A new population-based term prediction model vs. two traditional sample-based models: validation on 9046 ultrasound examinations

Abstract

Objectives: To compare results of predictions of date of delivery from a new population-based model with those from two traditional regression models.

Methods: We included 9046 fetal biparietal diameter (BPD) measurements and 8776 femur length (FL) measurements from the routine ultrasound examinations at Stavanger University Hospital between 2001 and 2007. The prediction models to be validated were applied to the data, and the resulting predictions were compared with the actual time of the subsequent deliveries. The primary measure was the median bias (the difference between the true and the predicted date of delivery), calculated for each method, for the study population as a whole and for three subgroups of BPD/FL measurements. We also assessed the proportion of births within ± 14 days of the predicted day, and rates of preterm and post-term deliveries, which were regarded as secondary measures.

Results: For the population-based model, the median bias was -0.15 days (95% confidence interval (CI), -0.43 to 0.12) for the BPD-based, and -0.48 days (95% CI, -0.86 to -0.46) for the FL-based predictions, and both biases were stable over the inclusion ranges. The biases of the traditional regression models varied, depending on the fetal size at the time of the examination; the extremes were -3.2 and + 4.5 days for the BPD-based, and -1.0 and + 5.0 days for the FL-based predictions.

Conclusions: The overall biases, as well as the biases for the subgroups, were all smaller with the population-based model than with the traditional regression models, which exhibited substantial biases in some BPD and FL subcategories. For the population-based model, the FL-based predictions were in accordance with the BPD-based predictions.