Use of linear mixed models to detect irregular fetal growth

Ultrasonographic fetal biometry data can be used to estimate fetal weights repeatedly over time. Different types of reference bands have been presented to judge such data. But relevant aspects of an individual growth curve are not taken in to account. How can individual growth curves be used efficiently to detect irregular fetal growth? Linear mixed model analysis of a sample of 167 uncomplicated pregnancies with 304 ultrasonic measurements revealed that a random intercept model is sufficient and heterogeneity in slopes is negligible. Hence irregularity of a sequence of weights can be captured by three quantities' residual level, trend and mean squared error' for which reference ranges are derived from the fitted model. After standardization they are independently normal and chi-square distributed in the reference population. Regularity of fetal growth may then be reported in terms of level and trend either in the form of Z-scores or as percentile indices. A sample of 97 pregnancies diagnosed as cases of chronic placental insufficiency postnatally is analysed to validate the ability to detected abnormal growth. Only the mean level and slope revealed to be able to detect IUGR. MSE was not discriminative. The sensitivity of the method depends on the number and placement of observation times, which opens the possibility of a model-based cost-effectiveness analysis. The derived scores make efficient use of available information and are easily implementable into any computer based monitoring device.