Modeling Breast Cancer Risks Using Artificial Neural Network: A Case Study of Kenyatta National Hospital in Nairobi, Kenya

Background: Early diagnosis of breast cancer is crucial to the survival of breast cancer patients. In the last years, improved technology has been adopted to aid data collection and store patients information in a database. Data mining may be used on such databases to come up with patterns that can help in predictability of diseases such as cancer. Use of statistical models may be used to aid doctors and not substitute their opinion. One search model is an artificial neural network model (ANN). Methods: Secondary data was collected from Kenyatta National Hospital (KNH), which is Kenyas national referral hospital located in the capital city, Nairobi. A total of 370 breast cancer patients information was obtained from both the inpatient and outpatient files. A three layer feed-forward artificial neural network was trained using 320 records. The ANN model obtained was used to predict malignancy in the remaining data set. A logistic regression was used to test which independent variables were significant. Receiver operating curve was used to evaluate the ANNs discriminative performance. Coefficient of determination (R2) was used to evaluate the goodness of fit. Results: ANN demonstrated a superior sensitivity performance over the logistic regression. There were no false positive and no false negative, however for the logistic regression there were seven false positives and eight false negative. The full logistic regression model showed that there were 4 significant independent variables. A reduced logistic regression model was obtained which consisted of 5 independent variables down from 21. ANN was found to have better discriminative performance (AUC=1) as compared to logistic regression (AUC=0.98909). Conclusions: The authors artificial neural network has high discriminative performance and can accurately predict breast cancer