Optimization of Multi-Factor NIPT Detection Time Points Based on Random Forest Feature Screening and K-Means Clustering

Abstract

Aiming at the problems that multiple factors (BMI, age, etc.) synergistically affect the detection time point in non-invasive prenatal testing (NIPT) and the accuracy of the existing single-factor optimization schemes is insufficient, this paper proposes a multi-factor optimization scheme of "random forest feature screening - K-means clustering - risk - compliance rate constraint". The study was based on the NIPT data of male fetuses from the prenatal diagnosis center of a tertiary hospital. After preprocessing (cleaning abnormal samples, standardizing gestational weeks format, and calculating derived indicators), random forest was used to screen the key factors affecting the attainment of Y chromosome concentration standards, and then multi-factor homogeneous grouping was achieved through K-means clustering. Finally, the optimal detection time points for each group were screened in combination with the principle of "compliance rate ≥85% and lowest risk", and the impact of detection errors was analyzed. The results show that random forest effectively identifies the key influencing factors and eliminates the interference of redundant variables. K-means achieves efficient grouping. The compliance rate of the optimal detection time points in each group meets the clinical requirements, and the time deviation is small when the detection error is ≤3%. This scheme can precisely match the differences of multiple factors, solve the problems of feature redundancy and inefficient grouping, and provide reliable decision support for the timing planning of NIPT clinical testing.