Time-Varying Exposures and Miscarriage: A Comparison of Statistical Models Through Simulation

Area of Science:

• Epidemiology
• Reproductive Health
• Biostatistics

Background:

• Early pregnancy involves dynamic maternal exposures and distinct developmental stages.
• Time-varying exposures during gestation pose challenges for epidemiological risk assessment.
• Alcohol consumption is a common example of a time-varying exposure with potential adverse pregnancy outcomes.

Purpose of the Study:

• To evaluate different statistical modeling approaches for assessing risks associated with time-varying exposures in early pregnancy.
• To compare the bias, precision, and statistical power of various models using simulated data.
• To highlight the importance of considering exposure timing in epidemiological research.

Main Methods:

• Simulation study using data from the Right From the Start cohort (over 5,000 women).
• Modeled alcohol consumption patterns and miscarriage likelihood by gestational week.
• Compared five different statistical modeling approaches for effect estimation.

Main Results:

• Model accuracy and precision depend on the alignment between model assumptions and the true exposure-outcome relationship.
• Approaches incorporating exposure patterns demonstrated reduced bias and increased statistical power compared to simpler models.
• The timing and duration of exposure significantly influenced the accuracy of risk estimates.

Conclusions:

• Careful definition of exposure timing is critical for uncovering causal relationships in early pregnancy.
• Statistical models that account for time-varying exposure patterns are superior for assessing pregnancy risks.
• Accurate risk assessment requires aligning analytical methods with the temporal dynamics of exposure and development.