Logistic regression generated probability estimates for trisomy 21 outcomes from serum alpha fetoprotein and beta human chorionic gonadotrophin: Simplification with increased specificity

Traditional Gaussian approaches to biochemical screening make calculations very cumber-some. We have sought to develop a logistic regression mathematical model for the calculation of risk for trisomy 21 in multiple marker screening that would be more "user friendly." Biochemical screening data from 15,444 patients with confirmed normal outcomes and 28 with trisomy 21 were analyzed by logistic regression analysis (LRA) to produce a direct mathematical approach to risk calculation that combines the improvements in sensitivity and specificity already demonstrated by us with extreme value cumulative function surface fittings, but which can be extended for other exploratory variables. LRA yields a 71.4% sensitivity with a 95% specificity which is comparable to our previously published discriminant aneuploidy detection (DADs) approach; both are an improvement over the commonly used Gaussian approach of risk assessment. In our study population, LRA performed as well as DADs; both performed better than the Gaussian method, and LRA is a simpler approach.