Prostate Specific Antigen (PSA) Growth Curves: A Method to Improve Prostate Cancer Screening
Abstract
Prostate cancer (PrCA) screening aimed at detecting aggressive disease represents a significant public health issue. Development of biomarkers to predict PrCA that is likely to kill if left untreated is a major challenge. This dissertation focused on analyzing existing repeated measures of prostatic specific antigen (PSA) to develop and validate a tool to improve both sensitivity and specificity of the PSA-based screening test to detect high-risk PrCA. We used the Prostate Lung Colorectal and Ovarian trial data (PLCO) for PSA growth model building. Using 6 years of annual PSA measurements we established the PSA growth curves for four groups of men; those who developed high-risk PrCA, those who developed low-risk PrCA, those who developed benign prostatic hyperplasia (BPH) and those who were not diagnosed with either PrCA or BPH. We used these curves to estimate PSA annual rate of change at defined time points; one and two years before diagnosis for each individual in the cohort. We then examined the area under the curve (AUC) to estimate the specificity and the sensitivity of PSA annual rate thresholds. We validated our work by replicating the PSA growth models in a cohort of screened men in The Department of Veterans Affairs. Our results show that PSA annual change rates varied significantly by cancer status in both cohorts. The difference between the means of PSA rate values across the four groups of men was high and robust. Annual individual PSA rates showed substantial variability; however, a distinct range and significantly higher values were observed among men who developed high-risk PrCA. This resulted in high AUC (0.97) in the logistic regression model. A threshold of 0.37ng/ml/year had the best combination of sensitivity and specificity; i.e., of 97.2%, and 97.3% respectively. In the VA validation cohort, the same pattern was observed. However, men in the low-risk PrCA group had higher annual PSA rates as compared to the same group in the PLCO cohort. This resulted in a lower AUC of 93.3 (92.86-93.71) and the threshold of 0.37ng/ml/year predicted high-risk PrCA with a sensitivity and specificity of 95.5% and 86.2 % retrospectively.