Abstract:
|
Modeling longitudinal cervical dilation curves remain a challenge due to the fact that women's onset of cervical dilation process is commonly unknown. Often researchers take the time to full dilation (i.e., 10 cm) as the benchmark time and then run time backwards. However, this method does not include women that do not get to 10cm. We view each woman as having unknown time-shift, which when adjusted for appropriately aligns her curve. Further, longitudinal cervical dilation measurements exhibit non-linear trends and between women heterogeneity. In this talk, we propose a penalized cubic B-spline mixed effects model with random shift parameters for analyzing longitudinal cervical dilation trajectories. Our method corrects for the time-shift in the data by introducing a random shift parameter. Our model is implemented via a Monte Carlo expectation maximization procedure where the random shift parameters are generated using an adaptive rejection algorithm in E - step. In M - step, variance components are estimated using ML while fixed and random effect parameters are given as best prediction. We demonstrate the proposed method using data from Consortium of Safe Labor study.
|
ASA Meetings Department
732 North Washington Street, Alexandria, VA 22314
(703) 684-1221 • meetings@amstat.org
Copyright © American Statistical Association.