Prostate cancer prognosis via integrative and co-localized glandular morphometry and immunofluorescent protein biomarker expression.

Abstract

262 Background: Quantitative image analysis of both the prostate needle biopsy (PNB) and the radical prostatectomy sample (RPS) have proven to be robust and predictive for prostate cancer (PCa) prognosis. While metrics assessing morphology and protein biomarker expression separately have been published in prognostic risk prediction models, we sought to determine the performance of novel approaches which combine protein biomarker expression with the morphometry of the gland wherein they express. In order to assess robustness, we evaluated these metrics in both PNB and RPS cohorts. Methods: Novel quantitative biometric features were derived from computerized analysis of morphology and immunofluorescent (IF) biomarker expression in PCa images. A Bayesian probabilistic approach to co-locate and quantify IF biomarker expression within various glandular sub-types was developed. This permitted differentiable analysis of PCa biomarkers such as androgen receptor (AR) in high and low Gleason grade glands of the same patient. Performance in predicting significant clinical disease progression (including metastasis and death-of-disease) was assessed via the concordance index (CI). Performance was evaluated independently in a cohort of 326 PNB samples (median 8-year follow up) and a cohort of 373 RPS (median 5-year follow up). Results: In a univariate analysis, previously published glandular morphology features had a CI of 0.68 (PNB) and 0.76 (RPS). Quantitative AR had a CI of 0.67 (PB) and 0.72 (RPS). Novel metrics of integrative co-localization resulted in a significant increase in prognostic accuracy; CIs of 0.72 (PNB) and 0.83 (RPS). These features separately assessed AR expression depending on the multiple glandular sub-types within the tumor sample. Conclusions: Integrating quantitative image analysis of morphology and IF biomarker expression in the PNB and RPS outperformed previous quantitative metrics. Such robust and reproducible metrics may prove beneficial in improving risk stratification models for future treatment decision making.