A systems-based modelling approach using transurethral resection of the prostate (TURP) specimens yielded incremental prognostic significance to Gleason when predicting long-term outcome in men with localized prostate cancer

摘要

OBJECTIVE To develop a systems-based model for predicting prostate cancer-specific survival (PCSS) using a conservatively managed cohort with clinically localized prostate cancer and long-term follow-up. PATIENTS AND METHODS Transurethral prostate (TURP) specimens in tissue microarray format and medical records from a 758 patient cohort were obtained. Slides were stained with haematoxylin and eosin (H&E), imaged and digitally outlined for invasive tumour. Additional sections were analysed with two multiplex quantitative immunofluorescence (IF) assays for cytokeratin-18 (epithelial cells), 4′-6-diamidino-2-phenylindole(nuclei), p63/high-molecular-weight keratin (basal cells), androgen receptor (AR) and α-methyl CoA-racemase, Ki67, phosphorylated AKT (pAKT)and CD34. Images were acquired with spectral imaging software. H&E and IF images were evaluated with image analysis algorithms; feature data were integrated with clinical variables to construct prognostic models for outcome. RESULTS Using a training set of 256 patients with 24% events, one clinical variable (Gleason score) and two tissue-specific characteristics (H&E morphometry and tumour-specific pAKT levels) were identified (concordance index [CoI] 0.79, sensitivity 76%, specificity 86%, hazard ratio [HR] 6.6) for predicting PCSS. Validation on an independent cohort of 269 patients with 29% events yielded a CoI of 0.76, sensitivity 59%, specificity 80% and HR of 3.6. Both H&E and IF features were selected in a multivariate setting and added incremental prognostic value to the Gleason score alone (CoI 0.77 to CoI 0.79). Furthermore, global Ki67 expression and AR levels in Gleason grade 3 tumours were both univariately associated with outcome; however, neither was selected in the final model. CONCLUSION A previously validated prostate needle-biopsy systems modelling approach that integrates clinical data with reproducible methods to assess H&E morphometry and biomarker expression, provided incremental benefit to the TURP Gleason score for predicting PCSS. Ki67 and AR, known to be associated with outcome in the prostate needle biopsy, were not associated with PCSS in multivariate models using TURP specimens.