MENU

Click here to

×

Are you sure ?

Yes, do it No, cancel
  1. Home
  2. Meeting Program
  3. All Sessions
  4. All Multi-Disciplinary General ePoster Discussion
  5. Predictive Modelling of Cardiac Events Due to Sub-Structure Exposure During Radiotherapy of Esophageal Cancer

Predictive Modelling of Cardiac Events Due to Sub-Structure Exposure During Radiotherapy of Esophageal Cancer

N Sasankan1*, C Cheng2 , K Men3 , H Geng4 , J Mohiuddin5 , A Barsky6 , H Fu7 , E Josef8 , R O'Quinn9 , J Zou10 , Y Xiao11 , (1) ,Philadelphia, ,(2) University of Pennsylvania, Philadelphia, PA, (3) University of Pennsylvania, Philadelphia, PA, (4) University of Pennsylvania, Philadelphia, PA, (5) ,Philadelphia, ,(6) University of Pennsylvania, Philadelphia, PA, (7) ,Philadelphia, ,(8) ,Philadelphia, ,(9) ,Philadelphia, ,(10) University of Pennsylvania, Philadelphia, PA, (11) University of Pennsylvania, Philadelphia, PA

Presentations

(Sunday, 7/14/2019) 4:00 PM - 4:30 PM

Room: Exhibit Hall | Forum 5

Purpose: Radiation related toxicity is one of the major concerns for long-term esophageal cancer survivors. Exposure of heart with its sub-structures to radiation may cause cardiac complications for cancer survivors. This can negatively affect the quality of life and overall survival. A predictive model built from dosimetry parameters of heart sub-structures may help us identify the key structures and associated dosimetry that are prognostic for post-treatment cardiac events.

Methods: A dataset of 81 patients with esophageal cancer undergoing radiotherapy was included in the study. The dosimetry of corresponding heart sub-structures was used to build the model to predict the occurrence of cardiac events. About 20 patients in our dataset reported cardiac events. A total of 190 variables was extracted for model building, far exceeding the number of patients and associated events. To reduce the dimension to prevent over fitting, two techniques were studied independently for feature selection, namely Multivariate Adaptive Regression Splines (MARS) and Minimum Redundancy Maximum Relevance (MRMR). RUSBoosted tree method, was applied to the feature selected form MRMR method. RUSBoosted and a quadratic SVM model was applied to the feature selected from MARS method.

Results: The RUSBoosted tree method had the following performance characteristics: accuracy, 72.8%, sensitivity, 75%; specificity 72% and AUC, 0.78. The RUSBoosted tree model had the following performance characteristics: accuracy, 74.1%; sensitivity, 65%; specificity. 77%, and AUC, 0.76. The SVM model exhibited the following performance characteristics: accuracy, 82.7%; sensitivity, 65%; specificity, 89%, and AUC, 0.85. All performance parameters were evaluated using 50-fold cross-validation.

Conclusion: We have demonstrated that models built from dosimetry parameters of heart sub-structures are prognostic for posttreatment cardiac events. With further refinement and validation, the models can potentially be used to aid clinical decision-making.

Funding Support, Disclosures, and Conflict of Interest: This project was supported by NCI grants U24CA180803(IROC), U10CA180868(NRG), and PA CURE grants.

Keywords

Not Applicable / None Entered.

Taxonomy

Not Applicable / None Entered.

Contact Email

Additional Links