Comparative predictive performance of three machine learning algorithms for acute radiation enteritis risk among patients with cervical cancer undergoing radiotherapy: A prospective cohort study.
The Random Forest model showed excellent predictive performance for acute radiation enteritis in cervical cancer patients, but further validation is needed before clinical implementation.
Where it sits
this study against the rest of the lanreotide corpusSummary and findings
This study aimed to develop a machine learning-based risk prediction model for acute radiation enteritis (ARE) in patients with cervical cancer undergoing radiotherapy. The study included 386 patients and evaluated three models: Logistic Regression, Decision Tree, and Random Forest. The Random Forest model showed the highest predictive performance.
Abstract
<h4>Objective</h4>To develop a machine learning-based risk prediction model for acute radiation enteritis (ARE) in patients with cervical cancer, providing a new method for early and accurate prediction of ARE during radiotherapy.<h4>Methods</h4>This prospective study enrolled patients with cervical cancer undergoing radiotherapy from March 2024 to March 2025. The patients were randomly divided into training and test sets at a 7:3 ratio. Prediction models were constructed using Logistic Regression (LR), Decision Tree (DT), and Random Forest (RF) algorithms. Model performance was evaluated based on the area under the receiver operating characteristic curve (AUC), accuracy, precision, sensitivity, specificity, and F1-score.<h4>Results</h4>The incidence of ARE was 52.85% (204/386). Among the three models, the Random Forest model demonstrated the best performance, with an AUC of 0.961, sensitivity of 0.934, and F1-score of 0.905. These performance metrics were consistently higher than those of the LR (AUC, 0.860; sensitivity, 0.739; F1-score, 0.736) and DT (AUC, 0.910; sensitivity, 0.887; F1-score, 0.873) models. The RF model showed good clinical utility in effectively identifying high-risk patients for early intervention. Feature importance ranking derived from the RF model identified the parametrial dose, radiotherapy time, clinical stage, rectal V40, age, Platelet-to-Lymphocyte Ratio (PLR), concurrent chemotherapy, and hypertension as the most influential predictors, in descending order of importance.<h4>Conclusions</h4>The RF-based risk prediction model exhibited excellent performance in assessing the risk of ARE among patients with cervical cancer undergoing radiotherapy, thereby enabling individualized risk assessment and facilitating early preventive strategies.
Background
This paper addresses the need for accurate prediction of acute radiation enteritis (ARE) in cervical cancer patients undergoing radiotherapy. Prior studies have indicated that ARE can significantly impact treatment outcomes, but predictive models have been limited. This study is important as it explores machine learning algorithms to enhance risk prediction and potentially improve patient management.
Methods
This prospective cohort study enrolled patients with cervical cancer undergoing radiotherapy from March 2024 to March 2025. Patients were randomly divided into training and test sets at a 7:3 ratio. Three machine learning models were constructed: Logistic Regression, Decision Tree, and Random Forest. Model performance was evaluated using metrics such as AUC, accuracy, sensitivity, specificity, and F1-score.
Results
The incidence of acute radiation enteritis (ARE) was 52.85% (204/386). The Random Forest model demonstrated the best performance with an AUC of 0.961, sensitivity of 0.934, and F1-score of 0.905. In comparison, the Logistic Regression model had an AUC of 0.860, sensitivity of 0.739, and F1-score of 0.736, while the Decision Tree model had an AUC of 0.910, sensitivity of 0.887, and F1-score of 0.873.
Interpretation
The findings suggest that the Random Forest model outperforms traditional methods in predicting ARE risk, which aligns with previous literature highlighting the utility of machine learning in clinical predictions. However, while the AUC of 0.961 indicates strong statistical performance, the clinical significance of these findings remains to be fully validated in diverse populations. Limitations such as single-site data and potential overfitting must be considered when interpreting the results.
Key findings
- Incidence of ARE was 52.85% (204/386).
- Random Forest model AUC was 0.961, sensitivity was 0.934, F1-score was 0.905.
- Logistic Regression model AUC was 0.860, sensitivity was 0.739, F1-score was 0.736.
- Decision Tree model AUC was 0.910, sensitivity was 0.887, F1-score was 0.873.
Limitations
- Single-site study limits generalizability.
- Potential for overfitting in machine learning models.
- Short follow-up duration may not capture long-term outcomes.
- Not reported in abstract.