Predicting 5-year survival of colorectal carcinoma patients using data mining methods

Area of Science:

• Oncology
• Biostatistics
• Machine Learning

Background:

• Colorectal cancer (CRC) survival prediction is crucial for treatment planning.
• Accurate prognostic models are needed to improve patient outcomes.
• Data-mining techniques offer potential for developing robust predictive tools.

Purpose of the Study:

• To compare the predictive accuracy of neural network, decision tree, and logistic regression models.
• To assess the performance of these models in predicting 5-year survival in colorectal cancer patients.
• To identify the most effective data-mining approach for CRC survival prediction.

Main Methods:

• Utilized a database of colorectal cancer patient demographics and pathological features.
• Included expression levels of p53 and Bcl-2 biomarkers in the analysis.
• Applied and compared three distinct data-mining algorithms: neural network, decision tree, and logistic regression.

Main Results:

• All three models achieved acceptable accuracy, ranging from 64% to 70%.
• The neural network model exhibited the highest accuracy (70%) and specificity (80%), but the lowest sensitivity (59%).
• Logistic regression and decision tree models demonstrated comparable and higher sensitivity (72%) than the neural network.

Conclusions:

• Data-mining models show promise for predicting colorectal cancer survival.
• Model selection involves balancing accuracy, sensitivity, and specificity based on clinical needs.
• Further research may refine these models for enhanced clinical utility in CRC patient management.