Designing a model for predicting colorectal cancer risk based on regression-logistic data mining technique

Abstract

Aim: Using machine learning for the early detection of this disease has an important role in improving disease indicators. Therefore, this study aims to design a disease prediction model based on data mining techniques to help in early diagnosis and provide evidence-based services. Methods: This is an applied descriptive studyconducted in 2020. The study population was all patients (800 people) referred to Taleghani Hospital in Abadan for diagnostic tests. The data were derived from the electronic records of during 2009-2010. The Spearman correlation method was used to identify the effective factors in determining the risk of CRC. Then, Binary Logistic Regression (BLR) analysis and Enter method, effective factors in determining the risk of CRC were identified. Finally, the regression prediction model for CRC was developed. SPSS 17 was used to analyze statistical data. P-value ≥ 0.05 was considered significant. Results: Eleven variables using the Spearman correlation coefficient showed a significant correlation with the output class (with and without colorectal cancer). The results of regression-logistic analysis using Enter 7 variables obtained a higher chance than other variables. The results of classifying the research samples using the Forward LR method showed that with this model, accuracy, precision, and sensitivity (91%, 93.5%, and 94.5%, respectively) had high performance. Conclusion: Designing a risk prediction model based on logistic regression plays an important role in rapid, accurate, and timely screening of patients in improving the quality of care and increasing the life expectancy of patients. The proposed model in the present study can help gastroenterologist to improve the diagnosis accuracy, precision, and effective prediction of high-risk groups.