Assessment of prognostic factors of prostate cancer : limitations and possibilities of morphology

Abstract: Prostate cancer is a leading cause of cancer morbidity and mortality. It is a morphologically, genetically, and clinically heterogeneous disease. Stage and grade are important predictors of patient outcome. Extraprostatic extension (EPE) of prostate cancer is a key component of staging but it is not fully understood how its histopathological characteristics correlate with outcome. Gleason grading takes the morphological heterogeneity into account and is considered one of the best prognostic factors of prostate cancer. The grading system has evolved considerably over time and it is essential to understand how this affects its utility. The aim of this thesis was to classify patients with EPE into prognostic groups and to evaluate Gleason grading trends over time and how grading reproducibility can be improved. We reviewed 1051 radical prostatectomy (RP) specimens and found 470 cases with EPE. Men with EPE had a higher risk of biochemical recurrence. When stratified by the extent and other pathological features of EPE, radial extent predicted recurrence, while perineural invasion at the site of EPE and circumferential extent did not. We analyzed trends in Gleason grading practices in Sweden and assessed the impact of the 2005 International Society of Urological Pathology (ISUP) revision. Data on 97,168 men with a primary diagnosis of prostate cancer in needle biopsy from 1998 to 2011 were obtained from the National Prostate Cancer Register (NPCR). There was a shift towards higher Gleason scores (GS) at diagnosis over the period but more evident after the ISUP revision. The trend remained when stage migration was factored in. This grade inflation has consequences for therapy decisions, such as the eligibility for curative treatment or active surveillance. The concordance between GS in biopsies and subsequent RP specimens was analyzed in 15,598 men registered by the NPCR between 2000 and 2012. The agreement improved from 55% to 68% during the period, but most of the improvement occurred before 2005. When adjusted for GS and year of diagnosis, the GS prediction became less accurate over time. A limitation of Gleason grading is that it is subjective and suffers from interobserver variability. We analyzed causes of disagreement in 87 prostate cancer biopsies, included in a reference image database for standardization of pathology. A group of 23 international experts failed to reach consensus in 41% of cases. The most frequent cause of disagreement was between GS 3+3 with tangential cutting artifacts and GS 3+4 with poorly formed or fused glands. An artificial intelligence (AI) system trained in grading assessed the grades of non-consensus cases and obtained a weighted kappa value of 0.53 compared to 0.50 for the pathologists, placing AI as the sixth most reproducible observer. In conclusion, prostate cancer is a heterogeneous disease calling for individualized diagnosis and treatment. These studies have highlighted some limitations of histopathological prognostic factors and suggested more standardized assessments. In a near future, AI may serve as a decision support for more consistent diagnoses.