Cribado de preeclampsia en el primer trimestre del embarazocomparativa de tres modelos de cribado

Laburpena

Objectives: Many algorithms have been published for pre-eclampsia (PE) screening during the first trimester of pregnancy; however, few have been validated for populations other than those for which they were developed. The prevalence of this pathology as well as the specific characteristics of the sample itself may interfere with the external validation of the algorithms. The main objective of this doctoral thesis is to compare the usefulness of three prediction models for PE during the first trimester of pregnancy implemented on our population. The specific objectives are to determine the predictive capability of the analysed prediction models for PE on each type of PE and to create our own screening algorithm adapted to our population's features. Material and methods: All pregnant women who attended the University Hospital Virgen de la Arrixaca from June 2011 to June 2015 to undergo an ultrasound screening for chromosomal abnormalities and who met all the inclusion criteria and none of the exclusion criteria were recruited for the data collection. The variables observed were mean blood pressure, pregnancy-associated plasma protein-A (PAPP-A), uterine artery pulsatility index (UtA-PI) and maternal medical history. The algorithms published by Poon et al. (Ultrasound Obstet Gynecol 2009), Akolekar et al. (Fetal Diagn Ther 2013) and Scazzocchio et al. (Am J Obstet Gynecol 2013) were implemented, taking into account the detection rates obtained and comparing their areas under the ROC curve. Furthermore, a new screening model for early and total PE was developed. Results: 16.521 patients were recruited for the study, 13.241 of whom met all the inclusion criteria and could provide all the needed data. 185 patients developed PE. The area under the curve for early PE in each one of the algorithms were 0.82, 0.85 and 0.83 for Poon's, Akolekar's and Scazzocchio's respectively. No statistically significant differences were found among them. Detection rates (DR) observed for 10% of the false positive rate (FPR) were 41%, 70% and 51% for Poon's, Akolekar's and Scazzocchio's. Detection rates for late PE observed for 10% of the FPR were 31%, 52% and 36% for Poon's, Akolekar's and Scazzocchio's. Our own model reached an area under the curve after Boostrap validation of 0.83 for early PE, with 64% of DR for 10% of FPR, and an area under the curve for total PE of 0.76 and 47% of DR for 10% of FPR. Discussion: The three analysed models work in a similar way and can be applied interchangeably on the sample. The external validation of the models on our population achieved screening rates lower than those published in the original published works. The algorithm developed in this doctoral thesis is, thus, comparable to the analysed models.