Combined application of information theory on laboratory results with classification and regression tree analysis: Analysis of unnecessary biopsy for prostate cancer - Abstract

BACKGROUND:The probability of a prostate cancer-positive biopsy result varies with PSA concentration.

Thus, we applied information theory on classification and regression tree (CART) analysis for decision making predicting the probability of a biopsy result at various PSA concentrations.

METHODS:From 2007 to 2009, prostate biopsies were performed in 664 referred patients in a tertiary hospital. We created 2 CART models based on the information theory: one for moderate uncertainty (PSA concentration: 2.5-10ng/ml) and the other for high uncertainty (PSA concentration: 10-25ng/ml).

RESULTS:The CART model for moderate uncertainty (n=321) had 3 splits based on PSA density (PSAD), hypoechoic nodules, and age and the other CART for high uncertainty (n=160) had 2 splits based on prostate volume and free PSA. In this validation set, the patients (14.3% and 14.0% for moderate and high uncertainty groups, respectively) could avoid unnecessary biopsies without false-negative results.

CONCLUSIONS: Using these CART models based on uncertainty information of PSA, the overall reduction in unnecessary prostate biopsies was 14.0-14.3% and CART models were simplified. Using uncertainty of laboratory results from information theoretic approach can provide additional information for decision analysis such as CART.

Written by:
Hwang SH, Pyo T, Oh HB, Park HJ, Lee KJ.   Are you the author?
Department of Laboratory Medicine, Center for Diagnostic Oncology, Research Institute and Hospital, National Cancer Center, Goyang-si, South Korea.

Reference: Clin Chim Acta. 2012 Oct 16. pii: S0009-8981(12)00479-2.
doi: 10.1016/j.cca.2012.10.012

PubMed Abstract
PMID: 23078854 Prostate Cancer Section