Assessing human fertility using several markers of ovulation

Area of Science:

• Biostatistics
• Reproductive Epidemiology
• Statistical Modeling

Background:

• Accurate modeling of human fertility requires precise timing of intercourse relative to ovulation.
• Measurement error in ovulation markers can lead to biased fecundability estimates and attenuated covariate effects.
• Existing methods often rely on single, error-prone ovulation markers.

Purpose of the Study:

• To develop a semi-parametric mixture model to account for measurement error in ovulation timing.
• To correct bias in estimates of day-specific fecundability using multiple independent ovulation markers.
• To jointly estimate error distributions, error-corrected fertility parameters, and couple-specific random effects.

Main Methods:

• Proposed a semi-parametric mixture model incorporating multiple independent ovulation markers.
• Assigned distinct non-parametric error distributions to each ovulation assessment method.
• Employed a Monte Carlo Expectation-Maximization (EM) algorithm for joint estimation.
• Applied the model to data from a North Carolina fertility study.

Main Results:

• Quantified the magnitude of measurement error in ovulation markers (urinary luteinizing hormone, ovarian hormone metabolites).
• Provided corrected estimates for day-specific probabilities of clinical pregnancy.
• Demonstrated the model's ability to correct bias in fecundability parameter estimation.

Conclusions:

• The proposed semi-parametric mixture model effectively addresses measurement error in ovulation timing.
• Utilizing multiple ovulation markers improves the accuracy of fertility parameter estimation.
• The methodology offers a robust approach for analyzing fertility data with imperfect ovulation markers.