Postoperative Epidural Disease and Local Control after Separation Surgery and Spine Stereotactic Body Radiation Therapy

Abstract

Purpose:

Limited data are available on outcomes of patients with spinal metastases treated with separation surgery with postoperative stereotactic body radiation therapy (SBRT).

Methods and materials:

This is a retrospective cohort study of 497 patients treated to 532 lesions between 2013–2024. We evaluated local failure (LF) and adverse events (hardware failure [HF] requiring surgical intervention, radiation myelitis [RM] and vertebral compression fracture [VCF]). Incidences of LF, VCF, RM, and HF were analyzed considering death as a competing risk. Dosimetric variables, including minimal biological effective dose to the clinical target volume (CTV Dmin BED10) were obtained. Preoperative and postoperative epidural spinal cord compression (ESCC) scores were calculated based on preoperative MRI and postoperative myelogram simulation.

Results:

Median follow-up time among surviving patients was 74.5 months (95% CI 70–94). Median overall survival was 14 months (95% CI 13–17). The risk of LF (95% CI) at 1, 2, and 3 years was 7.7% (5.6–10%), 10% (7.9–13%), and 13% (10–16%), respectively. At 2 years, cumulative incidences of VCF, RM, and HF requiring surgery were 2% (95% CI, 1–3.5%), 1% (0.39–2.3%), and 3.8% (2.4–5.8%), respectively. On univariable and multivariable regression, CTV Dmin BED10 correlated with risk of LF (on multivariable regression, hazard ratio 0.96, 95% CI 0.93–0.99, p = 0.005). After excluding previously radiated lesions, postoperative ESCC ≤1C was associated with an approximately 2-times higher likelihood of achieving the median CTV Dmin BED10 than ESCC 2 (OR: 2.06, 95% CI 1.31–3.26, p=0.002), whereas a postoperative ESCC ≤1B was not associated with higher odds of achieving the median CTV Dmin BED10 compared to ≥1C (OR: 0.75, 95% CI 0.50–1.11, p=0.15).

Conclusions:

In this large series of separation surgery followed by spine SBRT, local control was excellent with few adverse events. Higher CTV minimum biological dose—enabled by postoperative ESCC ≤1C —was the strongest predictor of durable local control.

Introduction

As per the neurologic, oncologic, mechanical, systemic (NOMS) decision framework for patients with spinal metastases, separation surgery should be strongly considered in the setting of patients with radioresistant tumor histologies (defined based on sensitivity to conventionally fractionated radiotherapy), tumors with epidural spinal cord compression (ESCC) scores of 2 or 3, i.e. “high-grade” epidural disease, in the setting of myelopathy, and in the setting of mechanical instability.^1,2 The goal of separation surgery, a term coined by Benzel and Angelov, as per the initial publication of the NOMS decision framework, is “minimal tumor resection … to separate the tumor margin from the spinal cord, leaving the bulk of the tumor mass to be treated with radiation.”¹

Stereotactic body radiation therapy (SBRT) has become a standard for the treatment of spinal metastases,³ supported by level 1 evidence from the SC.24/TROG 17.06 study, which showed that SBRT outperformed conventionally fractionated radiotherapy (CFRT) for pain control⁴ and local control.⁵ However, there is a relative paucity of data on outcomes for patients treated with separation surgery followed by spine SBRT, particularly after the development of updated research on the role of histology in determining radiosensitivity to SBRT rather than CFRT.⁶

To garner further information on the outcomes and adverse events associated with this regimen—particularly in the setting of updated understanding of radiosensitivity to SBRT regimens—and to determine the ideal patients to be considered for separation surgery followed by spine SBRT, we analyzed a large database of patients treated at a tertiary cancer center between 2013–2024 in a retrospective cohort study.

Methods and Materials

Patient Selection

The Institutional Review Board approved this retrospective cohort study without the need for informed consent. Figure 1 shows the CONSORT Diagram for patients included in the study. Patient selection for separation surgery and SBRT was conducted using a multidisciplinary approach utilizing the NOMS criteria, weighing neurologic symptoms, histology/radiosensitivity and receipt of prior radiation, mechanical symptoms, and systemic disease as well as performance status. A total of 1,470 patients underwent separation surgery between January 2013 and December 2024. Patients were included if they received post-separation surgery SBRT to a spinal level between C1-L5 with a biological effective dose (BED) assuming an α/β of 10 of greater than 48 Gy in 5 or fewer fractions. All patients received a minimum follow up of 1 month with post-SBRT magnetic resonance imaging (MRI). Patients with prior overlapping radiotherapy to the spine were eligible, and patients who received SBRT with protons were also eligible.

Figure 1. CONSORT DIAGRAM.

Separation Surgery

The technique of separation surgery has been described previously.⁷ The first step involves a laminectomy to achieve posterior decompression and to identify a normal dural plane. The second step focuses on ventral decompression. At cervical and thoracic levels, this requires either unilateral or bilateral facetectomy followed by a transpedicular approach to access the ventral aspect of spinal cord. In the lumbar spine, a complete facetectomy and pediculectomy are generally not required. In all cases, resection of the posterior longitudinal ligament and the intraoperative use of ultrasonography are recommended to assess the ventral extent of tumor debulking and restoration of the spinal cord contour. Gross-total tumor resection is not pursued. Due to complete or partial resection of the joints and pedicles, posterior spinal fixation was performed using pedicle screws and connecting rods. When more than 50% of the vertebral body was removed, vertebral body reconstruction with polythemylmethacrylate (PMMA) was performed using a posterior-only approach. The 3mm target in separation surgery represents a theoretical safety margin for photon radiotherapy. To evaluate the optimal extent of tumoral resection in relation to tumor histology and its predictive value for local failure of postoperative SBRT, both preoperative and postoperative ESCC were included in this analysis.

Spine SBRT

The technique of spine SBRT has also been described at length in previous publications.^6–9 Briefly, patients were simulated supine with a CDR frame (CDR Systems) and a CT scan with 2 mm slices. In all but one case (531/532; 99.8%), a postoperative myelogram was conducted for simulation, and cord constraints were applied to the true spinal cord structure (Supplemental Table 1). This myelogram was obtained a median of 5 (interquartile range [IQR] 4–7) days after surgery while patients were still in the hospital. In the remaining case, a focal, 3D MRI with 2mm slices was obtained and fused with the CT simulation scan, and a 1mm isotropic planning organ at risk volume was applied to the spinal cord. Contours included clinical target volume (CTV), made up of gross tumor plus adjacent components of the vertebral structure as per standardized guidelines.¹⁰

The CTV was drawn utilizing pre-surgery MRI; if portions of the vertebral structure were removed, the CTV included preoperative extent of the tumor. The planning target volume (PTV) consisted of a 2mm expansion on the CTV, excluding esophagus and/or thecal sac. We examined various dosimetric variables including total structure volume, minimum (Dmin), maximum (Dmax), mean (Dmean), and minimum dose covering 95% of CTV or PTV (D95%), all of which (except for the volume) were adjusted to biological effective dose (BED) assuming an α/β of 10. Our institutional constraints for the esophagus, spinal cord, and cauda equina in both the upfront and re-irradiation setting are shown in Supplemental Table 1.

Outcome Measures and Statistical Analysis

The primary outcome of interest in this study was local failure (LF) after separation surgery and spine SBRT, defined as increased size and/or increased perfusion on post-SBRT MRI. Overall survival (OS) was defined as time between SBRT and either last known follow up or death, if death occurred. Follow-up time was calculated using the reverse Kaplan Meier method. Secondary outcomes of interest included toxicities such as radiation myelitis (RM), vertebral body compression fracture (VCF), and hardware failure requiring surgical intervention (HF). RM was designated as characteristic radiological findings in the SBRT field on post-SBRT MRI, with or without characteristic clinical symptoms of myelitis.⁹ VCF was designated as progressive or new fracture without evidence of LF.

ESCC scores were designated on preoperative MRI and postoperative myelogram (and in 1 case, focal MRI) according the Bilsky et al.¹¹ Histological classes of radiosensitivity were assigned as per Jackson et al,⁶ with Class A being breast and prostate cancer, Class C being colorectal, hepatocellular, and cholangiocarcinoma, and Class B being all other histologies. Dosimetric variables included CTV and PTV D95% (minimum dose to 95% of the target volume), CTV and PTV volume, CTV and PTV maximal dose (Dmax), and CTV and PTV minimal dose (Dmin), variables previously found to be important factors in local control after spine SBRT. All variables, apart from the volumes, were calculated as BED to adjust for different dose-fractionation schemes. Our institutional practice is that gross tumor volume (GTV) is not a required target volume for treatment planning, particularly in the postoperative setting. As such, GTV is variably contoured, and was excluded from analysis.

OS was estimated using Kaplan-Meier methodology. LF, RM, VCF, and HF were each analyzed in a competing-risks framework, with death treated as a competing event. Cumulative incidence functions were generated, and differences between groups were assessed using Gray’s test. Univariable and multivariable cause-specific hazard regression modeling using Cox proportional hazards was performed to evaluate associations between clinical or dosimetric variables and the risk of LF, adjusted for clustering at the patient level. Variables were included on multivariable analysis if they were statistically significant on univariable analysis or if they had been previously shown to correlate with local control in previous analysis (histological class, previous radiation, and postoperative ESCC). A maximum of one dosimetric variable (the variable with the lowest p-value) was included in the multivariable regression for local failure due to potential collinearity.

To explore whether more extensive surgical separation conferred additional benefit, we performed a secondary logistic regression analysis in patients without prior overlapping radiation therapy, modeling the probability of achieving at least the median CTV minimum BED10 as a function of postoperative ESCC dichotomized at ≤1B versus ≤1C. Differences in the proportion of lesions with prior overlapping radiation therapy across tumor Class were assessed using the chi-squared test. Differences in CTV minimum BED10 across postoperative epidural spinal cord compression (ESCC) categories were evaluated using the Kruskal–Wallis rank-sum test. The linearity assumption for all continuous variables was assessed using restricted cubic splines with four knots. All continuous variables demonstrated no significant departure from linearity (all nonlinearity p > 0.05) and were therefore entered as continuous linear terms in the multivariable Cox proportional hazards model. Statistical significance was defined as p less than 0.05. Analyses were conducted using R version 4.5.1 (R Foundation for Statistical Computing).

Results

A total of 497 patients treated to 532 spinal segments (Figure 1) met criteria for analysis. Table 1 shows the baseline characteristics for patients and lesions. The majority of patients were male (55%), and the majority of patients were white (75.9%). Non-small-cell lung cancer (NSCLC) was the most common histology (n=92; 17%). Most lesions involved the thoracic spine (n=354; 67%). A total of 107 courses (20%) involved prior overlapping radiotherapy to the treated spinal lesion. Eight treatment courses (1.5%) were done with proton SBRT. The median BED₁₀ was 60 Gy (interquartile range (IQR) 51–72 Gy).

Table 1.

Baseline Characteristics for Treated Patients and Lesions

Characteristic	Value
Median Age at RT (Range), y	63 (19, 86)
Median Follow Up (95% CI), mo	74.5 (70, 94)
Median Overall Survival (95% CI), mo	14 (13, 17)
Sex
Female	224 (45%)
Male	273 (55%)
Race
White	376 (76%)
Asian	45 (9.1%)
Black or African American	36 (7.2%)
Some Other Race	12 (2.4%)
Unknown	24 (4.8%)
Asian-Far East/Indian Subcont	2 (0.4%)
White African Caribbean	1 (0.2%)
Native Hawaiian or Other Pacific Islander	1 (0.2%)
Histology
NSCLC	92 (17%)
Sarcoma	83 (16%)
Renal	85 (16%)
Prostate	42 (7.9%)
Thyroid	42 (7.9%)
Breast	34 (6.4%)
Colorectal	22 (4.1%)
Melanoma	20 (3.8%)
Head and Neck	18 (3.4%)
Bladder	11 (2.1%)
Pancreatic Adenocarcinoma	3 (0.6%)
Cholangiocarcinoma	8 (1.5%)
Thymic	4 (0.8%)
Esophageal	3 (0.6%)
Hepatocellular	5 (0.9%)
Other	60 (11%)
Segment
Cervical	40 (7.5%)
Cervicothoracic	16 (3.0%)
Thoracic	354 (67%)
Thoracolumbar	3 (0.6%)
Lumbar	118 (22%)
Prior Overlapping RT	107 (20%)
Type of RT
Photons	524 (98%)
Protons	8 (1.5%)
Dose (Gy) / Fraction(s)
27 / 3	227 (43%)
30 / 3	144 (27%)
40 / 5	86 (16%)
24 / 1	45 (8.5%)
35 / 5	21 (3.9%)
36 / 3	6 (1.1%)
32 / 4	1 (0.2%)
37.5 / 5	1 (0.2%)
18 / 1	1 (0.2%)
Median BED (Gy), IQR	60 (51, 72)
Median BED CTV D95% (Gy), IQR	54 (51, 61)
Median BED PTV D95% (Gy), IQR	53 (50, 60)
Median BED CTV Volume (cc), IQR	102 (63, 180)
Median BED PTV Volume (cc), IQR	146 (88, 240)
Median BED CTV Dmax (Gy), IQR	76 (68, 88)
Median BED PTV Dmax (Gy), IQR	76 (68, 87)
Median BED CTV Dmean (Gy), IQR	64 (57, 71)
Median BED PTV Dmean (Gy), IQR	63 (56, 70)
Median BED CTV Dmin (Gy), IQR	33 (22, 39)
Median BED PTV Dmin (Gy), IQR	31 (21, 37)

Table 2 shows the characteristics and parameters of separation surgery. Most cases (n=217; 41%) involved 2 levels of laminectomy. The most common preoperative ESCC was grade 2 (n=165; 31%), whereas ESCC 1B was the most common postoperative ESCC (n=131; 25%). Indications for surgery were neurologic in 450 cases (86%) and oncologic and mechanical in 523 cases (98%). There were 17 patients with ESCC scores of 0 and 1A. A total of 7 of them (41%) had prior overlapping RT, and the histologies were relatively radioresistant (e.g. osteosarcoma, renal cell carcinoma, and rectal adenocarcinoma) in all cases. Furthermore, 13/17 (76%) of those patients had a mechanical indication for separation surgery. The median (IQR) estimated blood loss during surgery was 350 mL (200–700 mL). The median (IQR) time between surgery and discharge was 5 days (4–7 days), and the median time (IQR) between surgery and RT was 18 days (15–23 days). Most patients completed myelogram simulation before discharge.

Table 2.

Characteristics and Parameters of Separation Surgery

Characteristic	N = 5321
Indication
Neurologic	450 (86%)
Oncologic	523 (98%)
Mechanical	523 (98%)
Number of Laminectomy Levels
1	76 (14%)
2	217 (41%)
3	200 (38%)
4	28 (5.3%)
5	7 (1.3%)
6	1 (0.2%)
7	1 (0.2%)
Preoperative ESCC
0	12 (2.3%)
1A	5 (0.9%)
1B	21 (4.0%)
1C	44 (8.3%)
2	165 (31%)
3	159 (30%)
Lumbar	124 (23%)
Postoperative ESCC
0	104 (20%)
1A	34 (6.4%)
1B	131 (25%)
1C	106 (20%)
2	32 (6.0%)
3	0 (0%)
Lumbar	124 (23%)
Estimated Blood Loss (mL)	350 (200, 700)
Time Between Surgery and Discharge (Days)	5 (4, 7)
Time Between Surgery and Myelogram Simulation (Days)	4 (3, 7)
Time Between Myelogram and RT (Days)	13 (10, 15)
Time Between Surgery and RT (Days)	18 (15, 23)

¹n (%); Median (Q1, Q3)

A total of 324 patients presented with preoperative high-grade cord compression (ESCC 2–3). Of those, 69 (21%) had myelopathy (Supplemental Table 3). The scores for Medical Research Council (MRC) for muscle strength, Eastern Cooperative Oncology Group (ECOG) performance status, and American Spinal Injury Association (ASIA) impairment scale are shown in Supplemental Table 3. The majority of patients (67%) presented with ASIA D (50% of muscles more than MRC 3/5, can raise arms or legs off bed). Fifty-six (81.1%) of patients had improvement of their symptoms of myelopathy after separation surgery and SBRT.

There were 73 instances of LF. The cumulative incidence of LF (95% confidence interval (CI)) at 12, 24, and 36 months was 7.7% (5.6–10%), 10% (7.9–13%), and 13% (10–16%), respectively (Figure 2A). The median follow-up time was 74.5 months (95% CI 70–94 months). Median OS after spine SBRT was 14 months, with a 95% CI between 13–17 months (Supplemental Figure 1A). Table 3 shows the univariable and multivariable cause-specific hazards regression for LF after separation surgery and spine SBRT. On multivariable regression, the only variable associated with a statistically significant risk of LF was CTV Dmin BED10 (on multivariable regression, hazard ratio [HR] 0.96, 95% CI 0.93–0.99, p = 0.005). Class C histology had a higher risk of LF (HR 3.41, 95% CI 1.26–9.23, p=0.016) relative to Class A histology on univariable analysis.

Figure 2. Cumulative Incidence of Local Failure after Separation Surgery and Spine SBRT for Spinal metastases.

A) Cumulative incidence of local failure overall after spine SBRT. B) Cumulative incidence of local failure by CTV minimal dose BED10 above or below median for all histologies. C) Cumulative incidence of local failure by CTV minimal dose BED10 above or below median for non-breast and non-prostate histologies D) Cumulative incidence of local failure by CTV minimal dose BED10 above or below median for breast and prostate cancer lesions.

Table 3.

Univariable and Multivariable Competing Risks Regression for Local Failure

	Univariable Regression				Multivariable Regression
Characteristic	N	HR	95% CI	p-value1	N	HR	95% CI	p-value1
BED10 (Gy)	532	1.00	0.98, 1.02	>0.9
Class
Class A	76	–	–		51	–	–
Class B	421	1.54	0.70, 3.38	0.3	313	1.77	0.64, 4.91	0.3
Class C	35	3.41	1.26, 9.23	0.016	25	3.00	0.79, 11.4	0.11
Postoperative ESCC
0–1A	138	–	–		129	–	–
1B-1C	237	1.42	0.81, 2.49	0.2	230	1.28	0.70, 2.35	0.4
2–3	32	1.41	0.58, 3.42	0.4	30	1.44	0.54, 3.85	0.5
Prior Overlapping RT
No	425	–	–		310	–	–
Yes	107	1.37	0.81, 2.33	0.2	79	0.78	0.38, 1.61	0.5
CTV Volume (cc)	513	1.07	0.86, 1.33	0.5
PTV Volume (cc)	516	1.04	0.87, 1.25	0.7
CTV DMin BED10 (Gy)	511	0.97	0.95, 0.99	0.004	389	0.96	0.93, 0.99	0.005
PTV DMin BED10 (Gy)	512	0.97	0.95, 1.00	0.029
CTV DMean BED10 (Gy)	511	1.00	0.98, 1.02	>0.9
PTV DMean BED10 (Gy)	512	1.00	0.98, 1.02	>0.9
CTV DMax BED10 (Gy)	511	1.00	0.99, 1.02	0.5
PTV DMax BED10 (Gy)	512	1.00	0.99, 1.02	0.7
CTV D95% BED10 (Gy)	513	0.99	0.96, 1.01	0.3
PTV D95% BED10 (Gy)	515	0.99	0.96, 1.01	0.3

¹Cox Proportional Hazards Model

Abbreviations: CI = Confidence Interval, HR = Hazard Ratio

Figure 2B shows the cumulative incidence of LF after separation surgery and spine SBRT stratified by CTV Dmin BED10 above or below the median of 32.75 Gy. The 1-, 2-, and 3- year cumulative incidence of LF (95% CI) were 9.6% (6.3–14%), and 13% (9.2–18%), and 16% (11–21%) for lesions treated to a CTV Dmin BED below the median and 4.8% (2.6–7.9%), 6.5% (3.9–10%), and 8.9% (5.7–13%) for lesions treated to a CTV Dmin greater than or equal to the median, respectively (p=0.025). Figure 2C shows the cumulative incidence of LF after separation surgery and spine SBRT stratified by CTV Dmin BED10 above or below the overall median, limited to Class B and C (non-breast and non-prostate) histologies. The 2-year rate of LF was 14% (95% CI 9.4–19%) for lesions treated to a CTV Dmin BED10 below the median, compared to 5.6% (95% CI 3–9.3%) for lesions treated to a CTV Dmin BED10 greater than or equal to the median (p=0.010).

For previously irradiated lesions, the CTV Dmin BED10 was lower (median 19.65 Gy). For lesions treated to a CTV Dmin BED10 greater than or equal to the median, there was a decreased risk of LF (2-year risk 8.2% (95% CI 2.6–18%) vs. 18% (95% CI 9.1–31%)) compared to lesions treated to a CTV Dmin BED10 less than the median, but this did not reach statistical significance (p=0.14, Supp Fig 2). There was no statistically significant difference in the risk of LF for Class A lesions treated to a CTV Dmin BED10 above or below the median (2-year risk 13% (95% CI 3.8–27%) vs. 9.4% (95% CI 2.9–21%) for CTV Dmin BED10 greater than or equal to vs. less than the median, respectively (p=0.7; Fig 2D)). It should be noted that a higher proportion of the Class A lesions were treated previously with RT (n=23 (30%), compared with n=77 (18%) for Class B and n=7 (20%) of Class C lesions, Chi-squared p=0.057).

Postoperative ESCC was not independently associated with LF on cause-specific hazards regression (Table 3). However, postoperative ESCC was associated with dose deliverability. After exclusion of previously irradiated lesions, increasing postoperative ESCC grade was associated with progressively lower achievable CTV Dmin BED10, with a significant overall difference across ESCC categories (Kruskal–Wallis p=0.044; Supplemental Table 2). When CTV minimum BED10 was dichotomized at the cohort median (32.75 Gy) and lesions previously irradiated were excluded, postoperative ESCC ≤1C was significantly associated with the ability to achieve this dose threshold. Lesions with ESCC ≤1C were approximately twice as likely to achieve CTV Dmin BED10 ≥32.75 Gy compared with lesions with ESCC 2 (odds ratio [OR] 2.06, 95% CI 1.31–2.26, p=0.002). Further restriction of postoperative ESCC to ≤1B did not improve dose deliverability compared with ESCC ≥1C (OR 0.75, 95% CI 0.50–1.11, p=0.15).

Supplemental Table 2 shows the cumulative incidence of HF requiring surgical intervention, RM, and VCF after separation surgery and postoperative spine SBRT. At 2 years, the cumulative incidence of of VCF was 2% (95% CI 1–3.5%), the cumulative incidence of RM was 1% (95% CI 0.39–2.3%), and the HF requiring intervention was 3.8% (95% CI 2.4–5.8%).

Discussion

This is the largest retrospective cohort study published to date of patients treated with separation surgery followed by spine SBRT, a total of 497 patients treated to 532 lesions, with a median follow-up time (95% CI) of 74.5 months (70–94 months) and a median OS (95% CI) of 14 months (13–17 months). These numbers are similar to previous publications examining this patient population.^6,7,12–14 Overall, separation surgery followed by spine SBRT is associated with excellent long term local control rates (1-year LF rate 7.7%) and a low risk of serious side effects.

Among the 324 patients who presented with high-grade epidural spinal cord compression (ESCC grade 2–3), 69 (21%) had myelopathy attributable to malignant cord compression at presentation. Following separation surgery and spine SBRT, 56 of these 69 patients (81%) experienced neurological improvement within weeks to months. This supports the role of this approach in patients presenting with cord compression and neurological deficit.

Laufer et al. published in 2013 a series of 186 patients treated with separation surgery followed by SBRT, showing a cumulative incidence of local failure (LF) of 16.4% 12 months after SBRT, with decreased risk of LF associated with higher biological effective dose (BED), and no difference in risk of local failure by histology-specific radiosensitivity.⁷ Al-Omair et al published in 2013 a series of 80 patients treated to 18–26 Gy in 1–2 fractions following separation surgery found a 1-year local control rate of 84%, with the most common site of failure in the epidural space.¹⁵ In 2018, Ito et al published a retrospective series of 28 cases of patients treated with spine SBRT and separation surgery after prior overlapping RT to the same spinal area. The authors found a 1-year local control rate of 70%, with 1 instance of radiation myelitis and 3 instances of VCF.¹⁶ In 2020, Redmond and colleagues published a phase II clinical trial of 35 patients with spinal metastases (patients with prior overlapping spine RT were excluded) treated with separation surgery followed by spine SBRT (30 Gy in 5 fractions). The 1-year local control rate was 90%; 3 patients developed local recurrence, all with radiosensitive (to CFRT) tumor histologies, and all in the epidural space.¹³ In 2022, Ito et al. published another phase II clinical trial of 33 patients treated with separation surgery followed by spine SBRT (24 Gy in 2 fractions). The 12-month LF rate was 13%, and adverse events were minimal.¹² Chakravarthy and colleagues published a series of 104 patients with NSCLC spinal metastases treated with separation surgery and SBRT from 2012–2019 and found a local control rate of 95% 2 years after surgery plus SBRT,¹⁴ as well as another series of 50 patients with colorectal cancer treated with separation surgery followed by SBRT with a local control rate of 86.7% 2 years after therapy.

Separation surgery and postoperative spine SBRT is associated with low risk of adverse events. The cumulative incidences of side effects of VCF, RM, and HF requiring surgical intervention were all in the single digit percentages 8 years after separation surgery and spine SBRT. This result is further strengthened by previous retrospective series^7,16 and early-phase clinical trials,^12,13 supporting separation surgery and postoperative spine SBRT as a modality of care for patients with certain indications.

Regarding the indications for separation surgery followed by spine SBRT, most patients had some component of neurologic, oncologic (including receipt of prior overlapping RT in 20% of cases), or mechanical factors as reasons to proceed with this treatment paradigm. The NOMS framework continues to serve as a guiding strategy to approach multidisciplinary discussions for these patients.¹ We have also provided our institutional single, 3-, and 5-fraction constraints, both in the upfront and re-irradiation settings, for reference. The most salient result of this study is the strength of the association between CTV Dmin BED10 and risk of LF. As the spinal cord constraint is the primary driver of minimal dose to the target volume with spine SBRT,¹⁷ it has previously been shown that minimal dose to target volumes is associated with risk of LF.^18,19 Prior postoperative SBRT series consistently demonstrate that local failure most commonly occurs within the epidural space.¹⁵ Chan et al. further showed that postoperative ESCC and epidural disease location alone do not predict failure patterns; rather, high-grade preoperative epidural compression was associated with posterior epidural progression.²⁰

Our findings extend these observations by providing a mechanistic link between postoperative anatomy and oncologic outcome. Specifically, we demonstrate that achieving a postoperative ESCC of 1C or lower is associated with delivery of a typical median CTV Dmin BED10, whereas further decompression beyond ESCC 1C does not confer additional benefit in dose deliverability. Importantly, CTV Dmin BED10—not postoperative ESCC itself—was the variable most strongly associated with local control on competing-risks analysis. This suggests that postoperative ESCC influences local control indirectly by governing the dose that can be safely delivered to the epidural target adjacent to the spinal cord.

These findings have important clinical implications. For patients undergoing separation surgery followed by SBRT—particularly those with non-breast and non-prostate histologies or prior overlapping radiation—achieving postoperative ESCC ≤ 1C appears sufficient to enable delivery of an effective minimum biological dose. More aggressive decompression to ESCC ≤ 1B did not improve dose deliverability and therefore may not be necessary in all cases. This supports a pragmatic surgical goal that balances neurologic decompression, surgical morbidity, and postoperative radiation feasibility.

Limitations

This study is limited by its retrospective design and the inherent potential for selection bias. Fractionation schemes and prescription doses were heterogeneous, reflecting real-world practice across an extended period. Although competing-risks methods were used to account for death as a competing event, unmeasured confounding cannot be excluded. For example, systemic therapy is an important factor for local control and has evolved significantly over the study period. While systemic therapy is beyond the scope of the current publication, we plan to examine systemic therapy in a histology-specific fashion in subsequent manuscripts.

Conclusions

Separation surgery followed by spine SBRT is associated with excellent local control and a low risk of serious adverse events. In this large cohort, CTV Dmin BED10 most strongly correlated with local control, providing a mechanistic explanation for treatment success. Achieving postoperative ESCC ≤ 1C appears sufficient to permit delivery of an effective minimum biological effective dose near the spinal cord, whereas further decompression does not confer additional benefit. These findings support postoperative ESCC ≤ 1C as a practical surgical target and warrant validation in other large series and prospective clinical trials.

DATA SHARING STATEMENT

Data contain PHI and therefore will not be shared.