Selection of BRCA1/2 negative cases using data mining analytical approach for hereditary breast cancer prediction in high risk breast cancer patients

Abstract

Background:The causes of hereditary breast cancer are classified in two groups as modifiable and non-modifiable (genetic) in our study.Some of the non-modifiable factors selected are breast density, menarche age, menopause, BRCA1 and BRCA2 genes, family history, height and weight of patients.On the other hand, oral contraceptive usage, alcohol consumption, hormone therapy, breastfeeding, exposure to radiation and smoking are among the modifiable factors.The risk of having hereditary breast cancer could be identified by BRCA1/BRCA2 gene tests which requires long lasting experiments and incurs high costs. Before performing the required genetic test, the modifiable and non-modifiable factors of the high risk patients are collected by geneticist or genetic counselor and then calculated risk score of the patients.After observing the test results, the effect of existing factors are analyzed.The aim of this study is to develop a hereditary breast cancer prediction algorithm by using data mining techniques for high risk breast cancer patients.Methods:Different applications may require different data mining methods.The current study involves the data classification technique for predicting hereditary breast cancer through a function that incorporates modifiable and non-modifiable factors. The data consists of 562 BRCA negative, and 68 BRCA positive instances with 117 categorical and numerical characteristics.The first step of this approach is preprocessing the data on hand.Individuals with incomplete data are excluded from the list, and some logical adjustments are performed.Results:As a result, 440 patients (392 BRCA negative, 48 BRCA positive), and 75 characteristics (3 numerical, 72 categorical attributes) are obtained.In the second step, important factors are determined by attribute selection algorithms. After applying 23 selection and 7 ranking methods, the important attributes are determined as: age, diagnosis, height, weight, menopause age, menarche age, FIB/FIB<40, SEB/SEB<40, THB/THB<40, FIO/FIO>40, SEO, cancer status of the family, clinical stage, and pathological stage.As a result, three data sets are suggested for classification. Finally, a total of 182 data classification results are derived for the aforementioned data sets.All analyses are performed by employing WEKA software.Conclusion:Comparison of the results shows that, the best data classification method's predictions on the BRCA negative class has 100% accuracy.Hence, a new individual's situation can be assessed without undergoing detailed gene tests, resulting in cost and required workforce reduction.The results are found to be promising and applicable.As an extension of the study, a user friendly interface and a decision support tool can be developed.