Prediction of prostate cancer using decision tree algorithm

Aim: Serum Prostate Specific Antigen (PSA) level is used for prediction of cancer, but this approach suffers from weak sensitivity and specificity. We applied binary-split decision tree (DT) algorithm to prostate cancer prediction problem. Materials and methods: Quick, Unbiased and Efficient Statistical Tree (QUEST) algorithm was used in 750 patients who had a serum PSA levels between 0 and 10 ng/mL. Results: The analysis indicated that following five nodes had different levels of cancer possibility: (1) PSA 5.98 ng/mL;(2) PSA = 5.98 ng/mL and Digital Rectal Examination (DRE) is suspicious or positive; (3) PSA =5 .98 ng/mL and DREis negative and free PSA 0.81; (4) PSA = 5.98 ng/mL and DRE is negative and free PSA = 0.81 and age = 57 years; (5)PSA = 5.98 ng/mL and DRE is negative and free PSA = 0.81, and age more than 57 years. The incidences of cancer detection in these groups were 25%, 15%, 0%, 4%, and 16%, respectively. In cases where the nodes 3 and 4 were evaluated as negative, the system would detect 97 of 98 cancer cases with 0.99 sensitivity, saving 74 patients from biopsy (13% ofthe patients). Conclusion: DT seems to be a valuable tool to increase specificity in prediction of prostate cancer.