MAAUA: 2006 Posters - Improved Prediction Of PSA Biochemical Recurrence By Quantitative Nuclear Grade (QNG) Signature Compared To Pathology Findings Post-prostatectomy

Back to Scientific Program
Back to Annual Meeting

Improved Prediction Of PSA Biochemical Recurrence By Quantitative Nuclear Grade (QNG) Signature Compared To Pathology Findings Post-prostatectomy
Robert W Veltri*, Cameron Marlow*, Danil A Makarov*, Michael C Miller*, Alan W Partin
Johns Hopkins University School of Medicine, Baltimore, MD

Introduction: We used image analysis of nuclear morphometry to predict biochemical (PSA) recurrence in men with prostate cancer (PCa). These men underwent radical prostatectomy (RP) and had long term follow-up (FU).
Methods: The NCI Cooperative Prostate Cancer Tissue Resource (CPCTR) tissue microarray was prepared from RP cases treated in 1991-92. The CPCTR-TMA cores were available from n=78 cases (n=39 non-recurrences and 39 PSA recurrences). Feulgen-stained nuclei were captured from all TMA spots using the AutoCyte Pathology Workstation. We applied Multivariate Logistic Regression (MLR) for selection of 40 nuclear morphometric descriptors (MND) to calculate Quantitative Nuclear Grade (QNG) and Pathology models, which derived predictive indices and predictive probabilities (PP) of the two status groups. Kaplan-Meier plots were also produced for survival analysis.
Results: A MLR model for QNG predicted PSA recurrence using eighteen NMDs with the area of receiver operator characteristic curves (AUC-ROC) = 0.865, a sensitivity of 89% and specificity of 60% and an accuracy of 74%. RP pathology yielded an AUC-ROC = 0.70 and the model retained only p_Stage and Gleason Sum Score variables and yielded a sensitivity of 53%, specificity of 60% and an accuracy of 56%. A combination of QNG and Pathology yielded an AUC-ROC = 0.88, sensitivity of 87%, specificity of 72% and an accuracy of ~80%. Kaplan-Meier plots concur with these observations.
Conclusions: We were able to significantly exceed the prediction of PSA recurrence over available routine RP pathology using QNG. In combination with pathology specificity of the model was increased.

Back to Scientific Program
Back to Annual Meeting