///2009 Abstract Details
2009 Abstract Details2018-05-01T17:45:11+00:00

Development and Validation of a Model for the Progress of labor

Abstract Number: 12
Abstract Type: Original Research

Pamela Flood MD, MA1 ; Richard M. Smiley PhD, MD2; Steven L. Shafer MD3

Graphic models of labor were pioneered by Friedman in the 1950s (1). He modeled labor with a sigmoid curve that has been simplified to 2 lines representing different rates (slopes) for early ("latent") labor and later ("active") labor. Friedmans models have been used to define and even diagnose abnormal labor although several authors have raised doubts regarding applicability to modern obstetrical practice (2). Mixed-effects modeling permits robust separation of individual variability from randomly distributed measurement noise. We developed a population model of labor progress using PLT Tools (P< Company, San Francisco, CA) for NONMEM (Globomax, Hanover, MD) to describe the progress of labor in 500 nulliparous parturients who had normal spontaneous vaginal deliveries.

The study was approved by the Columbia University IRB. Requirement for informed consent was waived. All cervical exams and time of exam were recorded. The first diagnosis of 10cm dilation was defined as time 0. The times of cervical examinations prior to full dilation were converted to hours before full dilation. We attempted to fit the data to two linear equations, tethered with inflection point (CD=10-Mactive x Time), for the active phase and (CD=Inflection - Mlatent x (Time - (10-Inflection)/Mactive)) for latent labor (figure, dotted line). (CD = cervical dilation in cm, M = slope of the latent or active phase of labor in cm/hr, Inflection = point (in cm dilation) where the active phase of labor begins). The parameters of the bilinear relationship were highly sensitive to starting estimates, making it unsuitable for statistical comparison.

The data suggested a biexponential model (CD = C1(-L1 x TIME) + C2(-L2 x TIME)) describing labor progression with a continuous function. C1 = cervical dilation (cm) associated with the larger exponent (active labor), C2 = cervical dilation (cm) associated with smaller exponent (latent labor), and C1+C2 = 10. The exponent L1 (1/hours) quantifies the rapid (active) phase of cervical dilation and the exponent L2 defines the slower (latent) phase. The -2 Log Likelihood of the biexponential model was 325 points less than the bilinear model, a huge improvement in model fit. The median absolute residual error improved from 0.73 to 0.65 cm. The derived coefficients (confidence intervals) were: C1=3.4 (2.8-4.1) cm, L1= 0.622 (0.49-0.77)/h, L2 = 0.089 (0.078-0.099) /h. Confidence intervals were calculated from 1000 bootstrap replications. These values can be converted to rate constants and suggest that the rate of change in cervical dilation accelerates continuously during labor.

In summary, our biexponential model of labor progress describes the data better than a bilinear model, and has the advantage of being stable with regard to initial estimates. This model may prove useful for testing the influence of patient covariates and treatment effects on the time course of labor.

SOAP 2009