, and 0.5808 respectively, for fertile window prediction and 75.90%, 36.30%, 84.40%, and 0.6759 for menses prediction.

Conclusions: By combining BBT and HR recorded by the Huawei Band 5, our algorithms achieved relatively ideal performance for predicting the fertile window and menses among regular menstruators. For irregular menstruators, the algorithms showed potential feasibility but still need further investigation.

Trial registration: ChiCTR2000036556. Registered 24 August 2020.

Keywords: Basal body temperature; Fertile window; Heart rate; Machine learning; Menstrual cycle; Wearable device.

Fig. 1

Flow chart of the study population

Fig. 2

Physiological parameters of the regular group in the different phases of the menstrual cycle. Changes in BBT (A), HR (B), SDNN (C) and the LF/HF ratio (D) during the menstrual cycle of regular menstruators are depicted. The horizontal line represents the medians, boxes represent the values between 25–75%, and lines represent the values between 5–95%. BBT: basal body temperature; HR: heart rate; LF/HF: low frequency /high frequency ratio; SDNN: standard deviation of normal-to-normal intervals

Fig. 3

Prediction performance of different models. ROC curves of fertile window (A) and menstruation (B) prediction models based on BBT and HR for regular and irregular groups. AUC: area under the receiver-operating characteristic curve; BBT: basal body temperature; HR: heart rate; ROC: receiver operating characteristic