Logistic回归在建立产前筛查唐氏综合征风险估计模型中的应用

利用64772例12~20周筛查孕妇(正常孕妇64718例, 唐氏患者54例)的孕妇血清学指标, 采用Logistic回归, 建立三联筛查方案和二联筛查方案的风险估计模型. 采用2种方案, 计算全部筛查孕妇的风险值, 通过ROC曲线下面积、检出率、阳性率评价2种筛查方案的效果. 结果表明: 年龄、甲胎蛋白、游离人绒毛膜促性腺激素和游离雌三醇在Logistic回归模型中均有统计学意义, 三联方案ROC曲线下面积大于二联方案, 但无显著性差异(z统计量为1.382, P=0.1669); 在同一截断值下, 三联方案检出率高于二联方案, 假阳性率低于二联方案, 差异有统计学意义(t=-9.44, P<0.001). 三联方案根据约登指数计算, 最佳筛查截断值为1:400, 假阳性率为4.3%.

Predicting Risk of Prenatal Screening for Down Syndrome Using Logistic Regression Model

In this study, a total of 64772 pregnant weman in gestational timeframe between 12 to 20 weeks are selected for analytical purposes, including 64718 normal pregnant weman and 54 suffering Down’s syndrome. The maternal serum markers are converted to serum with multiple of median (MoM) adjusted by grouping the gestational age and weight. The MoM is adjusted by twin, smoking and diabetes in the Logistic regression model. Risk estimation is calculated for each pregnant woman using two different screening methods. Area under the curve of receiver operator characteristic (AUC), detection rate and false positive rate are calculated individually to assess the effect of these two different methods. Results show that age, alpha-fetoprotein (AFP), free human chorionic gonadotrophin (HCG) and unconjugated estriol (uE3) are significantly dominant in the Logistic regression model. The AUC of tripe serum screening is larger than that of double serum screening, but manifests no significant difference (z=1.382, P=0.166 9). With the same cutoff value, the detection rate of triple serum screening is higher than that of double serum screening, and the false positive rate of triple serum screening is lower than that of double serum screening. The significant difference is detected between these two different models (t=-9.44, P<0.001). Triple serum screening is found to be the superior todouble serum screening compared with other screenings. Also noted is that when the false positive rate is 4.3%, the optimal cutoff value turns out to be 1:400.