THE ASSOCIATION ANALYSIS AND TIMING OPTIMIZATION OF NIPT DETECTION INDICATORS BASED ON GENERALIZED ADDITIVE MODELS AND CLUSTERING DECISION-MAKING

Abstract

Addressing the challenge of insufficient accuracy in non-invasive prenatal testing (NIPT) caused by traditional empirical grouping and a uniform testing timeline, this study focuses on the dynamic evolution mechanism of Y-chromosome concentration in male fetuses and the scientific optimization of the clinical intervention window. The study first utilized the Generalized Additive Model (GAM) to overcome the limitations of linear assumptions, providing an in-depth analysis of the nonlinear driving effects of gestational age, BMI, and their interaction terms on Y-chromosome concentration. Empirical results indicate that concentration exhibits significant three-stage fluctuations with gestational age, and BMI demonstrates a strong negative inhibitory effect beyond the critical threshold of 28. Subsequently, the study applied the K-means clustering algorithm combined with the elbow rule to achieve scientific stratification of the pregnant population, dividing it into two core subgroups: normal-to-high BMI and high BMI. Building on this, a logistic regression model was constructed and coupled with a comprehensive risk function incorporating weights for testing failure and diagnostic delay, thereby establishing a testing paradigm aimed at minimizing risk. The study confirmed that the optimal testing time point for both groups is 12 weeks of gestation, effectively balancing detection accuracy and clinical timeliness. Residual diagnosis and risk validation demonstrated that this modeling system possesses high statistical robustness, providing mathematical support for improving the quality of decision-making in precision prenatal screening.