Title: Dosimetric Variables Predictive of Local Control Following Stereotactic Radiosurgery (SRS) for Non-Small Cell Lung Cancer (NSCLC) Brain Metastasis
Abstract: Local control (LC) is a challenging problem in patients treated with SRS for brain metastasis, with scant data suggesting actionable dosimetric variables to improve outcomes. We evaluated the impact of various patient, tumor, and dosimetric variables on LC of NSCLC brain metastases with SRS. Patients with NSCLC brain metastases treated with single fraction SRS without prior metastasectomy or SRS to the same lesion were identified from an IRB approved SRS registry. Prescription dose was in general per RTOG 90-05 criteria, with adjustments at the discretion of the treating MD. All brain MRI imaging was reviewed retrospectively from the time of SRS until death or last follow-up. LC was graded in each treated lesion and was defined as stability or decrease in the size of the treated metastasis at any time post-SRS. Patient demographics, tumor descriptors, and basic treatment parameters were obtained through retrospective chart review. Dose volume histograms were collected for each lesion and evaluated with a custom dose-volume histogram analysis tool. V18-V50Gy, D10, D50, D90, dose mean, maximum, minimum, and median were extracted as were dose gradient index (DGI) and RTOG conformity index (CI). Univariate and multivariate Cox Regression were used to identify factors predictive of LC, and LC was estimated using the Kaplan-Meier method. Optimal cut point for continuous variables were determined using a combination of graphical diagnostic plots, minimum p-value approach with adjusted p-values, and receiver operating characteristic (ROC) curves. One hundred forty-one brain metastases in 60 patients were identified. Median # of metastases per patient was 2.33, and median tumor volume was 279 mm3. On univariate analysis (UVA), squamous histology, tumor volume, number of shots, V18-V30, D50, D90, dose mean, median, and minimum as well as DGI and CI were predictive of local control (p<0.05). On multivariate analysis (MVA), only increasing tumor volume and decreasing CI were independent predictors of poor LC after GKRS (p<0.05), while V30 strongly trended towards significance (p = 0.06). Prescription dose was not predictive of LC. 1-year local control overall was 69%. Above and below cut points of 1.52 for CI, 463 mm3 for volume, and 23% for V30, 1-year local control was 98 vs 59% (p<0.0001), 100 vs 61% (p<0.0001), and 93 vs 75% (p = 0.012). A large potential for variation in plan quality exists for a given prescription dose for SRS, and prescription dose is not predictive of LC. Our results suggest that dosimetric variables such as CI and V30 may represent actionable metrics to drive treatment planning during SRS for brain metastases. Continued analysis of a larger number of lesions within a recently initiated multi-institutional registry (CONDR-IMD) will further clarify such parameters.