Objective: To confirm the results from two previous evaluations of term prediction models, including two sample-based models and one population-based model, in a third population.
Methods: In a study population of 23,020 second-trimester ultrasound examinations, data were prospectively collected and registered over the period 1988-2009. Three different models for ultrasonically estimated date of delivery were applied to the measurements of fetal biparietal diameter (BPD) and two models were applied to the femur length (FL) measurements; the resulting term estimations were compared with the actual time of delivery. The difference between the actual and the predicted dates of delivery (the median bias) was calculated for each of the models, for three BPD/FL-measurement subgroups and for the study population as a whole.
Results: For the population-based model, the median bias was + 0.4 days for the BPD-based predictions and - 0.4 days for the FL-based predictions, and the biases were stable over the inclusion ranges. The biases of the two traditional models varied with the size of the fetus at examination; median biases were - 0.87 and + 2.2 days, respectively, with extremes - 4.2 and + 4.8 days for the BPD-based predictions, and the median bias was + 1.72 days with range - 0.8 to + 4.5 days for FL-based predictions. The disagreement between the two sample-based models was never less than 2 days for the BPD-based predictions.
Conclusion: This study confirms the results from previous studies; median biases were negligible with term predictions from the population-based model, while those from the traditional models varied substantially. The biases, which have clinical implications, seem inevitable with the sample-based models, which, even if overall biases were removed, will perform unsatisfactorily.
Copyright © 2012 ISUOG. Published by John Wiley & Sons, Ltd.