J. Chang1,2, C. Lundquist1, J. S. Lutz1, B. S. Wessler1, D. M. Kent1, H. C. Jen1,2 1Tufts Medical Center,Predictive Analytics And Comparative Effectiveness Center,Boston, MA, USA 2Floating Hospital For Children,Pediatric Surgery,Boston, MA, USA
Introduction:
Advances in fetal medicine have led to the identification of risk factors associated with clinical outcomes for different congenital abnormalities. Prenatal clinical prediction models (prenatal-CPMs) incorporate these risk factors to estimate the probability of outcomes and have the potential to improve decision making and individualized care in fetal medicine. The aim of our study was to analyze existing prenatal-CPMs and identify strategies to enhance prenatal-CPM development.
Methods:
We conducted a systematic literature review for articles containing prenatal-CPMs for surgically correctable congenital abnormalities published before December 31, 2015. Prenatal-CPMs were defined as models that were developed from a fetal cohort with at least two independent risk factors that predicted a clinically significant outcome. Prenatal-CPM characteristics such as index condition, covariates, predicted outcomes, model development method, model performance, and model evaluation were extracted according to the CHecklist for critical Appraisal and data extraction for systematic Reviews of prediction Modeling Studies (CHARMS) Criteria.
Results:
Of the 478 abstracts that were screened, only 9 (1.88%) articles were included in our study (Figure). There were 4 unique index conditions, including congenital cardiac abnormalities (n=4 prenatal-CPMs), congenital diaphragmatic hernias (n=2), twin-twin transfusion syndrome (n=2), and micrognathia (n=1). The majority of prenatal-CPMs included in the study were retrospective (66.6%), single-centered (88.9%), prognostic (77.8%) studies that produced a clinical score (55.5%) predicting perinatal mortality (77.8%) using data during the second trimester (66.7%). Only 1 prenatal-CPM (11%) explicitly addressed sampling bias from elective termination of pregnancies (TOP). All prenatal-CPMs excluded TOPs from the analysis. In addition, two prenatal-CPMs (22%) also excluded fetal deaths from final analysis. Only one prenatal-CPM addressed how missing data was handled. In terms of model performance, model discrimination was only available for 3 prenatal-CPMs (33.3%) while one model provided calibration statistics. Furthermore, only one study provided internal validation.
Conclusion:
Our study revealed that only a handful of prenatal-CPMs were developed over the last decade for the management of fetal congenital malformations. Current prenatal-CPMs show significant methodological limitations, such as selection bias, and lack reporting on model performance measures and on handling of missing data. Transparent and systematic reporting of multivariate prediction models and novel statistical methods that take into account the selection bias from fetal loss are needed in future prenatal-CPM development.
