Comparison of Survival Estimation Methods in the Analysis of Breast Cancer Data

Abstract

Objective: In many epidemiological studies or clinical trials, Cox Proportional Hazard model is the most commonly used technique for the analysis of effects of covariates on survival time when time are continuous and ties are present. It is important to handle the presence of ties in the data to prevent the inherent effect which may lead to unreliable result in the analysis. Therefore, this study was carried out handle the inadvertent inclusion of ties in survival time with three different estimation methods; Breslow, Efron and exact partial likelihood estimation methods of Cox regression model. Material and Methods: Analysis were carried out using data collected on 300 breast cancer patients from University of Ilorin Teaching Hospital to evaluate the factors affecting the survival of patients with breast cancer. The estimation methods used in handling ties in the breast cancer data are compared using Akaike Information Criterion. All analyses were performed using STATA version 14.0. Result: Of a total of 300 patients, 97 (32.33%) died of breast cancer. The result of the data analysis using Cox model with different estimation methods were approximately alike. The risk of dying from the breast cancer was associated with age of the patients; patients with Cytological diagnosis had an increased risk for death. Conclusion: The hazard ratios of the estimators are approximately similar but based on Akaike Information Criteria, exact partial likelihood outperformed the other methods of handling ties in breast cancer data since it has the smallest value.