Ability to predict biochemical progression using gleason score and a computer-generated quantitative nuclear grade derived from cancer cell nuclei

• Published in: Urology (Ridgewood, N.J.) Vol. 48; no. 5; pp. 685 - 691
• Main Authors: Veltri, Robert W., Miller, M. Craig, Partin, Alan W., Coffey, Donald S., Epstein, Jonathan I.
• Format: Journal Article
• Language: English
• Published: New York, NY Elsevier Inc 01.11.1996; Elsevier Science
• Subjects: Adult; Aged; Aged, 80 and over; Biological and medical sciences; Cell Nucleus - pathology; Diagnosis, Computer-Assisted; Disease Progression; Humans; Logistic Models; Male; Medical sciences; Middle Aged; Multivariate Analysis; Nephrology. Urinary tract diseases; Predictive Value of Tests; Prostatic Neoplasms - mortality; Prostatic Neoplasms - pathology; Survival Analysis; Tumors of the urinary system; Urinary tract. Prostate gland; Adenocarcinoma; Disease development; Human; Chromatin; Urinary system disease; Shape; Prostate disease; Cytologic investigation; Size; Malignant tumor; Texture; Result; Biomedical data processing; Image analysis; Predictive value; Male genital diseases; Prostate; Computer aid
• ISSN: 0090-4295; 1527-9995
• DOI: 10.1016/S0090-4295(96)00370-6

Abstract To determine the ability to predict prostate cancer progression using shape, size, and chromatin texture nuclear grading features preselected by logistic regression analyses based on expert-selected prostate cancer cell nuclei captured using a computer-assisted image analysis system. One hundred fifteen patients with clinically localized prostate cancer were identified at the Johns Hopkins medical institutions. The mean follow-up period was 10.4 ± 1.7 years in 70 patients without disease progression, whereas the mean time to progression for the entire group was 3.8 ± 2.5 years. Using 5-μm Feulgen-stained tissue sections, approximately 150 cancer cell nuclei were selected and captured for each case using a CAS-200 Image Analysis System. Thirty-eight different nuclear morphometric descriptors (NMDs) were calculated for each cell nucleus. The variance of the NMDs for each tumor was examined by univariate and multivariate logistic regression analyses and by Cox survival analyses to assess their ability to predict prostate cancer progression. Postoperative Gleason scoring was significantly correlated with disease progression ( P < 0.00001; sensitivity, 73%; specificity, 84%; receiver operating characteristic curve area under the curve (ROC-AUC), 83%). Using backward stepwise logistic regression at a stringency of P < 0.05, the variances of 11 of the NMDs were found to be multivariately significant for progression prediction ( P < 0.00001; sensitivity, 78%; specificity, 83%; ROC-AUC, 86%). A single value, termed the quantitative nuclear grade (QNG), was created from the variances of these 11 multivariately significant NMDs using the logistic regression function. The QNG and the postoperative Gleason score were combined to create a model for the prediction of progression having a sensitivity of 89%, specificity of 84%, and ROC-AUC of 92%. These two parameters (QNG and Gleason score) clearly separated the patient sample into three statistically distinct risk groups and predicted the time to progression on the basis of Kaplan-Meier survival probability analysis. The QNG, combined with the postoperative Gleason score, may assist in the more accurate stratification of patients undergoing radical prostatectomy into low-, moderate-, and high-risk groups for cancer recurrence and may permit the early initiation of adjuvant therapy.