(Sunday, 7/12/2020) [Eastern Time (GMT-4)]
Room: AAPM ePoster Library
Purpose:
Precise determination of Dose Volume Histogram (DVH) indices most predictive of toxicity is difficult because of multiple comparisons concerns. We develop a DVH analysis technique which determines the threshold dose with high precision while minimizing multiple comparisons concerns.
Methods:
A database of 79 prostate cancer patients treated with radiation therapy was developed. Patients were treated with IMRT to 77.4Gy in 43 fractions, with simultaneous, integrated boost to areas of disease identified on multi-parametric MRI to 81-83Gy. Grade >= 2 acute rectal toxicity in these patients was scored by experienced physicians using CTCAE v4.0 criteria. DVHs extracted from treatment plans were used to model the likelihood of acute rectal toxicity. A multivariate logistic regression model was constructed using an array of equally spaced V%_D indices (percentage of volume receiving dose D, or greater) with a step of 1Gy. The model was augmented with the Fused Lasso Operator to compensate for correlations between V%_D indices. Area Under the Curve (AUC) and fit quality (p-value) were used to compare the performance of the multivariate model with the performance of a family of univariate logistic regression models which used a single V%_D index.
Results:
20% of patients developed grade >=2 acute rectal toxicity. The best AUC value achieved by a univariate logistic regression model was 0.67, while the best fit quality p value was 0.035. A multivariate logistic regression model with Fused Lasso operator showed two distinct dose thresholds, at 34Gy and 77.5Gy (2Gy dose equivalent). V%_D indices above these thresholds become predictive of rectal toxicity. The fit quality p value for the multivariate model was equal to 0.008, while AUC of the model was 0.68.
Conclusion:
Fused Lasso technique can be used in DVH analysis to precisely determine dose thresholds predictive of toxicity in radiation therapy.
Keywords
Prostate Therapy, Risk, ROC Analysis
Taxonomy
IM- Radiation Dose and Risk: Models
Contact Email
