ROC analysis of an erythroblast morphologic scoring system to improve identification of fetal cells in maternal blood

Nucleated red blood cells are used in research settings as a target cell type for investigations of noninvasive prenatal diagnosis, and these cells have a characteristic nuclear morphology and hemoglobin staining pattern that makes them distinguishable from maternal cells. Recently, we developed an erythroblast scoring system based on these characteristics. Here, we employ statistical analyses to further characterize the utility of this scoring system.

A total of 170 nucleated red blood cells isolated from peripheral blood of four women undergoing elective termination of a trisomy 21 male fetus were analyzed. Each of the four scoring system parameters, whose values range from 0 to 3 points, served as an independent variable to determine its significance for the correct identification of a cell as fetal or maternal. A logistic regression was used as a multivariable statistical tool.

Forty-four patterns of the four parameters were found in the overall series. Some patterns were exclusively associated with maternal cells (e.g. 1-1-1-1 and 1-1-2-2), and others were exclusively associated with fetal cells (e.g. 3-2-2-2 and 3-2-3-2). A detection rate of 73.9% at a false-positive rate of 5% resulted from a random simulation model performed with a 1:5 case:control matched set. The variables most predictive of a cell being fetal were gamma hemoglobin staining intensity of cytoplasm and nuclear roundness.

The modified scoring system presented here improves upon the previously reported, unmodified erythroblast scoring system. These statistical analyses suggest that the scoring system is a promising method that aids in distinguishing fetal and maternal NRBCs for prenatal diagnostic applications and that it may be amenable to automated microscopy by applying the discrete morphological parameters as computational classifiers.