Related Experiment Video
Updated: Sep 17, 2025
08:52
Profiling Sensitivity to Targeted Therapies in EGFR-Mutant NSCLC Patient-Derived Organoids
Published on: November 22, 2021
4.2K
Multiomic based Bayesian network toxicity modeling for simultaneous prediction of multiple toxicity outcomes in NSCLC
Saurabh S Nair1, Ramon M Salazar1, Ting Xu2
1Departments of Radiation Physics and Thoaracic Radiation Oncology, The University of Texas MD Anderson Cancer Center, Houston, TX, USA.
Summary
A new Bayesian network model accurately predicts both radiation pneumonitis and radiation esophagitis in non-small cell lung cancer patients. This approach integrates multiomic data for improved toxicity prediction in radiotherapy.
Area of Science:
- Oncology
- Radiotherapy
- Bioinformatics
- Medical Physics
Background:
- Radiation Pneumonitis (RP) and Radiation Esophagitis (RE) are significant dose-limiting toxicities in non-small cell lung cancer (NSCLC) radiotherapy.
- Accurate prediction of these toxicities is crucial for optimizing treatment plans and patient outcomes.
Purpose of the Study:
- To develop a multi-objective Bayesian network (BN) model for simultaneous prediction of RP and RE.
- To integrate dose-volume histograms (DVH), clinical, and multiomic features for enhanced predictive accuracy.
Main Methods:
- A cohort of 179 NSCLC patients was analyzed, with toxicity data collected per CTCAE v5.0.
- A parallel dimensionality reduction technique with L1-norm penalty and penalized logistic regression was used for feature selection from 672 extracted features.
- A score-based structure learning algorithm (Tabu search) built the BN model, with a 70%/30% train/test split for cross-validation.
Main Results:
- The BN model achieved an Area Under the Curve (AUC) of 0.86 and Area Under the Precision-Recall Curve (AUPRC) of 0.76 for RP prediction on the test set.
- For RE prediction, the model yielded an AUC of 0.81 and AUPRC of 0.50.
- Joint prediction performance, evaluated by Free-response AUC (FRAUC), was 0.74 on the test set.
Conclusions:
- A highly predictive multi-objective BN model was successfully developed.
- The model effectively utilizes multiomic covariates to simultaneously predict RP and RE.
- This advancement offers a promising tool for managing key toxicities in NSCLC radiotherapy.
