Presenter(s)
A. U. Kishan1, K. McGreevy2, L. Valle1, J. E. Juarez Casillas3, K. Flores4, K. Taparra5, J. Nikitas6, J. M. Lamb7, E. Rivera8, D. Telesca9, M. L. Steinberg1, and J. B. Weidhaas1; 1Department of Radiation Oncology, David Geffen School of Medicine, University of California, Los Angeles, Los Angeles, CA, 2MiraDx, Los Angeles, CA, 3Department of Radiation Oncology, University of California, Los Angeles, Los Angeles, CA, 4UCLA, Los Angeles, CA, 5Department of Radiation Oncology, Stanford University School of Medicine, Stanford, CA, 6Department of Radiation Oncology, University of Pennsylvania, Philadelphia, PA, 7UCLA, Department of Radiation Oncology, Los Angeles, CA, 8UCLA, los angeles, CA, 9Department of Biostatistics, Fielding School of Public Health, University of California, Los Angeles, Los Angeles, CA
Purpose/Objective(s):
Late grade =2 genitourinary (GU) adverse events (AEs) limits broader adoption of post-operative radiotherapy despite established oncologic benefit. Germline biomarkers may enable individualized counseling and risk-adaptive planning. PROSTOX Ultra (PROSTOXu), a 32–mirSNP based signature, predicts late GU AEs after intact-prostate SBRT using a prespecified 0.50 threshold. Its performance post-prostatectomy is unknown. We hypothesized that genetic risk ordering would persist, but the clinical decision threshold might shift. We thus evaluated PROSTOXu in two prospective post-op SBRT cohorts (SCIMITAR, EXCALIBUR), comparing continuous and dichotomized discrimination and exploring threshold recalibration.
Materials/Methods:
SCIMITAR (NCT03541850) enrolled 100 post-prostatectomy patients treated with 30–34 Gy in 5 fractions ± pelvic nodes; 89 were evaluable. EXCALIBUR (NCT04915508) included 67 evaluable patients treated with 5-fraction SBRT to the prostate bed ± nodes. Germline DNA (saliva) isolation and PROSTOXu testing was performed in a CLIA-certified lab, and risk scores were generated using the original intact-prostate model. Discrimination was assessed by continuous AUC and dichotomized performance at the 0.50 threshold. Score distributions were compared by Wilcoxon testing. Exploratory recalibration identified thresholds maintaining sensitivity =0.65 and specificity =0.70. AUCs were bootstrapped (1,000 iterations) and pooled by fixed-effects meta-analysis. Logistic regression evaluated clinical covariates.
Results:
In SCIMITAR, 21/89 (23.6%) developed late grade =2 GU AEs. At the 0.50 threshold, binary AUC was 0.755 with high specificity (0.985) but limited sensitivity (0.546). In EXCALIBUR, dichotomized AUC was <0.50 despite high specificity (0.926), indicating threshold misalignment. Continuous discrimination was strong in both cohorts (EXCALIBUR AUC 0.824; SCIMITAR 0.814), substantially exceeding binary AUCs and demonstrating preserved risk ordering. PROSTOXu scores were significantly higher among patients with toxicity (EXCALIBUR median 0.338 vs 0.043, p=0.009; SCIMITAR 0.547 vs 0.049, p=1.1×10?5). Recalibration identified candidate thresholds ~0.10–0.31; 0.25 was selected. At 0.25, dichotomized AUC improved (EXCALIBUR 0.727; SCIMITAR 0.750) with pooled AUC 0.748, supporting a stable genetic signal with a downward-shifted clinical trigger post-operatively. Age, baseline pad use, IPSS, and time from prostatectomy to SBRT were not associated with toxicity (all p>0.2).
Conclusion:
Across two independent post-operative SBRT cohorts, PROSTOX Ultra preserved biological discrimination as a continuous predictor, while binary performance depended on threshold selection. A lower post-op threshold (0.25) improved discrimination without redeveloping the signature, indicating that mirSNP-based germline risk stratification generalizes post-prostatectomy but requires context-specific calibration.
