METHOD FOR PREDICTION OF CLINICAL OUTCOME OF BREAST CANCER Russian patent published in 2015 - IPC A61B5/00 G06N5/00 

FIELD: medicine.

SUBSTANCE: naive-Bayesian networks with optimised number of assemblies are used to derive a maximum area under ROC curve. Significant parameters having an influence on the clinical outcome of breast cancer are determined. Each assembly, except for a base one, accords with one of the parameters of a patient's data base. The base assemblies accord with final points specifying a disease progression or a lethal outcome in a pre-set time slot. To predict the disease progression, the following parameters are considered to be significant: age at the surgery, menstrual health, N category - lymph nodes, progesterone receptor expression, HER-2/neu receptor expression, pre-operative radiation treatment, neoadjuvant chemotherapy. The lethal outcome is predicted by the significant parameters as follows: menstrual health, T category - tumour size, N category - lymph nodes, biomolecular subtype, mRNA YB-1 expression level, pre-operative radiation treatment, hormone therapy. The data derived by constructing ROC curves are used to plot risk histograms of the clinical outcome of breast cancer connecting a conditional probability of the clinical outcome and an absolute probability of the respective clinical outcome. The conditional probability of the clinical outcome is divided into four ranges in accordance with four risk groups. The pre-taught Bayesian network is questioned with using the significant parameters of a specific patient. The conditional probability derived by questioning the Bayesian networks of the respective clinical outcome provides a basis to refer the patient to one of the risk group. The respective outcome - the disease progression or the lethal outcome - is predicted by the produced value with the use of the risk histograms of the absolute probability.

EFFECT: more accurate prediction of the clinical outcome of breast cancer by optimising the Bayesian network by means of deriving the maximum area under ROC curve, facilitated doctor's aim to make decision on the patient's therapeutic approach by a small number of the prognostic parameters, applicability of the probability model with the great number of assemblies.

5 dwg, 2 ex