Research on the Time-Point Prediction Model for NIPT Based on DOA-CEO Optimization Algorithm

Abstract

Fetal cell-free DNA in a pregnant woman's blood doesn't reach high levels, and nobody really knows when to time the non-invasive
prenatal test for best results. That's the problem we tackled. We built a prediction model that mixes machine learning with some custom optimization tricks. The model uses what we call a DOA-CEO algorithma hybrid that pulls together the strengths of two different approaches.
DOA brings strong search capability while CEO adds randomness that keeps things from getting stuck. Together they tune the model's grouping settings and starting weights. We tested the approach across 11, 033 real clinical cases. Results show the model performs well within
5% error margins. The R value hit 0.984, which we'd consider pretty good. When we look at error distribution, 95.3% of predictions stayed
within that 5% detection band. We compared our method against standard multiple linear regression and the basic DOA approachboth of
them fell short on prediction errors. The hybrid algorithm also converged noticeably faster, clocking about 34.6% improvement over the original DOA alone. What this means: we now have a tool that estimates when NIPT should run and what risk level to expect. Clinicians could use
this when designing personalized testing schedules for their patients. The work gives both a scientific foundation and something practical to
work with in real hospital settings.