3098 - Clinical Outcome Modeling of Spatially Fractionated Radiation Therapy Treatments of Unresectable Large and Bulky Tumors using the Same-Day 3D MLC-Based Method

Purpose/Objective(s): The MLC-based SFRT becoming a useful clinical modality for unresectable large (> 8 cm) or radioresistant tumors while sparing skin. This study models clinical outcomes of SFRT treatments delivered using same-day 3D MLC-based crossfire method.

Materials/Methods: A total of 131 SFRT patients (median tumor volume, 361.7 cc maximum, 1500 cc) of different histopathology with follow up data (median interval 6 months) consisting of either tumor control, pain relief, toxicity, and overall survival (OS) were used. Utilizing 6/10MV beams and AcurosXB dose engine, 3D MLC-based SFRT treatments were planned on same-day CT scan and delivered treatment via CBCT guidance, bypassing the need for complex, time-intensive, inverse-planning and patient-specific QA. A single dose of 15 Gy prescription providing a peak to valley dose ratio of >3 while protecting surrounding critical structures was delivered. The linear-quadratic model was used to calculate biological effective dose (BED) metrics (D10%, D50%, D90%, D95%) and equivalent dose in 2 Gy per fraction (EQD2) from the combined effective dose of SFRT plus follow up treatments using an a/ß of 10 and 3 Gy for tumors and normal tissues, respectively. Combination therapy radiation doses were either palliative (median: 30 Gy in10 fraction) via 3D-conformal or curative (median: 60-70 Gy in 30-35 fractions) with VMAT plans depending on histopathology; typically starting 2-3 days after SFRT. Multivariable logistic regression was used to model tumor control probability (TCP) and pain relief using BED metrics. OS was modeled with an univariable Cox hazard model using BED metrics and area under curve (AUC). Toxicity was modeled with univariable logistic regression using maximum EQD2 to critical organs.

Results: TCP model (75/98 tumors reported clinical benefit) yielded an AUC of 0.667 with no dose metrics independently associated with tumor local control (p > 0.05). The pain relief model (59/82 reported benefit) had an AUC of 0.567, again with no statistically significant metrics. In the Cox model, increases in all dose metrics were associated with improved OS (p < 0.05). Seven patients presented with Grade 3+ events. Maximum EQD2 of skin was associated with skin redness (p = 0.014), and the parotid’s maximum EQD2 was associated with xerostomia (p = 0.002).

Conclusion: We present predictive models using clinical outcomes from a single institution’s experience with SFRT patients planned and treated on same-day via 3D MLC-based crossfire technique. These models provide a framework for identifying key metrics that may improve SFRT planning and delivery approach (in future) including re-irradiation and improve patient’s comfort and compliance. By utilizing large patient cohorts of different tumor histopathology from multi-institutional clinical trials, further refining of these SFRT predictive outcome models is warranted.