OA Epidemiology

Prediction Models in Obstetrics: A Systematic Review of Methodology and Reporting

Proceedings of the 2013 annual meeting of the Netherlands Epidemiology Society

Volume 1 Issue S1 Abstract 20

 

C.E. Kleinrouweler, Department of Obstetrics and Gynaecology, Academic Medical Center, Amsterdam
E. Schuit, Julius Center for Health Sciences and Primary Care, University Medical Center Utrecht
G.S. Collins, Centre for Statistics in Medicine, University of Oxford
B.W.J. Mol, Department of Obstetrics and Gynaecology, Academic Medical Center, Amsterdam
K.G.M. Moons, Julius Center for Health Sciences and Primary Care, University Medical Center Utrecht

Background
Medicine, including obstetrics, is increasingly focused on risk-based or personalized medicine. Prediction models may aid in more individualized  medicine and as such many have been developed and published. We systematically reviewed the literature for prediction models in obstetrics and assessed their methodological quality, reporting and clinical applicability.

Methods
PubMed was searched for prediction models in obstetrics published between 1 January 2003 and 7 July 2012. Studies with the primary aim to develop a prediction model with three or more predictors were included. Two reviewers independently identified eligible studies and three  reviewers  assessed items relating to methodology and reporting from full text papers. Assessment was based on previously published checklists for methodology and reporting in prediction modeling research.

Results
The search retrieved 6602 papers describing the development of 126 prediction models. The most common outcomes to be predicted were hypertensive disorders, preterm birth, successful induction of labour, vaginal birth after caesarean section and gestational diabetes. Based on the assessment of 35 papers evaluated so far, we found that the majority of papers did not describe details and handling of missing values, many used univariable pre-selection of candidate predictors and clinical applicability of the model was rarely discussed. Methodological quality of all 126 papers will be presented in June.

Conclusions
Although all models were developed with the aim to allow for risk-based or personalized medicine, many suffer from methodological deficiencies and inadequate reporting. Furthermore, how the prediction models should guide decisions or treatment in clinical practice is rarely described.

Published: 06 Jun, 2013

 
Licensee OA Publishing London 2013. Creative Commons Attribution License (CC-BY)