Comparing the capacity of 3 mathematical models to forecast tumor volume dynamics and predict individual head and neck cancer patient responses to radiotherapy
Location
Moultrie, GA
Start Date
8-5-2024 1:00 PM
End Date
8-5-2024 4:00 PM
Description
BACKGROUND: Currently, planning of radiotherapy (RT) and assessment of response do not take into consideration tumor volume dynamics. If response to a particular RT schedule can be predicted accurately, then there is a potential for treatment adjustment. The objective of this study is to conduct an in-depth evaluation of the individual predictive capabilities of 3 mathematical models that describe tumor volume dynamics and patient response to radiation therapy in different ways.
METHODS: We combine two models of growth with two models of radiation therapy response to derive the 3 mathematical models: Model 1 - Exponential Growth with Direct Volume Reduction, Model 2 - Logistic Growth with Direct Volume Reduction, Model 3 - Logistic Growth with Carrying Capacity Reduction. All models were evaluated through our previously presented forecasting pipeline that combines new patient measurements with training data to forecast changes and tumor volume and use those changes to predict whether or not a patient will have locoregional recurrence of disease. Tumor volume data was collected from a cohort of head and neck cancer patients (n = 39) from Moffitt Cancer Center and MD Anderson Cancer Center that received 66-70 Gy RT in 2 Gy daily fractions. Tumor volume measurements were derived from CT scans: 2 before treatment and weekly scans during treatment.
RESULTS: The predictive capability of the models are measured using Youden's J-statistic, which ranges from -1 to 1 (perfect prediction) with negative values indicating a model that performs worse than random guessing. Model 3 (LOG+CCR) outperforms both Models 1 and 2 with a final J-statistic value of 0.597 for Locoregional Control (LRC) and 0.527 for Disease Free Survival (DFS). Model 1 (EXP+DVR) significantly outperforms Model 2 (LOG+DVR) in LRC prediction by 150%. However, Model 2 performs better than Model 1 in DFS prediction by 19.7%.
CONCLUSION: The ability to make accurate predictions regarding patient outcomes during a treatment cycle allows clinicians to make informed decisions on treatment adjustment and personalization. The differential performance of these 3 models in predicting patient outcomes with the same dataset indicates that the selecting the right underlying model is critical for accurate outcome prediction.
Embargo Period
5-23-2024
Comparing the capacity of 3 mathematical models to forecast tumor volume dynamics and predict individual head and neck cancer patient responses to radiotherapy
Moultrie, GA
BACKGROUND: Currently, planning of radiotherapy (RT) and assessment of response do not take into consideration tumor volume dynamics. If response to a particular RT schedule can be predicted accurately, then there is a potential for treatment adjustment. The objective of this study is to conduct an in-depth evaluation of the individual predictive capabilities of 3 mathematical models that describe tumor volume dynamics and patient response to radiation therapy in different ways.
METHODS: We combine two models of growth with two models of radiation therapy response to derive the 3 mathematical models: Model 1 - Exponential Growth with Direct Volume Reduction, Model 2 - Logistic Growth with Direct Volume Reduction, Model 3 - Logistic Growth with Carrying Capacity Reduction. All models were evaluated through our previously presented forecasting pipeline that combines new patient measurements with training data to forecast changes and tumor volume and use those changes to predict whether or not a patient will have locoregional recurrence of disease. Tumor volume data was collected from a cohort of head and neck cancer patients (n = 39) from Moffitt Cancer Center and MD Anderson Cancer Center that received 66-70 Gy RT in 2 Gy daily fractions. Tumor volume measurements were derived from CT scans: 2 before treatment and weekly scans during treatment.
RESULTS: The predictive capability of the models are measured using Youden's J-statistic, which ranges from -1 to 1 (perfect prediction) with negative values indicating a model that performs worse than random guessing. Model 3 (LOG+CCR) outperforms both Models 1 and 2 with a final J-statistic value of 0.597 for Locoregional Control (LRC) and 0.527 for Disease Free Survival (DFS). Model 1 (EXP+DVR) significantly outperforms Model 2 (LOG+DVR) in LRC prediction by 150%. However, Model 2 performs better than Model 1 in DFS prediction by 19.7%.
CONCLUSION: The ability to make accurate predictions regarding patient outcomes during a treatment cycle allows clinicians to make informed decisions on treatment adjustment and personalization. The differential performance of these 3 models in predicting patient outcomes with the same dataset indicates that the selecting the right underlying model is critical for accurate outcome prediction.
Comments
Winner of PCOM South Georgia Research Day 2024 Best Original Research award.