Application of Principal Component Analysis to Newborn Screening for Congenital Adrenal Hyperplasia

Abstract

Purpose: Newborn screening laboratories are challenged to develop reporting algorithms that accurately identify babies at increased risk for congenital adrenal hyperplasia (CAH) due to 21-hydroxylase deficiency (21OHD). Screening algorithms typically use cutoff values for a key steroid(s) and include considerations for covariates, such as gestational age or birth weight, but false-positive and false-negative results are still too frequent, preventing accurate assessments. Principal component analysis (PCA) is a statistical method that reduces high-dimensional data to a small number of components, capturing patterns of association that may be relevant to the outcome of interest. To our knowledge, PCA has not been evaluated in the newborn screening setting to determine whether it can improve the positive predictive value of 21OHD screening. Methods: PCA was applied to a data set of 920 newborns with measured concentrations of 5 key steroids that are known to be perturbed in patients with 21OHD. A decision tree for the known outcomes (confirmed 21OHD cases and unaffected individuals) was created with 2 principal components as predictors. The effectiveness of the PCA-derived decision tree was compared with the current algorithm. Results: PCA improved the positive predictive value of 21OHD screening from 20.0% to 66.7% in a retrospective study comparing the current algorithm to a tree-based algorithm using PCA-derived variables. The streamlined PCA-derived decision tree, comprising only 3 assessment points, greatly simplified the 21OHD reporting algorithm. Conclusions: This first report of PCA applied to newborn screening for 21OHD demonstrates enhanced detection of affected individuals within the unaffected population.