Predictive models of pregnancy based on data from a preconception cohort study.

Area of Science:

• Reproductive epidemiology
• Machine learning in healthcare
• Predictive modeling for conception

Background:

• Previous research focused on infertility risk factors, with limited predictive models (AUC: 59-64%).
• A North American preconception cohort study provided data for developing new predictive models.

Purpose of the Study:

• To derive adequate models predicting conception probability in couples actively trying to conceive.
• To improve the discrimination of predictive models for conception.

Main Methods:

• Utilized data from 4133 female participants (aged 21-45) in a preconception cohort study (2013-2019).
• Collected baseline and follow-up questionnaire data on sociodemographic, lifestyle, diet, medical history, and male partner characteristics (163 predictors).
• Employed machine learning algorithms (logistic regression, SVM, neural networks, gradient boosted trees) and Cox models to predict pregnancy probability.

Main Results:

• Models achieved an area under the receiver operating characteristic curve (AUC) of approximately 70% for predicting pregnancy within 12 cycles.
• Positive predictors for conception included prior breastfeeding and multivitamin/folic acid use.
• Negative predictors included female age, female BMI, and history of infertility.

Conclusions:

• Machine learning models can effectively predict conception probability using extensive epidemiologic data.
• Model performance (~70% AUC) exceeds that of earlier predictive models for conception.
• Limitations include reliance on self-reported data and lack of external validation.