Predicting neutropenia in breast cancer patients undergoing chemotherapy using a 2-stage FOS-3NN model trained on first cycle blood counts.

摘要

Previous studies have indicated that the ability to administer a full 6-cycle course of chemotherapy in the prescribed time period results in an increased patient survival rate for breast cancer patients. Neutropenia is a serious and common barrier to this goal. This low white blood cell condition necessitates either a dose reduction or delay in treatment while bone marrow cells recover. The ability to classify patients into low- or high-risk groups for developing neutropenia would allow physicians to more closely monitor high-risk patients and allow supportive care treatments such as growth factor support---which can lead to quicker recovery from or avoidance of neutropenia---to be administered earlier.; A combination of the nonlinear systems modeling technique Fast Orthogonal Search and the Nearest Neighbour classification scheme produced a model that was tested on 21 patients. The model was trained on an even split of low- and high-risk patients and validated on an entirely separate testing set. High-risk characteristics were defined prior to the study based on an evaluation by the participating oncologist. The data on which the model was built consisted of white blood cell, absolute neutrophil, platelet and haemoglobin counts including baseline counts (day 0 or prior), day 7 and day 28 of the first cycle of the chemotherapy regimen.; Nineteen out of 21 patients in the test sets were correctly classified. This results in a Fisher's exact test probability of P 0.00023 (2-tailed) and a Matthews' correlation coefficient of +0.83. This work is highly significant. Developing clinical-support tools to identify high-risk patients will lead to lower occurrence of neutropenia, more intensive chemotherapy regimens, and hence better prognosis for the patient's survival.