Dosimetric evaluation of target coverage as a predictor of local failure following stereotactic body radiation therapy for spinal tumors

Abstract

Purpose

To perform a dosimetric analysis of target coverage and determine parameters predictive for local failure (LF) in patients undergoing spinal stereotactic body radiation therapy (sSBRT).

Materials and Methods: Sixty-seven spinal tumors in 59 patients were treated with image-guided linac-based sSBRT from 2008-2012. Median prescription dose was 18Gy (8-35) delivered in 1-5 fractions (87% single-fraction). Prescription dose was targeted to cover ≥ 80% of PTV within spinal cord (SC) dose constraints (9/11Gy to 0.1cc SC/SC+2mm). Twelve tumors had local failure (LF, median time-to-failure 3.7 months) and were compared to 14 tumors with >1-year follow-up and local control (LC). Univariate and multivariate analyses were performed to determine parameters predictive of LF.

Results

Median follow-up was 7.4 months and 24.7 months for LF and LC, respectively. Post-SBRT, 42% of LF patients had neurological symptoms due to tumor progression. No patients developed post-SBRT myelopathy. Pre-treatment PTV volumes were not statistically different (median/mean/range 61.8/74.5/19.9-206.4cc for LF vs 39.4/47.1/10.3-119.7cc for LC; p=0.13). LF tumors had larger volumes receiving <80% of prescription dose (5.2cc vs 1.9cc, p=0.02) and larger overlap volume between GTV/SC within 2 and 3mm (p=0.01/p=0.007). LF tumors had lower GTV minimum dose (5.6 vs 8.5Gy, p=0.001) and smaller GTV to SC distance (0.06 vs 0.19mm, p=0.049). Maximum SC doses were not statistically different (6.4Gy LC vs 9.2Gy LF, p=0.33). GTV minimum dose was predictive of LF, with a trend for overlapping GTV/SC volume within 2mm.

Conclusions

Minimum GTV dose, PTV volume receiving <80% prescription dose, smaller GTV-SC distance, and large overlapping volume of PTV/SC are predictive of LF after SBRT. Given the absence of SC toxicity but neurological progression upon LF, less conservative SC constraints should be considered.

1 INTRODUCTION

Approximately 40% of cancer patients will develop spinal metastases [1-2]. Goals of treatment include palliation of pain, prevention of fractures, maintenance of neurological function, and improving quality of life [3-7]. Conventionally fractionated radiation therapy (RT) offers effective palliation of pain and maintenance of neurological function, but its effectiveness in tumor control is limited by SC tolerance [7]. Upon failure after conventional RT, treatment options are limited, particularly in patients who are not surgical candidates and re-irradiation with standard fractionation is frequently not possible. As such, spinal stereotactic body radiotherapy (sSBRT) has emerged as a salvage option for re-irradiation, as well as primary treatment for localized spinal tumors.

SBRT delivers precise and highly conformal RT in 1-5 high-dose fractions. This delivers a higher biological equivalent dose, not achievable utilizing standard RT methods. With sharp dose gradients, high tumor dose, and low dose to surrounding tissues, SBRT allows the safe delivery of biologically higher RT doses lending to improved local control (LC) [6]. The safety and efficacy of sSBRT has been shown to be effective in multiple retrospective studies as well as prospective series [7-12]. LC rates ranging from 70-90% have been reported following SBRT as primary, post-operative, and re-irradiation treatment [7, 13-15]. Phase II results of RTOG 0631 have demonstrated successful and feasible delivery of single fraction sSBRT for localized spine metastases, and Phase III results evaluating pain response and quality of life are awaited [11].

While SBRT allows us to achieve excellent LC for spinal tumors, predictors for LF remain unclear. Prior studies have demonstrated a dose response for LC. Yamada et al. demonstrated LC rates of 95% with single-fraction treatment at prescription doses > 24 Gy [15]. Lovelock et al. found that a minimum GTV dose of < 15 Gy predicted for LF [16]. As the use of sSBRT increases, identifying factors that contribute to a higher risk for recurrence will prove helpful in treatment planning. The purpose of this study is to present a dosimetric analysis of target volume coverage to determine factors predictive of LF in spinal tumors treated with SBRT.

2 MATERIALS AND METHODS

Sixty-seven spinal tumors in 59 patients were treated with sSBRT at our institution from 2008-2012. Prior to data collection, approval by the hospital Institutional Review Board Human Investigations Committee was granted. Details regarding patient/treatment characteristics, follow-up, and outcomes were retrospectively reviewed. Patients were seen at 8- and 12-weeks post-SBRT, and then followed at approximately 3-month intervals, with additional follow-up as needed. If patients developed symptoms or neurological deficits prior to the first planned follow up, imaging and clinical assessments may have been performed earlier. Assessments included: spine CT and/or MRI, Karnofsky Performance Status (KPS), American Spinal Injury Association (ASIA) score [17], pain at the SBRT site (pain-free/mild-moderate/severe), vertebral fracture, and myelitis.

Patient and tumor characteristics of the entire cohort are summarized in Table 1. Tumors were malignant in 94% of cases, with the remaining representing meningioma or schwannoma. Indications for SBRT included pain, neurological deficit, or oligometastatic disease. Median patient age was 63 years (16-86). Eighty-four percent of patients had normal neurological function prior to SBRT (ASIA E), with the remainder having some deficit (ASIA D). Twelve percent of patients had prior surgery, done at the discretion of the referring neurosurgeon for acute neurological deficits and/or spinal instability. Thirty-three percent of patients had prior RT to the SBRT site, and 62% received prior chemotherapy.

Table 1.

Total Cohort Patient/Tumor Characteristics (N=67)

Characteristic	N (%)
Median Age (years)	63
Gender
Male	32 (48)
Female	35 (52)
Karnofsky Performance Status
100	10 (15)
90	19 (28)
80	26 (39)
70	12 (18)
ASIA Score
E	56 (84)
D	11 (16)
C	0 (0)
B	0 (0)
A	0 (0)
Primary Tumor Site
Breast	17 (25)
NSCLC	7 (10)
Colorectal	1 (2)
Kidney	7 (10)
Melanoma	3 (5)
Prostate	10 (15)
Other	22 (33)
Histology
Adenocarcinoma	39 (58)
Squamous Cell	3 (5)
Other	25 (37)
Tumor Location
C-Spine	5 (8)
T-Spine	48 (72)
L-Spine	13 (19)
Sacrum	1 (1)
Prior Surgery
Yes	10 (15)
No	57 (85)
Prior Radiation to SBRT Site
Yes	21 (31)
No	46 (69)
Solitary Spine Metastasis
Yes	18 (27)
No	46 (69)
Unknown	3 (4)
Indication for SBRT
Pain	38 (57)
Neurological Deficit	10 (15)
Oligometastasis	19 (28)

2.1 Treatment Planning and Target Delineation

All patients were treated with image-guided linear accelerator-based SBRT. Treatment delivery was performed utilizing 3-Dimensional Conformal RT (n=4), Intensity-Modulated RT (n=51), or Volumetric Modulated Arc Therapy (n=12). Patients underwent CT simulation with extracranial immobilization. If a pre-SBRT MRI was performed, images were fused to the planning CT for target delineation in the treatment planning system (Pinnacle, Phillips; Fitchburg, WI).

Gross tumor volume (GTV) was defined as gross tumor on imaging, which included radiographically visible tumor within the vertebral body as well as any epidural or paraspinal soft tissue component. Clinical target volume (CTV) included the GTV as well as the remaining vertebral body if not included in the GTV. The planning target volume (PTV) was defined as a 2 mm expansion of the CTV in three dimensions, excluding the volume represented by a 2 mm expansion of the spinal cord (SC). Required normal tissue contours varied according to tumor location. A SC contour was required in all cases. For SC and SC+2mm, the tissue dose constraints to 0.1cc of the volume were a maximum of 10 Gy and 13 Gy, respectively. Patients received 1-5 fractions, with a median dose of 18 Gy (8–35). The majority of patients (87%) received single-fraction treatment. The single fraction treatment dose was 10 (n=1), 15, (n=2), 16 (n=6), or 18 (n=14) Gy. Patients with larger treatment volumes were treated with 3-5 fractions in order to meet dose constraints, at the discretion of the treating physician. Multi-fraction treatment regimens included 19.5 Gy in 3 fractions (n=1), 35 Gy in 5 fractions (n=1), and 25 Gy in 5 fractions (n=1). The prescription isodose surface was chosen such that ≥ 80% of the PTV received the prescription dose. The biological effective dose (BED) was calculated using the linear quadratic model [18]. An α/β ratio of 10 Gy was assumed for spinal tumors.

2.2 Definition of Failure and Dosimetric Analysis Cohort

Each pre-treatment and post-treatment CT/MRI was reviewed by a single neuroradiologist and LC/LF was determined radiographically. LF was defined as the return of measurable tumor following initial complete response, or increase in tumor size following partial response. LC was defined as the absence of LF. Time to LF/LC was defined from the date of SBRT completion to the date of either imaging failure (LF) or last imaging follow-up (LC). Twelve tumors were identified as having LF, all representing malignant metastases. To determine factors predictive of failure, these tumors were compared to a control group of LC tumors. Tumors in the LC group included those with ≥ 1 year of follow-up (demonstrating durable response to SBRT) and complete dosimetric information for analysis (n=14). The combined 26 tumors served as the analysis cohort for the study.

The pre- and post-treatment MRI images were fused to the planning CT utilizing MIM Software (MIM Inc., Cleveland OH). Dosimetric parameters collected included: BED, GTV volume, GTV maximum (GTVmax), GTV minimum (GTVmin), and GTV mean (GTVmean) dose; PTV volume; PTV absolute volume outside the 80% isodose line (PTV80), PTV percent volume outside the 80% isodose line; minimum and maximum distance between GTV and SC (GTV-SC); maximum dose to 0.1 cc of SC (SCmax); and overlapping PTV/SC volume within 0-5 mm.

2.3 Statistical Analysis

Time intervals were calculated from the date of RT completion. Statistical analysis was performed using SPSS Version 20 (SPSS, Inc., Chicago, IL). All statistical tests were two-sided. The difference between two sample means of continuous variables was analyzed with an Independent samples t-test and categorical variables using Chi-Square. Univariate and multivariate analysis was performed to correlate LF using a Cox regression model. Numerical variables were run as both continuous and categorical, using the median value as a cutoff for categorical variables. Kaplan-Meier curves were used to demonstrate LC for categorical variables. All variables that were statistically significant in the univariate analysis were included in the multivariate analysis. A p-value of ≤0.05 was considered statistically significant.

3 RESULTS

3.1 Patient Characteristics

Patient and tumor characteristics of the analysis cohort are listed in Table 2. Overall, median clinical follow-up for the entire cohort (LF and LC) was 15 months. Follow up for patients who developed LF was 7.4 months (range: 1.3-36.5 months). Follow up for patients with LC at the time of analysis was 24.7 months (range: 8.4-41.4 months). Median patient age in the analysis cohort was 64 years (16-86). Sixty-two percent of patients had prior chemotherapy, 15% had prior surgery, and 38% had prior RT to the sSBRT site. Patients received SBRT for pain (65%), neurological deficit (8%), or oligometastatic disease (27%). Ninety-two percent of patients had an ASIA score of E (normal function). At the time of last clinical follow-up, 42% of LF patients developed progressive neurological dysfunction due to tumor progression. No patients in either group developed post-SBRT myelopathy. The presence of visceral metastases was significantly higher in the LF cohort, 67% vs 17% (p=0.04). There was a trend towards greater paraspinal tumor involvement in the LF cohort, 58% vs 21% (p=0.06). All other clinical variables were similar between the cohorts.

Table 2.

Analysis Cohort Patient/Tumor Characteristics (N=26)

	All Patients	Local Control Cohort (N=14)	Local Failure Cohort (N=12)	p-value
Characteristic	N (%)	N (%)	N (%)
Mean Age (years)	61	62	59	0.64
Gender					0.21
Male	14 (54)	6 (43)	8 (67)
Female	12 (46)	8 (57)	4 (33)
Karnofsky Performance Status					0.33
100	5 (19)	4 (28.5)	1 (8.5)
90	10 (38.5)	4 (28.5)	6 (50)
80	10 (38.5)	6 (43)	4 (33)
70	1 (4)	0 (0)	1 (8.5)
ASIA Score					0.72
E	24 (92)	13 (93)	11 (92)
D	2 (8)	1 (7)	1 (8)
C	0 (0)	0	0
B	0 (0)	0	0
A	0 (0)	0	0
Primary Tumor Site					0.54
Breast	6 (23)	3 (21.5)	3 (25)
NSCLC	3 (11)	2 (14)	1 (8)
Colorectal	1 (4)	0	1 (8)
Kidney	2 (8)	1 (7)	1 (8)
Melanoma	1 (4)	0	1 (8)
Prostate	6 (23)	5 (36)	1 (8)
Other	7 (27)	3 (21.5)	4 (33)
Histology					0.63
Squamous Cell	1 (4)	1 (7)	0
Adenocarcinoma	17 (65)	9 (64)	8 (67)
Other	8 (31)	4 (29)	3 (33)
Tumor Location					0.04
C-Spine	1 (4)	1 (7)	0
T-Spine	20 (77)	8 (57)	12 (100)
L-Spine	5 (19)	5 (36)	0
Epidural Spinal Cord Compression Grade [25]					0.05
0	13 (50)	9 (64)	4 (33)
1a	3 (11)	3 (21)	0
1b	2 (8)	1 (7)	1 (8)
1c	5 (20)	1 (7)	4 (33)
2	3 (11)	0	3 (25)
3	0	0	0

3.2 Failure Patterns

The one-year actuarial LC rate was 72% in the entire cohort. In the patients that developed LF, median time to failure was 3.7 months (0.7-45.7). Median SBRT dose in the LF cohort was 18 Gy (10-35), with all but 2 patients receiving single-fraction treatment. For the LC cohort, median SBRT dose was 18 Gy (15-19.5), with all but 1 patient receiving single-fraction treatment. Details of the dosimetric analysis are shown in Table 3. Pre-treatment GTV and PTV volumes were similar between the groups. Median, mean, and range GTV volume (cc) was 19, 29, and 1.2-91 vs 31, 44, and 3.7-136, for LC and LF, respectively (p=0.30). Median, mean, and range PTV volume (cc) for LC and LF was 39, 47, and 10.3-120 and 62, 75, and 20-206, respectively (p=0.13). Minimum GTV dose was significantly lower in the LF cohort, 5.6 Gy vs 8.5 Gy (p=0.001). There was no difference in GTVmax and GTVmean dose. The PTV80 was significantly higher for tumors with LF, 5.2 ± 3.7 vs 2.0 ± 2.3 cc (p=0.02). Similarly, the percentage of PTV outside the 80% isodose line was higher in the LF group, 8% ± 3% v 4% ± 3% (p=0.02). SCmax dose was not statistically different between the two cohorts, 6.4 Gy vs 9.2 Gy for LC and LF, respectively (p=0.33). While there was no difference in maximum GTV-SC distance (1.5 vs 1.2 cm, p=0.27), the GTV-SC minimum distance was shorter in the LF cohort (0.06 vs 0.19 cm, p=0.049). The GTV/SC overlapping volume within 0-5 mm was measured. GTV/SC overlap volumes within 2 and 3 mm were significantly larger in the LF cohort: 0.4 vs 0.1 cc and 0.9 vs 0.2 cc for volumes within 2 mm (p=0.01) and 3 mm (p=0.007), respectively. Univariate analysis was performed in an attempt to correlate factors predictive of LF. Results can be seen in Table 4 and Figures 1 and 2. GTVmin dose (continuous and < 7 Gy), PTV80 (> 2 cc), GTV-SC minimum distance (continuous), and overlapping GTV/SC volume within 2 (> 0.09 cc) and 3 mm (continuous and > 0.25 cc) were predictive of LF, with a trend seen for SCmax (continuous). On multivariate analysis, the only factor that remained significant was GTVmin dose, with a trend to significance for overlapping PTV/SC volume within 2 mm.

Table 3.

Dosimetric Analysis

Variable	Local Control Cohort, N=14, mean (SD)	Local Failure Cohort, N=12, mean (SD)	p-value
BED (Gy)	44.7 (9.3)	46.1 (6.7)	0.69
GTV Volume (cc)	28.8 (26.8)	43.6 (43.2)	0.30
GTV Minimum Dose (Gy)	8.5 (2.1)	5.6 (1.6)	0.001
GTV Mean Dose (Gy)	19 (2.7)	20 (3.5)	0.39
GTV Maximum Dose (Gy)	21.9 (3.4)	24.7 (4.9)	0.10
PTV Volume (cc)	47.1 (31.6)	74.5 (56.1)	0.13
PTV Volume Outside 80% IDL (cc)	2.0 (2.3)	5.2 (4.0)	0.02
% PTV Outside 80% IDL	4.3 (3.7)	7.9 (3.7)	0.02
Maximum SC Dose to 0.1 cc SC (Gy)	6.4 (2.6)	9.2 (9.7)	0.33
GTV to SC Minimum distance (cm)	0.19 (0.2)	0.06 (0.1)	0.05
GTV to SC Maximum distance (cm)	1.5 (0.9)	1.2 (0.4)	0.27
GTV/SC Overlap Volume w/in 2 mm (cc)	0.10 (0.3)	0.37 (0.2)	0.01
GTV/SC Overlap Volume w/in 3 mm (cc)	0.24 (0.4)	0.93 (0.7)	0.007

Table 4.

Univariate Analysis for Local Failure

Variable	Hazard Ratio (95% CI)	p-value
Patient Age
Continuous	0.98 (0.95-1.02)	0.38
> Median (64)	0.55 (0.18-1.90)	0.34
KPS
Continuous	0.96 (0.90-1.04)	0.31
Time from treatment planning to SBRT
Continuous	0.96 (0.87-1.05)	0.35
> Median (12 d)	0.35 (0.09-1.31)	0.12
GTV Volume (cc)
Continuous	1.01 (0.99-1.02)	0.26
> Median (21 cc)	1.85 (0.54-6.36)	0.33
PTV Volume (cc)
Continuous	1.01 (1.0-1.02)	0.16
> Median (42 cc)	1.55 (0.45-5.30)	0.49
BED (Gy)
Continuous	1.03 (0.95-1.10)	0.52
> Median (50 Gy)	0.79 (0.24-2.60)	0.96
GTV Maximum Dose (Gy)
Continuous	1.07 (0.96-1.2)	0.21
> Median (23 Gy)	1.15 (0.35-3.80)	0.82
GTV Minimum Dose (Gy)
Continuous	0.66 (0.48-0.92)	0.02
> Median (7 Gy)	0.13 (0.03-0.62)	0.01
GTV Mean Dose (Gy)
Continuous	1.07 (0.91-1.28)	0.42
> Median (19 Gy)	0.61 (0.18-1.99)	0.41
PTV Volume outside 80% IDL (cc)
Continuous	1.12 (0.99-1.25)	0.05
> Median (2 cc)	5.59 (1.19-26.20)	0.03
GTV to SC Minimum distance (cm)
Continuous	0.003 (0.00-0.91)	0.046
> Median (0.09 mm)	0.45 (0.12-1.63)	0.22
GTV to SC Maximum distance (cm)
Continuous	0.57 (0.22-1.46)	0.24
> Median (1.18 mm)	0.64 (0.19-2.12)	0.46
Maximum SC Dose to 0.1cc of SC (Gy)
Continuous	1.06 (0.99-1.13)	0.10
> Median (6 cc)	0.85 (0.26-2.81)	0.79
GTV/SC Overlap Volume w/in 2 mm (cc)
Continuous	5.10 (0.71-36.50)	0.11
> Median (0.09 cc)	4.10 (1.04-16.27)	0.04
GTV/SC Overlap Volume w/in 3 mm (cc)
Continuous	2.52 (1.08-5.85)	0.03
> Median (0.25 cc)	4.10 (1.04-16.27)	0.04
Prior Palliative Chemotherapy
Yes	2.79 (0.60-12.90)	0.19
Prior Surgery at SBRT Location
Yes	0.52 (0.07-4.07)	0.53
Prior RT at SBRT Location
Yes	1.03 (0.30-3.53)	0.97
Compression Fracture Prior to SBRT
Yes	1.1 (0.23-5.33)	0.90

Figure 1.

Kaplan-Meier actuarial curves for LC in: A) tumors with PTV80 < 2cc vs ≥ 2cc and B) tumors with GTVmin < 7Gy vs ≥ 7Gy.

Figure 2.

Kaplan-Meier actuarial curves for LC in tumors with: A) GTV/SC overlap volume within 2mm >0.09cc vs ≤0.09cc and B) GTV/SC overlap volume within 3mm >0.25cc vs ≤0.25cc.

4 DISCUSSION

In this cohort of 67 tumors treated with sSBRT, the 1-year actuarial LC rate was 72%. With further analysis of the 12 tumors which had LF compared to a LC cohort, we found dosimetric parameters that could predict for LF. As expected, larger tumor volume outside the prescribed dose was predictive of failure. Minimum GTV dose for LF tumors was significantly smaller than LC tumors. Additionally, smaller distances between tumor and SC and larger overlap volumes between GTV/SC predicted for LF. Tumors in the LF cohort had a numerically larger maximum SC dose, however, this was not statistically different. The larger doses likely reflect larger tumor/treatment volumes in the LF group, which may in turn partially explain the failure. On multivariate analysis, GTV minimum dose was predictive of LF, likely related to the need to respect SC tolerance. No patients developed post-SBRT myelopathy. Clinical factors such as prior chemotherapy, RT, or surgery did not impact LF. Additionally, tumor volume and BED did not predict for LF.

With regards to LC, our results compare less favorably with other studies in the literature [7, 13, 15]. This could be because many patients in our study were lost to follow-up or did not have follow-up imaging available for assessment. Given the shorter life expectancy of patients with metastatic disease and the often palliative nature of the radiation treatments given, many patients do not return for imaging or clinical assessments in the follow up period. If more data points with regards to imaging information had been available, the differences seen between the LF and LC cohorts may have lost significance. The Mayo Clinic reported an actuarial 1-year LC rate of 89% for a cohort of 85 tumors treated with sSBRT [13]. Gerszten et al. reported a LC rate of 88% for 500 cases treated with single-fraction sSBRT at a median dose of 20 Gy [7]. Yamada et al. reported outcomes of 103 spinal metastases treated with single-fraction IMRT to 18-24 Gy, with an actuarial LC of 90% at a median follow-up of 15 months. Prescribed dose was a significant predictor for failure. The median percentage of PTV receiving ≥ 95% prescribed dose was 95% with a SC dose constraint of < 12-14 Gy. No patients developed myelopathy. For the 7 patients developing failure, median time to failure was 7 months [15]. In our study, the median time to failure was 3.7 months, relatively shorter as compared to other reported data. It is possible that patients identified as having LF in our cohort represented false positive findings as a result of pseudoprogression, which is believed to occur following SBRT. Many of the patients who developed LF underwent surgery due to worsening pain or neurological deficit immediately at the time of the documented imaging failure. Others were initiated on systemic therapy. For patients whose disease progressed and no further treatment options were available, comfort care or hospice was initiated, and no further imaging studies were performed. Given these factors, further assessment of imaging following the documented LF was unable to be performed, limiting our ability to identify those with true pseudoprogression and potentially increasing the failure rate in this cohort.

While reported control rates are similar among institutions, none consistently report factors predictive of failure. Lovelock et al. correlated LF with measures of dose insufficiency for 91 tumors treated with high-dose single fraction sSBRT [16]. In comparing 7 tumors with LF to the whole patient group, they found that distributions of the minimum doses received by the hottest 100% (D_min), 98% (D₉₈), and 95% (D₉₅) of the GTV were statistically different. D_min for patients experiencing LF was lower (10.8 vs 15.7 Gy) and freedom-from-LF for D_min > 15 Gy vs < 15 Gy was significantly longer (p<0.02). No failures were observed for D_min > 15 Gy. These results indicate that D_min is a risk factor for LF, similar to the results in the current study. Chang et al. explored patterns of failure following sSBRT in a Phase I/II study of 74 lesions [19]. Prescription dose was chosen to meet the SC dose constraint of ≤ 9-10 Gy. At a median follow-up of 21 months, 23% of tumors demonstrated LF with 47% having progression in the epidural space. A pattern-of-failure analysis showed that failure occurred in the bone adjacent to the treatment site and/or in the epidural space adjacent to SC. The authors reported that under-dosing occurred in the epidural region as a result of SC constraints, with 9-10 Gy possibly an overly conservative constraint. While our study did not specifically explore geometric failure patterns, it was noted that the majority of our failures occurred in the epidural space.

For single-fraction SBRT at our institution, SC dose constraints are as follows: 10 Gy and 13 Gy to ≤0.1 cc of SC or SC+2 mm, respectively. This maximum point dose is low in comparison to other institutions performing sSBRT. The Quantitative Analyses of Normal Tissue Effects in the Clinic defines a 13 Gy D_max SC for single fraction SBRT [20], while other institutions such as Cleveland Clinic and The University of Glasgow use 14 Gy as a maximum allowance for SC point dose [6, 21]. Mayo Clinic uses a constraint of > 14 Gy to < 0. 5 mL of the SC.¹³ Memorial Sloan-Kettering limits the maximum SC point dose to 14 Gy [15].

At the time that SBRT was initiated at our institution, there were few institutions with experience in sSBRT and little published data on its safety and efficacy. Therefore, a more conservative SC constraint was chosen. To date, more data has been published on the safety of sSBRT, specifically with regards to radiation-induced myelopathy (RIM). In a retrospective analysis of patients developing permanent myelopathy following conventional RT, Wong et al. reported a latency time of 18 months for development of RIM following single-course treatment. In patients receiving re-irradiation, latency rates were significantly shorter (11 months, p=0.03) [22]. Data regarding RIM following sSBRT was more recently reported by Gibbs et al., showing a mean of 6 months for the development of myelopathy in 6/1075 patients treated [23]. Maximum SC doses ranged from 3.6-29 Gy, and 3/6 patients developing myelopathy received a ≥ 8 Gy dose equivalent to SC to a volume > 1 cm³. No specific dosimetric parameters contributed to complications. Gerszten et al. saw no clinically or radiographically detected RIM or SC damage at a median follow-up of 21 months following sSBRT, with maximum tumor doses of 22.5 Gy in a single-fraction [7]. Ahmed et al. reported the sSBRT experience at Mayo Clinic, where 85 lesions were treated to a median dose of 24 Gy (10-40) in 1-5 fractions with a median maximum PTV dose of 31 Gy. When limiting < 0.5 mL of SC to >14 Gy, the median maximum SC dose was 16 Gy. At a mean follow-up of 8 months, limited toxicity was noted with no cases of RIM [13]. Recently, Sahgal et al. evaluated the probability of RIM following SBRT in previously untreated patients, comparing 9 myelopathy cases to 66 without. Median follow-up in the cohort was 15 months, with a 12-month (3-15) median time to myelopathy. The authors reported logistic estimates for the probability of developing RIM, with doses to point volumes conferring a 1-5% risk of myelopathy following SBRT in 1-5 fractions. They concluded that limiting the contoured thecal sac volume (representing SC) to a maximum point dose of 12.4 Gy for single-fraction treatment results in a <5% risk of RIM [24].

In the current study, no patients developed RIM. The cohort represented a combination of single-course, re-irradiation, and post-operative SBRT. Mean follow-up was 20 months (1-45), indicating a sufficient follow-up period for the detection of RIM following SBRT, according to published data [22-23]. The mean maximum tumor dose was 23 Gy (13-36). Under these conditions, no adverse SC events were noted. When compared with Sahgal et al. [24], the maximum point doses in our study are well below this recommended level, indicating room for additional dose.

There are limitations to our analysis, including the inherent drawbacks of a retrospective study. Our study population represents a small group of heterogeneous tumors with varying histology and tumor location. However, the majority represented single-fraction treatment for malignant tumors. In comparing to other studies evaluating sSBRT, this is similar [7, 9, 15, 21]. While patient numbers are limited, there are no current studies examining several of these specific dosimetric parameters. Due to the selection process of the control group, there is potential for bias. We attempted to partially eliminate this by limiting the control group to patients with ≥ 1-year follow-up, to remove any patients that had LC but may have died within a year following treatment. A match-pair analysis could minimize potential bias, however given the small number of patients, this was not feasible. With these issues in mind, the investigations will be repeated on a larger group of patients in a multi-institutional setting, with attempts to match each LF to a LC tumor with similar follow-up. Although our study population was small and heterogeneous, we were still able to illicit factors significantly predictive for LF. We provide a unique set of dosimetric parameters which have not been previously studied in this population.

5 CONCLUSIONS

SBRT for spinal tumors is a safe and effective treatment to establish long-term tumor control. GTVmin dose, PTV volume receiving < 80% prescription dose, minimum GTV-SC distance, and overlapping GTV/SC volumes were dosimetric parameters predictive of LF after sSBRT in our cohort. Given the absence of SC toxicity with symptomatic progression upon failure, less conservative SC dose constraints should be considered to achieve improved coverage and tumor control.