A multistate competing risks framework for preconception prediction of pregnancy outcomes

Area of Science:

• Reproductive Medicine
• Biostatistics
• Maternal-Fetal Medicine

Background:

• Preconception pregnancy risk profiles are crucial for early obstetric intervention to improve conception and live birth rates.
• Constructing these profiles is statistically complex due to multinomial outcomes, competing risks in conception and gestation, and different time scales.
• Current risk tools often predict single adverse outcomes post-conception, limiting early intervention opportunities.

Purpose of the Study:

• To develop and validate a novel preconception pregnancy risk assessment tool.
• To reframe pregnancy attempts as a multistate model for improved prediction.
• To enable earlier intervention for better pregnancy outcomes.

Main Methods:

• A multistate model was developed, treating pregnancy attempts as two nested multinomial prediction tasks: conception and subsequent pregnancy outcome.
• The model accounts for missing outcome data at multiple stages.
• An inverse-probability-weighted Hypervolume Under the Manifold statistic was used for validating multivariate risk scores.

Main Results:

• The developed risk profiles demonstrated meaningful discrimination between the four key pregnancy attempt outcomes in the EAGeR trial population.
• These profiles significantly outperformed random chance in classification accuracy.
• The framework successfully constructed and validated a preconception pregnancy risk assessment tool.

Conclusions:

• The proposed prediction framework expands pregnancy risk assessment by considering diverse outcomes and enabling earlier, preconception predictions.
• This provides obstetricians and patients with enhanced information for guiding pregnancy attempts.
• The study highlights the utility of a multistate competing risks framework for preconception risk assessment.