Machine learning-based prognostic model for patients with anaplastic thyroid carcinoma

Abstract Objective Despite the identification of various prognostic factors for anaplastic thyroid carcinoma (ATC) patients over the years, a precise prognostic tool for these patients is still lacking. This study aimed to develop and validate a prognostic model for predicting survival outcomes for...

Full description

Saved in:
Bibliographic Details
Main Authors: Yihan Sun, Da Lin, Xiangyang Deng, Yinlong Zhang
Format: Article
Language:English
Published: Springer 2025-01-01
Series:Discover Oncology
Subjects:
Online Access:https://doi.org/10.1007/s12672-024-01703-9
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1832585648472588288
author Yihan Sun
Da Lin
Xiangyang Deng
Yinlong Zhang
author_facet Yihan Sun
Da Lin
Xiangyang Deng
Yinlong Zhang
author_sort Yihan Sun
collection DOAJ
description Abstract Objective Despite the identification of various prognostic factors for anaplastic thyroid carcinoma (ATC) patients over the years, a precise prognostic tool for these patients is still lacking. This study aimed to develop and validate a prognostic model for predicting survival outcomes for ATC patients using random survival forests (RSF), a machine learning algorithm. Methods A total of 1222 ATC patients were extracted from the Surveillance, Epidemiology, and End Results (SEER) database and randomly divided into a training set of 855 patients and a validation set of 367 patients. We developed an RSF model and a traditional Cox model using the training cohort and further compared their performance based on calibration and discrimination. integrated brier score (iBS) was used to estimate the calibration ability. The Brier score, C-index value, the receiver operating characteristic (ROC) curve with the area under the curve (AUC) and Decision Curves Analysis (DCA) were evaluated. Furthermore, we assessed the feature importance within the RSF model and validated its performance using the validation group. Results An RSF model and a traditional Cox model were successfully developed in training set. The Brier score for the RSF model was 0.055, which is lower than the Cox model’s score of 0.063, indicating better performance since a lower Brier score signifies superior model accuracy. The RSF model exceeded the Cox model in performance based on the C-index and AUC. Additionally, the DCA curve indicated that the RSF model provided substantial clinical benefit. And we further ranked the time-dependent features according to their permutation importance and observed that surgery, radiotherapy, and chemotherapy were the most influential predictors initially. Moreover, according to the RSF model predictions, the ATC patients were successfully stratified into 2 prognostic groups displaying significant difference in survival. Conclusions This prognostic study first revealed that RSF offers more precise overall survival predictions and superior prognostic stratification compared to the Cox regression model for ATC patients.
format Article
id doaj-art-890b30259aa84857a2c20f1499f9dcdb
institution Kabale University
issn 2730-6011
language English
publishDate 2025-01-01
publisher Springer
record_format Article
series Discover Oncology
spelling doaj-art-890b30259aa84857a2c20f1499f9dcdb2025-01-26T12:39:49ZengSpringerDiscover Oncology2730-60112025-01-0116111010.1007/s12672-024-01703-9Machine learning-based prognostic model for patients with anaplastic thyroid carcinomaYihan Sun0Da Lin1Xiangyang Deng2Yinlong Zhang3Department of Thyroid Breast Surgery, The Second Affiliated Hospital and Yuying Children’s Hospital of Wenzhou Medical UniversityDepartment of General Surgery, Hangzhou TCM Hospital Affiliated to Zhejiang Chinese Medical UniversityDepartment of Neurosurgery, The Second Affiliated Hospital and Yuying Children’s Hospital of Wenzhou Medical UniversityDepartment of Thyroid Breast Surgery, The Second Affiliated Hospital and Yuying Children’s Hospital of Wenzhou Medical UniversityAbstract Objective Despite the identification of various prognostic factors for anaplastic thyroid carcinoma (ATC) patients over the years, a precise prognostic tool for these patients is still lacking. This study aimed to develop and validate a prognostic model for predicting survival outcomes for ATC patients using random survival forests (RSF), a machine learning algorithm. Methods A total of 1222 ATC patients were extracted from the Surveillance, Epidemiology, and End Results (SEER) database and randomly divided into a training set of 855 patients and a validation set of 367 patients. We developed an RSF model and a traditional Cox model using the training cohort and further compared their performance based on calibration and discrimination. integrated brier score (iBS) was used to estimate the calibration ability. The Brier score, C-index value, the receiver operating characteristic (ROC) curve with the area under the curve (AUC) and Decision Curves Analysis (DCA) were evaluated. Furthermore, we assessed the feature importance within the RSF model and validated its performance using the validation group. Results An RSF model and a traditional Cox model were successfully developed in training set. The Brier score for the RSF model was 0.055, which is lower than the Cox model’s score of 0.063, indicating better performance since a lower Brier score signifies superior model accuracy. The RSF model exceeded the Cox model in performance based on the C-index and AUC. Additionally, the DCA curve indicated that the RSF model provided substantial clinical benefit. And we further ranked the time-dependent features according to their permutation importance and observed that surgery, radiotherapy, and chemotherapy were the most influential predictors initially. Moreover, according to the RSF model predictions, the ATC patients were successfully stratified into 2 prognostic groups displaying significant difference in survival. Conclusions This prognostic study first revealed that RSF offers more precise overall survival predictions and superior prognostic stratification compared to the Cox regression model for ATC patients.https://doi.org/10.1007/s12672-024-01703-9Anaplastic thyroid carcinomaTraditional Cox modelRandom survival forestsPrognosisSurvival prediction
spellingShingle Yihan Sun
Da Lin
Xiangyang Deng
Yinlong Zhang
Machine learning-based prognostic model for patients with anaplastic thyroid carcinoma
Discover Oncology
Anaplastic thyroid carcinoma
Traditional Cox model
Random survival forests
Prognosis
Survival prediction
title Machine learning-based prognostic model for patients with anaplastic thyroid carcinoma
title_full Machine learning-based prognostic model for patients with anaplastic thyroid carcinoma
title_fullStr Machine learning-based prognostic model for patients with anaplastic thyroid carcinoma
title_full_unstemmed Machine learning-based prognostic model for patients with anaplastic thyroid carcinoma
title_short Machine learning-based prognostic model for patients with anaplastic thyroid carcinoma
title_sort machine learning based prognostic model for patients with anaplastic thyroid carcinoma
topic Anaplastic thyroid carcinoma
Traditional Cox model
Random survival forests
Prognosis
Survival prediction
url https://doi.org/10.1007/s12672-024-01703-9
work_keys_str_mv AT yihansun machinelearningbasedprognosticmodelforpatientswithanaplasticthyroidcarcinoma
AT dalin machinelearningbasedprognosticmodelforpatientswithanaplasticthyroidcarcinoma
AT xiangyangdeng machinelearningbasedprognosticmodelforpatientswithanaplasticthyroidcarcinoma
AT yinlongzhang machinelearningbasedprognosticmodelforpatientswithanaplasticthyroidcarcinoma