Tiered prediction system for preeclampsia: An integrative application of multiple models

Abstract

For years, it has been a challenge to identify women at risk of Preeclampsia (PE), one of the leading causes of maternal and perinatal morbidity and mortality. This would be especially useful in early pregnancy when modifiable factors can be addressed to reduce the risk or severity of outcome. Despite an increasing number of clinical and statistical prediction models being developed, which have been shown to outperform traditional maternal history or Doppler ultrasound approaches, it is still difficult to make accurate predictions based on a single model at a single time-point. Hence, here we investigate the use of multiple models integrated by Bayes' theorem. Methods: Prediction models based on three stages of pregnancy, pre-pregnancy, 15 weeks and 20 weeks of gestation, were developed with varying levels of sensitivity and specificity specific to each stage. Post-test probabilities at each stage are then calculated based on the Likelihood of each test using Bayes' theorem. The accuracy measures and predictive values are evaluated for both pre-test and post-test probabilities. Results: The overall proportion of truly identified cases have improved in the integrated model, with 81% correctly identified at 20 weeks of gestation, compared to 75% by the individual model. A relatively balanced accuracy can be achieved even when individual tests have been specified for higher sensitivity or specificity. Conclusion: Through an integrated prediction system, the accuracy of prediction is further enhanced and tailored for individual women, as the risk is assessed and updated throughout pregnancy based on predictors at different stages, the likelihood of PE from prediction at earlier stages, and clinicians' knowledge or hypotheses.