Risk Analysis and Prediction for Prostate Cancer based on Data Mining

This data set was from the prostate cancer early warning data of PLA General Hospital. Logistic regression and decision tree were used to screen the influencing factors of prostate cancer (PCa) and to build a prediction model. Logistic regression showed that total prostate-specific antigen (total PSA, tPSA), serum albumin, alkaline phosphatase, low density lipoprotein cholesterol, Apolipoprotein Al and inorganic phosphorus increased, while f/tPSA(the ratio of free PSA to total PSA) decreased, which was risk factors for prostate cancer. The area under ROC curve predicted by models in the modeling group and verification group was 0.741 (0.706, 0.777) and 0.763 (0.705, 0.821), respectively. The decision tree results showed that PCa prediction rate was 75.2% when f/tPSA $\leq$ 0.098. The top four indicators in terms of importance were f/tPSA, free PSA (fPSA), inorganic phosphorus and tPSA, and the area under ROC curve was 0.735 (0.704, 0.767). This study has provided a basis for assisting clinical diagnosis and improving diagnostic screening of PCa.