Clinical results of mean GTV dose optimized robotic guided SBRT for liver metastases

This retrospective analysis was approved by the respective ethics committees of the treating centers. Between March 2011 and July 2015, 52 patients with a total of 91 metastatic liver lesions (one to four metastases per patient) were treated with SBRT using the CyberKnife real-time tracking system at two centers performing radiosurgery (Table 1). Three patients had two and three patients had three repeat SBRT procedures for new liver metastases which developed during follow-up, and were included in the analysis. All patients’ primary tumors were controlled at the time of SBRT treatment and SBRT was chosen as first therapeutic option at initial diagnosis of liver metastases in 26.9 %, at the diagnosis of recurrent metastases after local therapy in 9.6 %, or at the diagnosis of new metastases after systemic failure in 63.5 % of the patients. Furthermore, 63.5 % of the patients had no extrahepatic disease, while 22.8 % had stable extrahepatic disease, and 13.7 % underwent simultaneous treatment of extrahepatic disease at the time of SBRT, mostly consisting of lung, bone or lymph node metastases. All patients were considered oligometastatic having less than five metastatic sites. Median baseline Karnofsky Index was 90 % (range, 60–100 %) and median GTV volume was 12 cc (min 1 cc and max 372 cc).

Prior to treatment either GoldAnchor™ (Naslund Medical AB, Huddinge, Sweden) or solid gold fiducial markers (IZI Medical Products, Owings Mills, MD, USA) were implanted as close to the lesion as possible with computer tomography (CT) or ultrasound (US) guidance. Depending on the size, shape and number of the lesions, one (e.g., for a small spherical lesion) to five (e.g., for multiple lesions) fiducials were implanted.

Treatment planning was performed on standard non-contrast-enhanced CT scans at regular end expiration breath hold with 1.5-mm slice thickness. The planning CT was fused with a T2- and multiple T1-weighted magnetic resonance images (MRI) at 0–20 min after injection of intravenous contrast agent (Gadovist, Bayer, Germany). When MRI was not available (e.g., due to pacemaker) a secondary contrast-enhanced CT was additionally used for treatment planning. A composite GTV was defined as the sum of the GTVs contoured on each of the fused CT or MRI images according to a previously published CyberKnife protocol [24]. The clinical target volume (CTV) consisted of the GTV with an expansion of 5 mm in all directions within the liver (excluding extension beyond liver parenchyma) to encompass microscopic tumor spread [25]. The PTV included the CTV and an expansion of 3 mm in all directions to encompass the targeting uncertainties for the CyberKnife system [26, 27].

Beam optimization was performed using the MultiPlan® (Accuray) treatment planning software (version 3.5 and 4.5) and the Sequential Multi-Objective Optimizer [28] according to the consensus guidelines for treatment planning for robotic radiosurgery [29]. The main objectives for optimization were to maximize the GTV mean dose above 3 × 18 Gy (BED₁₀ = 151.2 Gy₁₀), to cover 95 % of the PTV with 3 × 15 Gy with a maximum dose of 3 × 20 Gy, and to minimize all critical structures according to the ALARA (As Low As Reasonably Achievable) principle. If due to critical organ constraints [30] 3 × 15 Gy was not achievable for the PTV, the prescription dose was lowered (3 × 8–14 Gy) in 34.1 % of the cases and/or more protracted fractionation schedules were used (4–5 fractions) in 23.1 % of the cases. The aim in these cases was to maintain a high GTV mean dose above 3 × 18 Gy (BED₁₀ = 151.2 Gy₁₀) while not raising the maximum dose of 3 × 20 Gy (BED₁₀ = 180 Gy₁₀). In other words, whenever the PTV prescription dose was lowered, we compensated by generating steeper dose gradients in the CTV/PTV margin zone outside the GTV (Fig. 1). The final prescribed dose to the PTV ranged between 23 Gy and 45 Gy (median 39 Gy) to the 60–83 % isodose (median 75 %) and the mean GTV dose ranged between 33 Gy and 57 Gy (median 52.5 Gy) resulting in a median BED₁₀ of 86.1 Gy₁₀ (min 40.6 Gy₁₀ and max 112.5 Gy₁₀) surrounding 95 % of the PTV and of 142.1 Gy₁₀ (min 60.2 Gy₁₀ and max 165.3 Gy₁₀) for the mean GTV.

SBRT was delivered using the Synchrony® Respiratory Tracking System (Accuray) (versions 8.5 and 9.5). All patients were loosely immobilized using a custom made vacuum mattress (HEK Medical, Germany) and initially aligned using the spinal vertebra closest to the lesions. Respiratory motion was initially modeled, updated during treatment, and compensated during beam-on periods using the prediction of two to three LED chest markers which were correlated to one to five previously implanted gold fiducial markers detected on orthogonal x-ray imaging [22, 23, 31]. Combined average tracking errors (i.e., correlation and predictions errors) were kept below 1 mm and average rotation errors were minimized by aligning the patient to the average breathing phase, if rotation was detectable by the CyberKnife system. Median fraction treatment time was 41 min (range 18–76 min), excluding setup time.

All patients were observed 6 weeks after the end of their SBRT treatment and every 3 months thereafter. Every follow-up included the recording of possible adverse events according to the Common Terminology Criteria for Adverse Events (CTCAE, Version 4.03) for acute toxicity and the Radiation Therapy Oncology Group (RTOG) and European Organization for Research and Treatment of Cancer (EORTC) criteria for late toxicity. Imaging during follow-up was kept similar to the planning imaging (contrast-enhanced MRI or CT where applicable) and evaluated using the Response Evaluation Criteria In Solid Tumors (RECIST) with a dedicated focus on differentiation of radiation effects in the liver versus actual tumor growth.

In our statistical analysis, we evaluated local control (LC), progression-free interval (PFI), overall survival (OS) and toxicity. Local control was defined as complete remission (CR), partial remission (PR) or stable tumor size (ST) and was independently confirmed by PET imaging when CT or MRI gave suspicious but inconclusive results. All time points for LC, PFI and OS were calculated from the end of SBRT treatment to the respective event; death of any cause was the endpoint for OS. In case of multiple SBRT treatments, PFI and OS were calculated from the end of the first SBRT series and for LC each lesion was observed separately from the end of the respective SBRT treatment. In case of local recurrence, time to first description of suspected recurrence (back dating) was recorded. Surviving patients without a disease progression were censored at last follow-up. All curves were estimated using the Kaplan-Meier method.

The comparison of different patient or dosimetry groups was performed using the log-rank-test for the univariate and using cox-regression for the multivariate analysis. For LC histology, GTV volume, PTV prescription dose expressed as BED and GTV mean dose expressed as BED and for PFI and OS gender, age, GTV volume, cumulative GTV volume, Karnofsky performance status, histology, previous systemic treatment, extrahepatic tumor status and number of metastases were used as variables in the univariate analysis. For the indication of statistical significance a p-value of ≤ 0.05 was considered. Kaplan-Meier survival estimates and curves were calculated using the statistical program SPSS (Version 20.0, IBM, Armonk, USA). Cox regression survival analysis was performed using the R programming language for statistical computing (Version 3.2.3, The R Foundation for Statistical Computing, Vienna, Austria).

Local control (LC) was analyzed per treated lesions (n = 91). The overall crude LC at time of analysis was 89.0 %. The 2-year actuarial LC rate for all treated liver metastases was 82.1 % (Fig. 2). All 10 local failures occurred in patients with primary colorectal cancer (Fig. 3). Nine out of the 10 local failures occurred in two patients (one patient with two treatments of one and four metastases and one patient with one treatment of four metastases). Two out of ten local failures were marginal recurrences after previous SBRT treatment (re-treatment recurrences), and the remaining eight were from two patients who were treated for four metastases each and experienced simultaneous in-field recurrences. Five recurrences were re-treated with SBRT (two patients with one and four treated metastases), one recurrence was resected and four recurrences (one patient with four treated metastases) were treated with further chemotherapy only. In univariate analysis histology (colorectal cancer vs. other primary cancers, p < 0.001), PTV prescription BED₁₀ (Hazard Ratio HR 0.95, 95 % CI 0.91–0.98, p = 0.002) and GTV mean BED₁₀ (HR 0.975, CI 0.954–0.996, p = 0.011), both variables considered as continuous variables, were predictive for local control. Interestingly, all local recurrences occurred in colorectal metastases (Table 2). The 2-year actuarial LC rate for PTV prescription BED₁₀ > 86.1 Gy₁₀ was 96.6 % and for ≤ 86.1 Gy₁₀ 68.1 % (Fig. 3), the difference being significant (p = 0.005). In multivariate analysis PTV prescription BED₁₀ seemed to be the dominating factor for LC, though not reaching significance (p = 0.069). However, the results of the multivariate analysis should be interpreted with caution due to the limited number of events (n = 10) and all events occurring in colorectal cancer patients harboring the risk of over fitting.

Progression-free interval (PFI) analysis was based on treated patients (n = 52). Twenty-eight patients (53.8 %) developed new intrahepatic metastases. Eight of these patients were re-treated locally (six with SBRT). Twenty-four patients (46.2 %) developed extrahepatic metastases, three of whom were treated locally (all with SBRT). The median progression-free interval was 9 months (range, 1–36 months) and the 1- and 2-year actuarial progression-free interval rates were 35.1 and 17.7 %, respectively (Fig. 2). In univariate analysis patient age at time of SBRT, gender, tumor histology, number of treated metastases per patient, and GTV volume (largest metastases or cumulative volume) were not predictive for PFI (Table 2). On the other hand, status of extrahepatic disease at time of SBRT (HR 3.05, CI 1.59–5.84, p < 0.001), Karnofsky Index (HR 0.51, CI 0.27–0.98, p = 0.041), and prior liver therapy (HR 2.10, CI 1.04–4.23, p = 0.034) had significant effects on the PFI in our analysis. Multivariate analysis showed only the status of extra-hepatic disease at time of SBRT to be a significant factor for PFI (HR 2.33, CI 1.07–5.07, p = 0.033).

Overall survival (OS) analysis was based on treated patients (n = 52). Median overall survival at time of analysis was 23 months and cause of death was further tumor progression in 91.2 %. Actuarial 1- and 2-year overall survival rates were 70.2 and 45.0 %, respectively (Fig. 2). In univariate analysis patient age at time of SBRT, gender, Karnofsky-Index, number of treated metastases, prior liver therapy and GTV volume (largest metastases or cumulative volume) were not predictive for OS (Table 2). The 2-year actuarial OS was better for patients with colorectal cancer as the primary tumor site (58.7 %) compared to other primary tumor sites (35.7 %), although the difference was not statistically significant (p = 0.364). The only statistically significant variable prognostic for OS on univariate (HR 2.64, CI 1.30–5.38, p = 0.005) and multivariate (HR 2.59, CI 1.26–5.32, p = 0.009) analyses was the status of extrahepatic disease (no evidence of disease vs. stable or progressive extrahepatic disease) at time of SBRT (Fig. 2).

One patient developed an infected encapsulated hematoma in the liver and a right-sided pleural effusion after fiducial implantation. Overall, radiation treatment itself was well tolerated. Grade 1 acute nausea or fatigue were observed in 24.1 % of the patients and nausea was generally handled with antiemetic. Only one patient (1.9 %) had a grade ≥ 2 side effect and required a stent implantation for hepatic vein occlusion after SBRT of a tumor attached to that vein.