Predicting clinical pregnancy using clinical features and machine learning algorithms in in vitro fertilization

Area of Science:

• Reproductive medicine
• Biostatistics
• Artificial intelligence in healthcare

Background:

• Assisted reproductive technology (ART) and in vitro fertilization (IVF) cycles are increasingly utilized for infertility treatment.
• Identifying factors influencing successful pregnancy in IVF is crucial for optimizing patient outcomes.
• Machine learning (ML) offers advanced analytical capabilities for predicting complex biological outcomes.

Purpose of the Study:

• To develop and evaluate ML-based prediction models for clinical pregnancy in IVF cycles.
• To identify and rank key variables affecting IVF success using ML algorithms.
• To explore the impact of specific treatment variables on pregnancy outcomes.

Main Methods:

• Utilized a dataset of 24,730 IVF and intracytoplasmic sperm injection cycles with clinical pregnancy outcomes.
• Applied ML algorithms, including random forest and logistic regression, to construct prediction models.
• Employed partial dependence plots to analyze the influence of individual variables on pregnancy prediction.

Main Results:

• The random forest algorithm demonstrated superior performance compared to logistic regression in predicting clinical pregnancy.
• Ovarian stimulation protocols were identified as the most critical factor influencing pregnancy outcomes.
• Long and ultra-long stimulation protocols positively correlated with clinical pregnancy, while female age and infertility duration showed negative associations.

Conclusions:

• ML models, particularly random forest, effectively predict clinical pregnancy in ART cycles by ranking influential variables.
• Understanding the impact of factors like ovarian stimulation, embryo transfer, age, and infertility duration can aid clinicians in patient management.
• This study provides valuable insights for optimizing IVF treatment strategies and improving patient counseling.