A novel machine-learning algorithm to screen for trisomy 21 in first-trimester singleton pregnancies.

Area of Science:

• Medical screening
• Machine learning in healthcare
• Genetics and genomics

Background:

• First-trimester antenatal screening for Trisomy 21 (T21) in the UK utilizes Nuchal Translucency (NT), gestational age, Free β-HCG, and PAPP-A.
• Current screening employs multivariate Gaussian distribution models to calculate T21 risk as an odds ratio.

Purpose of the Study:

• To investigate the efficacy of machine-learning algorithms for predicting T21 in first-trimester singleton pregnancies.
• To compare the performance of machine-learning models against established screening methods.

Main Methods:

• Utilized a dataset of 86,354 first-trimester singleton pregnancy screening cases, including 211 with T21.
• Trained and tested machine-learning models using adaptive boosting technology.
• Compared test case results with pregnancy outcomes to evaluate model performance.

Main Results:

• A machine-learning model demonstrated superior performance compared to current multivariate distribution models (AUC 0.978 vs 0.974).
• Reduced false positive rates from 3.82% to 2.28%.
• Decreased overall screen positive rates from 4.00% to 2.48%.

Conclusions:

• Machine-learning algorithms offer significant improvements for first-trimester T21 screening in the UK.
• These advancements can be integrated into the existing UK screening program with minimal changes.
• Further performance enhancements are anticipated with larger datasets and improved outcome data.