Ensemble learning-driven hybrid prediction model for improved prenatal down's syndrome screening: a comparative study with laboratory-based median equations.

OBJECTIVE

This study developed and validated a hybrid prediction model (HyPred) for prenatal Down's syndrome (DS) screening using ensemble learning（EL）-driven, including extreme gradient boosting, balanced random forest, and gradient boosting machine. Its performance was compared with laboratory-based median equations.

METHODS

A retrospective analysis was conducted using 8,363 first-trimester samples (Nov 2019-Aug 2022) for training and 1,943 samples (Sep 2022-Jul 2023) for validation. HyPred was developed in R and compared with a laboratory-based median equation in terms of sensitivity, specificity, and other metrics.

RESULTS

The lab-based median equation improved screening over the default equation. In the validation set, the EL model showed superior performance with higher accuracy, robustness, and adaptability. HyPred achieved an AUC of 0.97, surpassing the median equation in key metrics.

CONCLUSION

The EL model offers improved accuracy and robustness for prenatal DS screening. Despite higher computational demands, modern tools facilitate its optimization, supporting precision medicine through more customized screening.