Volume 5, Issue 1 (6-2022)
2022, 5(1): 0-0 |
Back to browse issues page
Ethics code: IR.MEDILAM.REC.1398.071
Download citation:
BibTeX | RIS | EndNote | Medlars | ProCite | Reference Manager | RefWorks
Send citation to:
BibTeX | RIS | EndNote | Medlars | ProCite | Reference Manager | RefWorks
Send citation to:
Deldar M, Sayehmiri K, anbiaee R, Jalilian A. Predicting Ovarian Cancer Survival Times: Comparison Of Parametric Methods With Random Survival Forests. Journal title 2022; 5 (1)
URL: http://newresearch.medilam.ac.ir/article-1-540-en.html
URL: http://newresearch.medilam.ac.ir/article-1-540-en.html
Ilam University of Medical Sciences
Abstract: (1137 Views)
Background and Aim: Predicting survival time has many effective consequences in managing the quality of life for the rest of the patientchr('39')s life. On the other hand, survival data are highly variable and make accurate predictions difficult. Random survival forests by repeating tree construction and averaging the results of these trees cause reducing the prediction error and further generalizability of these results. This thesis compares the prediction error of the random survival forest model with the Cox and Weibull models in predicting the time to the first recurrence in patients with epithelial ovarian cancer.
Materials and Methods: In this study, the data of 141 patients with ovarian cancer who referred to Imam Hossein Hospital in Tehran from 2008 to the end of 2018 were used. Cox regression, Weibull model, classification and regression tree, and random survival forest were fitted to the data to investigate the factors affecting the first recurrence of patients, respectively. Finally, using the C-Index and Brier score, the prediction error of these models were compared.
Results: According to the results of random survival forest, metastatic tumor using the criterion (VIMP), with relative importance of 2.665 and minimal depth (MD) 2.349, tumor stage with relative importance of 1.993 and depth of 2.678, and maximum platelet count with relative importance of 2.132 and depth of 2.683 were effective variables. According to the Brier score, the random survival forest prediction error was 0.16 and the Cox model was 0.24. The C-Index error in the random survival forest was 0.34 and the Cox model was 0.42. Brier scores for the Cox and Weibull models were calculated to be approximately the same, so the random survival forest prediction error of the less than both Cox and Weibull models.
Conclusion: Unlike classical methods, the random survival forest without the need for special presuppositions with less predictive error can well explain the variables of the response variable when exposed to high-dimensional data.
Materials and Methods: In this study, the data of 141 patients with ovarian cancer who referred to Imam Hossein Hospital in Tehran from 2008 to the end of 2018 were used. Cox regression, Weibull model, classification and regression tree, and random survival forest were fitted to the data to investigate the factors affecting the first recurrence of patients, respectively. Finally, using the C-Index and Brier score, the prediction error of these models were compared.
Results: According to the results of random survival forest, metastatic tumor using the criterion (VIMP), with relative importance of 2.665 and minimal depth (MD) 2.349, tumor stage with relative importance of 1.993 and depth of 2.678, and maximum platelet count with relative importance of 2.132 and depth of 2.683 were effective variables. According to the Brier score, the random survival forest prediction error was 0.16 and the Cox model was 0.24. The C-Index error in the random survival forest was 0.34 and the Cox model was 0.42. Brier scores for the Cox and Weibull models were calculated to be approximately the same, so the random survival forest prediction error of the less than both Cox and Weibull models.
Conclusion: Unlike classical methods, the random survival forest without the need for special presuppositions with less predictive error can well explain the variables of the response variable when exposed to high-dimensional data.
Received: 2019/04/6 | Accepted: 2019/07/10 | Published: 2022/06/19
Send email to the proposal executer
| Rights and permissions | |
 | This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License. |
