Prediction of the fertile window and menstruation with a wearable device via machine-learning algorithms.

RESEARCH QUESTION: Can algorithms for prediction of the fertile window and menstruation be developed through machine learning based on women's physiological parameters collected using a wearable device (Huawei Band 6 Pro) from both regular and irregular menstruators?

DESIGN

This prospective observational cohort study was conducted at the Obstetrics and Gynaecology Hospital of Fudan University (China). Participants were recruited from November 2021 to September 2022. Each participant used a wearable device to record their wrist skin temperature (WST), heart rate, heart rate variability and respiratory rate for two menstrual cycles. Algorithms were developed to predict the fertile window and menstruation based on WST and heart rate data.

RESULTS

Data from 270 and 84 qualifying cycles with confirmed ovulations from 136 regular menstruators and 47 irregular menstruators were included in this study. For regular menstruators, the prediction algorithm based on WST and heart rate data for the fertile window had accuracy of 85.47%, sensitivity of 70.07%, specificity of 89.77%, and area under the curve (AUC) of 0.869. The algorithms for labelling the first day of the menstrual cycle and menstrual prediction 3 days in advance had accuracy of 83.6% and 75.0%, respectively. For irregular menstruators, accuracy, sensitivity, specificity and AUC were 79.85%, 42.79%, 87.28% and 0.763, respectively, for prediction of the fertile window. The accuracy of algorithms for labelling the first day of the menstrual cycle and menstrual prediction were 61.2% and 50.8%, respectively.

CONCLUSIONS

Based on WST and heart rate data from the wearable device, the algorithms demonstrated reliable performance in predicting the fertile window and menstruation among regular menstruators. These algorithms also showed potential for assisting irregular menstruators in managing their cycles and planning for conception.