Prostate cancer prognostication based on clinical and histopathological tumor features

Abstract

Prostate cancer is the second most common cancer in men worldwide. Of almost 1.3 million newly diagnosed men per year, up to 80% will have localized disease with a characteristically prolonged natural history. Risk stratification and treatment decision-making for these men is currently based on the combination of standard clinical and histopathological predictors, such as the Gleason score, prostate specific antigen (PSA) level and clinical tumor stage at diagnosis. However, these standard predictors are not sufficient to capture the heterogeneity in prognosis for men with localized prostate cancer. As a consequence, these men are often overtreated and may suffer from treatment-related side effects. In this thesis we aimed to improve prognostication for men with localized prostate cancer through validation of existing risk stratification tools based on standard clinical and histopathological factors, and through validation of existing, and identification of novel, prognostic markers.

In Study I, we evaluated if the nested case-control study design is appropriate for estimating relative and absolute risks of dying from prostate cancer in the presence of competing risks. We used a case-control study (ProMort I) nested in the National Prostate Cancer Register of Sweden (NPCR). We found that the relative risks of dying from prostate cancer estimated in ProMort I were comparable to the relative risks estimated in the NPCR. The relative risks of dying from other causes estimated in ProMort I were biased, which led to biased estimates of the absolute risks of dying from prostate cancer. The bias in both the relative and absolute risks was reduced by augmenting competing-risks cases, and especially by augmenting both the competing-risks cases and the controls. Our results indicate that, without the additional extensions to the design, the nested-case control studies are not suitable for the development of models predicting death from prostate cancer in the presence of competing risks.

In Study II, we systematically compared the prognostic performance of the most commonly used pretreatment risk stratification tools in predicting death from prostate cancer using data from the Prostate Cancer data Base of Sweden. The Memorial Sloan Kettering Cancer Center nomogram, Cancer of the Prostate Risk Assessment score and Cambridge Prognostic Groups discriminated death from prostate cancer better than the D’Amico and D’Amico-derived risk grouping systems. The order of performance remained after stratifying by primary treatment and year of diagnosis. Using these tools could improve clinical decision-making.

In Study III, we evaluated if a virtual microscopy system which we developed for central re-review in ProMort I and Study IV can be used interchangeably with standard light microscopy for the histopathological evaluation of prostate cancer. We found good repeatability (i.e., intra-observer agreement) and reproducibility (i.e., inter-observer agreement) for several key prostate cancer histopathological features (i.e., core length, tumor length, primary and secondary Gleason pattern, the Gleason score and the Gleason Grade Groups (GGs)) both within and between light and virtual microscopy. The repeatability and/or reproducibility for some of the rare, or less commonly reported, features and for the percentage of Gleason pattern 4 was poor. The repeatability and/or reproducibility for these features should be improved before they are used in prognostic models. For all evaluated features, the agreement was similar within and between light and virtual microscopy indicating that light microscopy and our internally developed virtual microscopy system can be used interchangeably for the histopathological evaluation of prostate cancer.

In Study IV, we evaluated if the International Society of Urological Pathology (ISUP) revisions of the Gleason grading systems have improved prostate cancer prognostication. We used a nested case-control study (ProMort II) to compare the prognostic performance of the pre-2005 Gleason score and the ISUP 2014 Gleason score. In our study, the ISUP 2014 Gleason score discriminated death from prostate cancer better than the pre-2005 Gleason score. Our results also indicate that this improvement may be due to classifying all cribriform patterns, rather than poorly formed glands, as Gleason pattern 4. We then evaluated if other histopathological features can further improve the prediction of death from prostate cancer. The number of cores with ≥50% cancer involvement, comedonecrosis and high-grade prostatic intraepithelial neoplasia (HGPIN) predicted death from prostate cancer independently of the GGs. Only comedonecrosis and HGPIN remained independent predictors when added to the model with all the standard predictors (the GGs, age, PSA and clinical tumor stage at diagnosis). Adding these features had minimal impact on the model discrimination.