Analysis of Ordinal Logistic Regression Model on Breast Cancer Diagnosis by Birads Mammography

Abstract

The right diagnosis is needed for appropriate therapy. The diagnosis of breast cancer is quite ambiguous and requires high accuracy. Mammography is a method of diagnosing breast cancer using BIRADS (Breast Imaging-Reporting and Data System) assessment. This study aimed to assess the accuracy of BIRADS classification in the diagnosis of breast cancer and predictors that influence it through a logistic regression model test. The research method was cross sectional study by collecting data from the results of mammography examinations obtained from Medical Record documents, SIRS (Hospital Information Systems), and the radiologist's expertise of mammography. The data came from 47 hospital breast cancer patients that contained information on potential predictors of breast cancer namely tumor location, metastases, age, weight, and education. Logistic regression model analysis was performed to find the best statistical test model for breast cancer diagnosis classification based on BIRADS assessment. The diagnosis classification of BIRADS was consisting of normal, benign, and malignant grades. For this reason, hypothesis testing was conducted with G test for simultaneous model testing. Then, a development of an appropriate logit model by using a partial test. Followed by conducting a suitability and feasibility test model with the Goodness of Fit using the Hosmer-Lemeshow Test. The results of the analysis revealed that the ordinal logistic regression was the best model of BIRADS classification diagnosis with an accuracy value of 52.5%. The result of ordinal logistic regression model for malignant breast cancer: