Morphometric sum optical density as a surrogate marker for ploidy status in prostate cancer: an analysis in 180 biopsies using logistic regression and binary recursive partitioning

Abstract

We sought to identify morphometric descriptors predictive of nondiploidy in prostatic adenocarcinoma on prostate needle biopsies using logistic regression (LR) and binary recursive partitioning (BRP) and compare the equivalence of both methods. A total of 180 prostate needle biopsies diagnosed as prostatic adenocarcinoma were selected. Deoxyribonucleic acid ploidy and morphometry were performed separately on Feulgen-stained sections from these biopsies using the CAS-200 system and Nuclear Morphometry Suite, respectively. Seven morphometric predictors were tested as predictor variables, including nuclear area, circularity, elongation, sum optical density (OD), configuration run length, coefficient of variation (CV), and angularity. Logistic regression (LR) identified a two-parameter model including sum OD and circularity that had a 93.9% overall correct prediction rate (area under curve=0.950; 95% CI: 0.913, 0.987). A reduced model including only sum OD was equally good without any significant loss of predictive accuracy (93.3% correct overall classification rate). BRP also selected sum OD as the most predictive parameter; a sum OD cut-off of 7.73 in this model identified 93.3% of the nondiploid cases correctly. Morphometric OD can be used as a surrogate marker of nondiploidy. LR and BRP models are both equivalent in identifying and correctly classifying nondiploid cases of prostate cancer using sum OD as the predictor variable.