*3.1. Clinical Outcomes*

All 36 patients were referred for SABR after discussion in a multidisciplinary tumor board, and reasons for referral included a high surgical risk due to comorbidity (*n* = 9), which is unsuitable for other ablative therapies due to tumor size (*n* = 10) or location (*n* = 5), patient preference (*n* = 5), co-existing second malignancy (*n* = 3), use of anti-coagulants (*n* = 2), and chronic stage ≥IV kidney disease (*n* = 2). Baseline patient characteristics are summarized in Table 2. The mean age of this cohort was 78.1 years with a preponderance of men (66.7%). The mean tumor diameter was 5.6 cm (range 2.4–9.3 cm) with 86.1% of tumors measuring ≥4 cm in the largest dimension of which 23 patients have a cT1b tumor and 8 patients have a cT2a tumor. Five patients (13.9%) had metastasized renal cell carcinoma (RCC) at the time of diagnosis. Pathologic confirmation of RCC before treatment was achieved in approximately half of patients (55.6%) of which the majority was diagnosed with Fuhrman grade 2 (*n* = 14). Other patients with histology included Fuhrman grade 1 (*n* = 1), Fuhrman grade 3 (*n* = 1), a RCC with sarcomatoid features (*n* = 1), and a chromophobe tumor (*n* = 1). In two patients, no grading was available because pathologic confirmation was obtained from systemic metastases. All patients were able to complete adaptive MRgRT with an average time per fraction of 45 min. An overview of the average duration of the different components of adaptive MRgRT for RCC is shown in Figure 2. Three patients completed treatment while tracking on the kidney instead of the tumor.

**Figure 2.** Pie-chart of the average duration of the different components of breath-hold gated adaptive MR-guided radiotherapy with an average time per fraction of 45 min.


**Table 2.** Baseline patient characteristics (*n*=36). Abbreviations: RCC =renal cell carcinoma, GTV =gross tumor volume, PTV = planning target volume, CKD = chronic kidney disease.

The median follow-up was 16.4 months. Overall survival was 91.2% at one year (Figure 3), LC was 95.2% (Figure 3), and freedom from any progression was 91% at one year. Two patients had local recurrences. One patient had progressive distant disease at recurrence for which systemic therapy was delivered, and the second patient with an isolated local recurrence underwent radiofrequency ablation as salvage. Treatment-related acute toxicity grade ≥ 2 in the form of nausea was observed in a single patient, which responded to oral ondansetron. No other acute or late grade ≥2 toxicity was reported. The mean eGFR at baseline was 55.3 (SD <sup>±</sup>19.0) mL/min/1.73 m2. With a mean interval of 16 months and mean eGFR post-MRgRT was 49.3 (SD <sup>±</sup> 19.1) mL/min/1.73 m2, which indicates a decrease of 6.0 mL/min/1.73 m2. No patient in this cohort required dialysis during follow-up.

**Figure 3.** Kaplan-Meier plots for overall survival (left) and local control (right).

#### *3.2. The Need for Daily Plan Re-Optimization*

In 151 out of 180 fractions (83.9%), the predicted plans (without re-optimization) met all institutional target and OAR constraints. In these fractions, predicted and re-optimized plans were of similar quality with a mean GTV V38Gy of 98.8% and 99.1%, respectively, and mean V33Gy of 0 cc for both stomach, duodenum, and bowel. In the other 29 fractions, predicted plans were suboptimal with insufficient GTV coverage in two out of 180 fractions (1.1%) exceeding OAR constraints in 25 fractions (13.9%), and both insufficient GTV coverage and exceeded OAR constraints in another two fractions (1.1%). There was no significant difference in suboptimal predicted plans for left-sided or right-sided RCC (*p* = 0.56). For these suboptimal plans, on-couch re-optimization corrected the GTV V38Gy from a mean of 88.7% (predicted) to 97.4% (re-optimized). Similarly, re-optimization corrected OAR V33Gy ≤ 1 cc violations from on average V33Gy of 4.1 (predicted plans) to 0.3 cc (re-optimized plans). Analysis on a patient basis showed that the 29 insufficient predicted fractions were distributed among 11 patients (11/36, 30.6%). However, three or more suboptimal fractions were seen in only five patients (13.9%).

Decision tree analysis identified the baseline OAR V25Gy (combined structure of stomach, bowel, and duodenum) as the most significant predictor variable for daily adaptive planning needs with 0.5 cc as an optimal cut-off value (*p* < 0.001). In all cases with a baseline OAR V25Gy of ≤ 0.5 cc, plan adaptation was redundant as the predicted plans already complied with institutional constraints. In patients with baseline OAR V25Gy of more than 0.5 cc, plan re-optimization was needed in 32.2% of fractions in order to fulfill the preset target coverage and OAR constraints (Table 3). The correct classification rate of the decision tree was 86.1% with a sensitivity of 100% and a specificity of 67.7%. The difference between re-optimized and predicted dose parameters for target (GTV V95%) and OAR (V33Gy) stratified for split group 1 and 2 (Table 3) is shown in Figure 4.

**Table 3.** Results in the Chi-square automatic interaction detection (CHAID) tree table.


**Figure 4.** Difference of DVH parameters. Boxplots showing the relative volume difference in GTV V95% (%) and absolute difference in OAR V33Gy (cc) of the re-optimized compared to the predicted plans stratified for Split group 1 (re-optimization not needed) and 2 (re-optimization needed). Abbreviations: DVH = dose volume histogram, GTV = gross target volume, OAR = organs at risk.
