Predicting pregnancy using large-scale data from a women's health tracking mobile application.

Area of Science:

• Women's Health
• Reproductive Health
• Machine Learning in Medicine

Background:

• Pregnancy prediction is a long-standing challenge in women's health.
• Mobile health tracking apps offer a new avenue for data collection beyond traditional studies.
• The efficacy of using mobile app data for pregnancy prediction remains largely unexplored.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting pregnancy probability using data from a women's health app.
• To assess the feasibility of leveraging large-scale mobile health data for fertility insights.
• To identify interpretable trends in fertility data from deep learning models.

Main Methods:

• Developed four predictive models: logistic regression and three Long Short-Term Memory (LSTM) networks.
• Utilized a dataset comprising 79 million logs from 65,276 women using the Clue app.
• Incorporated ground truth pregnancy test data for model training and evaluation.

Main Results:

• The models successfully stratified women based on their predicted probability of pregnancy.
• Women in the top 10% predicted probability had an 89% chance of pregnancy over 6 cycles, versus 27% for the bottom 10%.
• Extracted interpretable time trends from deep learning models, aligning with existing fertility research.

Conclusions:

• Women's health tracking app data holds significant potential for broad-scale pregnancy prediction.
• Machine learning models can effectively stratify fertility risk using real-world mobile health data.
• Further development is needed to fully realize the potential of mobile health data in reproductive health prediction.