Body Composition-Integrated Random Survival Forest Model for Predicting Overall Survival in Locally Advanced Cervical Cancer Treated With Definitive Chemoradiotherapy

Sep 15, 2026

Study integrates body composition data to build a random survival forest (RSF) predictive model for overall survival (OS) in locally advanced cervical cancer (LACC) patients receiving definitive chemoradiotherapy (CRT). Retrospective enrollment included 374 LACC patients from the First Affiliated Hospital of Soochow University (2010–2018). Eleven features were selected with LASSO, while Optuna optimized RSF hyperparameters. RSF outperformed Cox regression and Deepsurv by C-index and AUC, with favorable Brier scores and decision curve analysis supporting good calibration and clinical benefit. SHAP highlighted skeletal muscle density as a leading contributor, and risk stratification showed strong performance.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages22

Reading Time55 min

AccessFree to download

Methods

Patients and data source

Feature selection and model training

Model evaluation and comparisons

Results

Predictive performance

Calibration and clinical utility

Feature importance and risk stratification

Background

Prognosis and treatment context

Rationale for body composition and machine learning

The First A liated Hospital of Soochow University The First A liated Hospital of Soochow University The First A liated Hospital of Soochow University The First A liated Hospital of Soochow University The First A liated Hospital of Soochow University The First A liated Hospital of Soochow University The First A liated Hospital of Soochow University The First A liated Hospital of Soochow University Cervical cancer, body composition, random survival forest, machine learning, overall survival, predictive model April 25th, 2024 [https://doi.org/10.21203/rs.3.rs-4268512/v1](https://doi.org/10.21203/rs.3.rs-4268512/v1) 􀁱 􀅏 This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License No competing interests reported. This study aims to integrate body composition data in developing a random survival forest (RSF) model for predicting overall survival (OS) in locally advanced cervical cancer (LACC) patients treated with de«nitive chemoradiotherapy (CRT) . 374 LACC patients from the First A liated Hospital of Soochow University were enrolled in the study. Eleven features were selected using least absolute shrinkage and selection operator (LASSO) regression in constructing the RSF model. The Optuna algorithm was applied to optimize the hyperparameters. The constructed RSF model was evaluated based on the C-index, AUC, Brier scores, and decision curve analysis (DCA) . Furthermore, we compared the RSF model with the traditional Cox regression and deep learning (Deepsurv) models. The C-index of the RSF model in the test set was 0.809 (95%CI: 0.708–0.911), superior to the Cox regression (0 .739, 95%CI: 0 .627–0.852) and Deepsurv (0 .774, 95%CI: 0 .656–0.892) models. The Brier scores for 1-, 3-, and 5-year OS in the test set (0 .044, 0 .069, and 0 .096) and DCA demonstrated that the RSF model has good calibration and clinical bene«ts. Furthermore, skeletal muscle density (SMD) ranked top in visualization using SHapley Additive exPlanations (SHAP) plotting. Lastly, we applied the RSF model in risk strati«cation, which showed a favorable performance. The RSF model, which is based on body composition characteristics, performs well in predicting OS in LACC patients treated with de«nitive CRT. Moreover, the RSF model demonstrates value in risk strati«cation. Cervical cancer is a common gynecological malignant tumor with an incidence rate second only to breast cancer, with approximately 604,127 new cases occurring annually around the world [ 1, 2] . As the fourth leading cause of cancer death, the 5-year OS rate of cervical cancer is below 70% and is lower in cases of locally advanced cervical cancer (LACC) [3, 4] . Concurrent chemoradiotherapy (CRT) is the «rstline treatment option for patients with locally advanced cervical cancer (LACC) and has been demonstrated to improve the 5-year OS rate by 6% compared with radiotherapy alone [5] . Recently, several molecular imaging biomarkers have been shown to predict outcomes in cervical cancer at baseline and/or early in treatment. This facilitates customized treatment plans, patient follow-up, and adjuvant therapy [6] . Body composition is described as the percentages of fat, bone, and muscle in human bodies. Cachexiain cancer patients will result in changes in body composition, including reduced muscle mass with or without the loss of adipose tissue [7] . Changes in body composition have already been found to remarkably impact on the survival, quality of life, postoperative complications, and treatment response of cancer patients [8–10] . Skeletal muscle index (SMI) during concurrent chemoradiotherapy was found to be associated with overall survival and cancer-speci«c survival in LACC [11] . Subcutaneous adipose tissue (SAT) and visceral adipose tissue (VAT) have been measured and found to be meaningful for predicting the outcomes of patients with breast cancer, prostate cancer, and pancreatic cancer, among others [ 1

FAQs

What data and imaging basis are used to build the prediction model?

The model integrates body composition characteristics derived from imaging-related body composition measurements, including muscle and adipose tissue components. These variables are used to construct features for survival prediction.

How was the random survival forest (RSF) model constructed and optimized?

Eleven features were selected using LASSO regression. The Optuna algorithm was applied to optimize RSF hyperparameters before training the final model.

How does the RSF model perform compared with Cox regression and Deepsurv?

On the test set, the RSF model achieved a higher C-index than Cox regression and Deepsurv. It also showed good discrimination and calibration supported by AUC, Brier scores, and decision curve analysis.

Which body composition feature was most important in the model?

Skeletal muscle density (SMD) ranked top based on SHAP visualization, indicating strong contribution to the model’s predictions.

Sep 15, 2026

Uploaded by Seraphina