Journal of Health and Biomedical Informatics

Introduction: Breast cancer is one of the most common cancers, and also it is the most common type of malignancy in Iranian women that has been growing in recent years. The risk of recurrence is usual in patient. Many factors may increase or decrease the recurrence rate. Data mining methods have been used to diagnose or predict cancer and one of the most application of data mining approaches is prediction of breast cancer recurrence
Method: This is a retrospective study. Collected data on 809 patients with breast cancer with 18 fields for each patient were used. Due to excessive missing data only about 665 cases have been used. Since the number of fields in the remaining records with null values have been observed, as a preprocessing and data preparation phases, these values have been estimated by the EM algorithm and using SPSS.v20 software. In this study, a model for prognosis of breast cancer recurrence among patients using J48 tree has been developed. Results: The specificity and sensitivity of the developed model are 53% and 85%, respectively. Moreover, only 14% of patients who have relapsed are known as false negative with developed model.
 Conclusion: Creating a predictive model with appropriate specificity and sensitivity can warn patients about recurrence and timely preventive measures to prevent progression of the cancer. The False Negative rate is very important in medical prediction models that can make serious results/consequences. In present study this rate is about 14% that seems reasonable amount in term of modeling.