Risk prediction in prostate cancer diagnostics : current challenges and improvements

Abstract

Prostate cancer has been considered a disease of elderly men, and thus historically less focus has been on prostate cancer research than many other cancer types. However, as life expectancy is increasing all over the world, more life years are lost when men are diagnosed with prostate cancer at the age of 70 years now than before. Therefore, it is increasingly important to improve the diagnostic pathway of prostate cancer in modern health care. My thesis aims to address some of the issues in the current prostate cancer diagnostic pipeline using risk prediction models.

Measuring the level of prostate-specific antigen (PSA) in blood is widely used as a blood test to screen for prostate cancer and evidence has shown that mortality decreases with PSA testing. However, because PSA testing has a high false-positive rate, many unnecessary biopsies are performed on healthy men and many men are overdiagnosed with indolent disease (International Society of Urological Pathology (ISUP) grade group 1). In Study I the objective was to predict the risk of clinically consequential cancer (ISUP ≥ 2) at biopsy and the cumulative probability of having a negative biopsy when being PSA tested with one, two, three, four, or five to eight year intervals. We found that men with a PSA level above 1 ng/mL had an increased risk of ISUP ≥ 2 prostate cancer when screened with longer then annual intervals, while men with a PSA level below 1 ng/mL had low risk of ISUP ≥ 2 prostate cancer regardless of time between testing. The benefit of a shorter screening interval needs to be balanced with the increased cumulative probability of having a negative biopsy which we found to be twofold for annual vs. biennial testing intervals and threefold for annual vs. triennial testing intervals.

Knowledge about the relationship between PSA, age and different grades of prostate cancer is important for clinicians working with prostate cancer diagnosis because of how widely used the PSA test is. In Study II we studied the association between the risk of indolent and clinically consequential prostate cancer (ISUP 1 and ISUP ≥ 2) and PSA and age, respectively. Our study cohort comprised of 6.083 biopsied men from the STHLM3 study and 72.996 biopsy cores from those men. In the overall ISUP grade system, lower grades can be masked by higher grades, and thus we studied the associations for both overall ISUP grade and for ISUP grade on each biopsy core. Our results showed that ISUP 1 prostate cancer was not significantly associated with PSA or age, on overall ISUP grade or on individual biopsy core level. In contrast, our results showed that ISUP ≥ 2 prostate cancer is significantly associated with increasing PSA level and older age. Our results indicate that PSA leakage of ISUP 1 prostate cancer cells is more similar to that of benign prostate tissue than ISUP ≥ 2 prostate cancer tissue. The use of magnetic resonance imaging (MRI) before biopsy to diagnose prostate cancer has increased in current clinical practice. Combining results from prostate MRI with existing risk prediction models can improve the predictive abilities of the models. The aim of Study III was to develop a risk prediction model (S3M-MRI), combining the Stockholm3 score and the PI-RADS (Prostate Imaging Reporting and data System) score from MRI to predict the risk of ISUP ≥ 2 prostate cancer. We developed the S3M-MRI model using data from the STHLM3-MRI diagnostic study and compared the model performance of the S3M-MRI to the Stockholm3 model and PI-RADS score. We also compared five diagnostic strategies for clinical outcomes. We found that the combined S3M-MRI model had better predictive abilities than both the Stockholm3 and the PI-RADS alone. However, when we compared it to different clinical strategies, the sequential use of the Stockholm3 test followed by MRI on Stockholm3 positive men resulted in similar numbers of performed biopsies and diagnosed ISUP ≥ 2 prostate cancers while saving many MRI scans.

Prostate cancer diagnosis is based on the result of the prostate biopsy and reclassification of ISUP grade on radical prostatectomy samples compared to biopsy is common. In Study IV our aim was to study what effect reclassification of disease status based on prostate biopsies has on the performance of prostate cancer risk prediction models using simulations and data from the STHLM3 Radical Prostatectomy Cohort. The cohort comprised of 780 men from the STHLM3 study who were diagnosed with prostate cancer and treated with a radical prostatectomy between 2013 and 2015. We compared four simulated prediction model scenarios with and without error in disease status and calculated the area under the receiver operating characteristics (ROC) curve (AUC) of the Stockholm3 score for predicting clinically significant prostate cancer assessed using biopsy and radical prostatectomy samples. Our simulations showed that fitting a risk prediction model using data with error in the disease status only leads to a small decline in the true predictive performance, but leads to a large decline in apparent predictive performance when evaluated against data with error in the disease status. Moreover, our results showed that the Stockholm3 test has stronger association with clinically significant prostate cancer defined on prostatectomy samples (without errors) than on biopsy samples (with errors).

In conclusion, in this thesis we have aimed to describe a part of the risk associated with diagnosis of prostate cancer as well as developing new prostate cancer risk prediction models. This thesis contributes to the constant pursue of improving the current prostate cancer diagnostic pipeline in order to improve the lives of men screened for or diagnosed with prostate cancer.