Screen: 5
Shanshan Tang, PhD
UT Southwestern Medical Center
Dallas, TX
Title: Progression Free Survival Prediction for Oligometastatic Renal Cell Cancer Patients after Stereotactic Ablative Radiation Therapy
Purpose/Objective(s): Stereotactic ablative radiotherapy (SAbR) has been demonstrated to be an attractive option for renal cell carcinoma (RCC) patients presenting with oligometastatic disease (omRCC). SAbR offers longitudinal disease control with favorable toxicity rates. However, not all patients will benefit from sequential SAbR and delayed systemic therapy (ST) due to progression of micrometastases. Therefore, patient selection is critical for the successful application of this strategy. The purpose of this work is to develop a progression-free survival (PFS) prediction model to identify omRCC patients who could benefit from SAbR.
Materials/
Methods: The study included 153 omRCC patients treated with curative intent SAbR. The PFS model was built on patients pre-SAbR clinical features. PFS was defined as the time period between the first SAbR treatment until initiation of ST or death. Features with predictive capability were identified using Hazard Ratio analysis and univariate Cox proportional hazard analysis with the training dataset. Spearman rank correlation analysis was applied to remove redundant features. Then, we constructed a multivariate Cox Proportional Hazard (CPH) model to predict PFS. Optimal features for generating a CPH model was selected one-by-one with a step forward feature selection strategy with C-index as the criteria. The optimal feature set for prediction was determined with the validation dataset, and applied to the testing dataset. Risk scores for each patient in testing dataset were generated and recorded. Five repeats of 5-fold cross-validation were conducted to mitigate the impact of random patient partition. The final risk scores, which was the average of the 5 repeats, was used to categorize patients into high- and low-risk groups. We used C-index and log rank test to assess the survival prediction model, and receiver operating characteristic (ROC) curve analysis to evaluate the performance of binary prediction for 1-year PFS prediction.
Results:The C-index for PFS prediction is 0.673 and the AUC for 1-year PFS being 0. 73. The most frequently selected features for PFS prediction include number of metastatic lesions, brain metastasis and bone metastasis. Kaplan-Meier analysis and log-rank test shows that our model is able to significantly differentiate high- and low-risk groups for disease progression (p<0.001).
Conclusion: The developed model utilizing the baseline clinical features is able to predict PFS in omRCC patients, which can be used to identify omRCC patients who are likely to benefit from local therapy with SAbR for longitudinal disease control with delayed ST. Abstract 3297 - Table 1: Summary of progression-free survival (PFS) prediction performance
Survival model | ||||
No. of patient | C-Index | |||
153 | 0.6730 | |||
Binary assessment of PFS at 1-year | ||||
No. of patient | AUC | Sensitivity | Specificity | Accuracy |
139 | 0.7274 | 0.6786 | 0.6145 | 0.6403 |