Radionecrosis risk and local control after CyberKnife^® fractionated stereotactic radiotherapy for brain metastases: an eight-year single-centre cohort

Abstract

Background

Stereotactic radiotherapy (SRT) is an effective treatment for patients with a limited number of brain metastases (BM). However, radionecrosis (RN) remains a concern, particularly in long-term survivors, as its distinction from tumor progression is challenging and may lead to inappropriate salvage strategies. Fractionated stereotactic radiotherapy (FSRT) has been proposed to improve the therapeutic ratio for larger lesions or postoperative cavities, but the optimal dose and fractionation remain uncertain.

Materials and methods

We conducted a retrospective single-center study involving all consecutive patients who received SRT with CyberKnife^® for BM between January 1, 2016, and December 31, 2023. Local control (LC) and RN were assessed on MRI according to RANO-BM criteria; RN was graded using CTCAE v5.0.

Results

A total of 131 patients with 197 BM were analyzed with a median follow-up of 41.7 months (range, 1.7–207.8). Median age was 63.4 years; lung cancer was the most common primary (70.2%). Most patients had a single BM (68.8%) and received treatment in the first-line metastatic setting (70.2%). At the lesion level (n = 197), most targets received 27 Gy in 3 fractions (52.3%) or 30 Gy in 5 fractions (45.7%). Crude LC was 85.3%, with 29 local failures. Larger BM was significantly associated with local failure (HR 1.03; 95% CI 1.01–1.05; p = 0.017). Median brain progression-free survival (bPFS) was 9.8 months (95% CI 7.4–13.6). Brain progression correlated with the initial number of BM (HR 1.41; 95% CI 1.10–1.8; p = 0.006). RN occurred in 42 patients (32.1%), symptomatic in 14.5% cases (11.4% grade 2, and 3.1% grade 3). Larger size of BM (HR 1.58; 95% CI 1.04–2.49; p = 0.048) and prior surgery (HR 2.12; 95% CI 1.05–4.33; p = 0.037) were independent predictors of RN.

Conclusion

FSRT achieved durable LC with acceptable rates of symptomatic RN in this cohort. Tumor volume and prior surgery were associated with both LC and RN, supporting individualized dose–fractionation strategies based on lesion size and clinical context. Further prospective data are needed to refine dose constraints and integrate evolving therapies.

Background

The occurrence of brain metastases (BM) is common, affecting up to 40% of patients. Local management of BM remains a critical issue, given the substantial functional impact, poorer prognosis, and the limited penetration of many systemic therapies across the blood–brain barrier [1–3]. Their management relies on surgery, radiotherapy (RT), and systemic therapies. For decades, whole-brain radiotherapy (WBRT), typically 30 Gy in 10 fractions (fx), was the standard of care for patients with BM. However, WBRT is now considered suboptimal due to its association with neurocognitive decline in the mid- to long term and improvement in survival with new systemic treatments. Stereotactic radiotherapy (SRT) allows high-dose irradiation of limited intracranial targets with uninvolved brain parenchyma sparing. The oncological benefits of sparing uninvolved brain tissue remain uncertain. SRT can be delivered either as single-fraction stereotactic radiosurgery (SRS) or fractionated stereotactic radiotherapy (FSRT) over 2–10 fx. Randomized trials comparing SRS or surgery with or without WBRT in patients with 1–3 (up to 4) BM have shown that adding WBRT improves distant intracranial control but does not prolong overall survival (OS) and significantly impairs neurocognitive function [4–8]. With SRS or surgery plus WBRT, relapse occurred more frequently in patients treated with SRS alone with no significant difference in OS. Consequently, current guidelines favor focal treatment alone (SRT or surgery) in patients with a limited number of BM and good performance status, reserving WBRT for selected cases. Surgical resection should be considered for BM causing mass effect, lesions ≥ 3.5 cm, or when tissue diagnosis is warranted, followed by postoperative RT. For patients with 5 or more BM, optimal local treatment remains controversial [9, 10]. High-dose focal RT can lead to radiation necrosis (RN), with risk influenced by dose, fractionation, target volume, and prior treatments. RN impact patient’s quality of life, notably in long-term survivors, where symptomatic RN may emerge. Studies reported up to 34% of RN at 24 months after SRT with significant morbidity in 10–17% of cases [11, 12]. Proposed mechanisms of RN include radiation-induced endothelial injury, blood–brain barrier disruption, and subsequent inflammatory and hypoxic cascades. Management of symptomatic RN remains suboptimal: long-term corticosteroid therapy, anti-VEGF monoclonal antibodies, hyperbaric oxygen, or thermal laser ablation carry only limited evidentiary support [13].

RN may present with clinical and radiological findings similar to those of tumor progression, making diagnosis difficult. It can lead to inappropriate therapies. In select cases, surgery can also relieve symptomatic RN but patients with advanced disease, poor performance status, deep-seated or eloquent‐area lesions are not suitable for such a resection. FSRT has been increasingly used, particularly for lesions > 2–3 cm or postoperative cavities, in an attempt to maintain high LC while reducing RN [14]. The aim of this study was to report long-term outcomes of patients with BM treated with SRT at our institution over an eight-year period, focusing on LC, RN incidence and risk factors, brain control, survival, and leptomeningeal disease (LMD), considering contemporary literature.

Materials and methods

Study period and population

We conducted a retrospective single-center analysis including all consecutive patients who received CyberKnife^® SRT for BM at Georges Pompidou European Hospital (HEGP) between January 1, 2016, and December 31, 2023.

Inclusion and exclusion criteria

Eligible patients were: 18 years or older, treated for BM with exclusive or post-operative intracranial SRT. Histological confirmation of the primary tumor was required. Patients with a history of WBRT could be included. Patients with primary malignant or benign intracranial tumor were excluded.

Data collection

Clinical data were extracted from electronic medical records and included: age, sex, performance status, primary tumor histology, extracranial disease control, high cardiovascular risk factors (hypertension, diabetes mellitus, dyslipidemia with elevated LDL cholesterol, current or past smoking, obesity, as defined by the 2021 European Society of Cardiology Guidelines [15]. Concurrent systemic therapy was defined as any systemic treatment administered during SRT or within one month of SRT. The Graded Prognostic Assessment (GPA) score was calculated for each patient according to cancer subtype. Extracranial disease status was classified as uncontrolled in cases of extracerebral recurrence or progression at the time of SRT. For intracranial disease, the number, anatomical location, and gross tumor volume (GTV) of BM were recorded.

Radiotherapy technique

The indication for SRT was reviewed and confirmed during a multidisciplinary tumor board meeting. SRT was delivered by CyberKnife^® system with thermoplastic mask immobilization. GTV was delineated on pre-treatment contrast-enhanced T1-weighted (CE-T1w) MRI sequences with 1 mm slice thickness using Eclipse™ (Varian) and planned with Precision^® (Accuray). A 0–2 mm margin was added to the GTV to create the planning target volume (PTV). A 0-mm margin was used as the institutional standard; a 1–2 mm margin was applied selectively when the MRI–SRT interval exceeded 15 days, when MRI–CT fusion confidence was suboptimal, or for postoperative resection cavities when cavity boundaries were not clearly delineated on MRI. Margin expansion remained selective and non-systematic and was performed at the discretion of the treating radiation oncologist, in agreement with the medical physicist. Dose prescription was adapted to the size and the location of the tumor: lesions ≤ 1 cm received 18 Gy in a single fx. Brainstem metastases were treated with 25 Gy in 5 fx. For lesions near critical structures, post-operative cavities, or following WBRT, 30 Gy in 5 fx was used; otherwise, 27 Gy in 3 fx were preferred. All plans used an 80% prescription isodose line to cover ≥ 95% of the PTV. Dosimetric parameters for normal brain tissue were calculated using a brain minus PTV structure, consistent with our institutional workflow throughout the study period, to ensure longitudinal reproducibility across evolving planning practices and to avoid including the high-dose target volume in normal-brain metrics. As part of institutional practice, V12Gy was routinely recorded for all plans, but it was considered a limiting constraint only for single-fraction treatments. For multi-fraction regimens, planning objectives focused on fractionation-specific normal brain metrics (brain minus PTV), including V23.1 < 7 cm³ and V20Gy < 20 cm³ in 3-fx SRT, V24Gy < 20 cm³ and V28.8 < 3–7 cm³ for 5-fx regimen. These objectives reflect institutional practice informed by published multi-fraction tolerance series and national practice (RecoRad guidelines and other FSRT cohorts) [16, 17]. The maximum point dose and PTV coverage were recorded.

Follow-up

Intracranial follow-up consisted of brain MRI every 3 months during the first 2 years and every 6 months thereafter. Extracranial disease was re-evaluated every 3 months for the first 2 years and every 6 months thereafter using CT-TAP and/or 18FDG PET/CT. The following variables were collected: date of each clinical visit, neurological status, MRI findings, relapse treatment, and date of progression or death.

Endpoints

Primary objectives were to evaluate LC and RN. Local control (LC) was defined as the absence of local failure (LF), with LF defined as radiological progression within the planning target volume (PTV). LC was assessed on MRI according to the Response Assessment in Neuro-Oncology Brain Metastases (RANO-BM) criteria. Local failure-free survival (LFFS) was estimated from end of SRT to LC of a treated lesion. Crude local control was the overall rate of LF during the entire study period. RN diagnosis was based on surveillance brain MRI combined with clinical evaluation or histopathological confirmation. In cases of diagnostic uncertainty between RN and tumor progression, additional advanced imaging (MRI spectroscopy or perfusion) was performed at the treating physician’s discretion. All MRI and corresponding reports were reviewed by an experienced radiation oncologist. Stability or regression of the enhancing lesion over serial imaging without treatment escalation was considered in favor of RN. RN was graded according to the Common Terminology Criteria for Adverse Events (CTCAE v5.0).

Secondary endpoints included: RN-free survival (RnFS) defined as the interval from the end of SRT to RN diagnosis; brain progression-free survival (bPFS), defined as the interval from the end of SRT to any intracranial progression (local failure or new BM); progression-free survival (PFS), defined as the interval from the end of SRT to progression at any site; overall survival (OS), defined as the interval from the start of SRT to death or last follow-up. We reported the incidence of leptomeningeal disease (LMD); and the proportion of deaths attributed to neurological causes.

All time-to-event and toxicity endpoints (OS, bPFS, LFFS, and RN) as well as all univariable and multivariable analyses were performed at the patient level using cumulative brain and dose–volume metrics, whereas crude LF counts were summarized at the lesion level. This approach avoids unit-of-analysis bias.

Exploratory comparisons were prespecified as follows. Among patients with a calculable histology-specific DS-GPA, we performed an exploratory benchmarking analysis against published median survival estimates [15–17]. Exploratory comparisons were prespecified as follows. Among patients with a calculable histology-specific DS-GPA, we performed an exploratory benchmarking analysis against published median survival estimates.

Statistical analysis

No variable prioritized for multivariable adjustment had more than 5% missing data. Univariable comparisons were performed using the Wilcoxon/Mann–Whitney test for continuous variables (reported as median with interquartile range (IQR)) and the Chi-square or Fisher’s exact test for categorical variables (reported as n (%)). Dose schedule (18, 27, or 30 Gy) was analyzed as a categorical variable and compared using Fisher’s exact test due to small cell counts in certain groups. Variables with p < 0.05 in exploratory analyses or considered clinically relevant were entered into multivariable Cox proportional hazards models; variables with sparse exposure were excluded to avoid model instability. Results of multivariable analyses are presented as hazard ratios (HR) with 95% confidence intervals (CI). A p-value < 0.05 was considered statistically significant. For DS-GPA benchmarking, we performed a descriptive, group-level calibration by comparing observed median OS within each histology- and GPA-defined stratum to the corresponding published diagnosis-specific GPA–predicted median OS. Observed-to-predicted median OS ratios were calculated to summarize group-level consistency. Because the diagnosis-specific GPA provides population-based median survival estimates rather than individual survival predictions, rank-based correlation analyses were not used as the primary performance metric. Analyses were performed using R software (version 4.2.1).

Ethics approval and declarations

The study was approved by the Ethics Committee of HEGP, Paris, France. Owing to the retrospective design and the high proportion of deceased patients at the time of chart review, the requirement for informed consent was waived by the ethics committee. For surviving patients seen in follow-up, an information notice and opt-out option were provided; all data were de-identified before analysis. The study was conducted in accordance with the Declaration of Helsinki.

Results

Population

A total of 131 patients with 197 BM were analysed (Table 1). The median age was 63.4 years (IQR, 57.1–72.8; range, 26–92), with 60.3% male patients. Lung cancer was the most common primary tumor (70.2%), followed by renal cell carcinoma (12.2%) and gastrointestinal cancers (11.5%). Eighty-eight patients, (67.2%) presented with a single lesion, and 70.2% received the first-line treatment. Prior WBRT and surgery had been performed in 10 (7.6%) and 19 (14.5%) patients, respectively. At the lesion level (n = 197), most targets received 27 Gy in 3 fx (52.3%) or 30 Gy in 5 fx (45.7%), while other schedules accounted for 4 lesions (Table 1). The mean cumulative GTV per patient at first SRT was 5.5 ± 6.8 cm³ (range, 0.1–30.9); the median was 2.4 cm³ (IQR, 0.5–8.0).

Table 1.

Baseline patient and treatment characteristics

Variable	N (%) *
Sex
Male	79 (60.3)
Female	52 (39.7)
Age
Median (IQR)	63.4 (57.1; 72.8)
Min-Max	26–92
Baseline KPS
≥ 70	94 (71.8)
< 70	6 (4.6)
NR	31 (23.7)
High Cardiovascular Risk Patient*
Yes	39 (29.8)
NR	7 (5.3)
GPA index
Mean ± SD	2.4 ± 0.8
Median	2.5
Estimated Median Survival (months)	15
Mean ± SD	17.9 ± 10.6
NA	12 (9.2)
Primary tumor histology
Lung overall	92 (70.2)
Lung adenocarcinoma	68 (51.9)
Lung epidermoid carcinoma	12 (9.2)
Lung neuroendocrine carcinoma	12 (9.2)
Ovarian carcinoma	2 (1.5)
Invasive breast carcinoma	4 (3.1)
Gastrointestinal carcinoma	15 (11.5)
Clear cell renal cell carcinoma	16 (12.2)
Sarcoma	2 (1.5)
Molecular biomarkers (genomic and PD-L1)
PDL-1 positive*	38 (29.0)
TP53	14 (10.7)
KRAS	14 (10.7)
EGFR	13 (9.9)
ALK	4 (3.1)
BRAF	4 (3.1)
HER 2	4 (3.1)
SMARCA4	4 (3.1)
STK11	1
MET	1
Hormonal receptors	1
ROS1	1
NR	22 (16.8)
Therapeutic line at first SRT
First	92 (70.2)
Second	28 (21.4)
Third and more	11 (8.4)
Clinical stage at first SRT
Oligometastatic	87 (66.4)
Multimetastatic	44 (33.6)
Extracranial disease status
Controlled	70 (53.4)
Active or new diagnosis	61 (46.6)
Prior brain treatment
WBRT	10 (7.6)
Resection of the BM	19 (14.5)
Radiotherapy schedules by BM ( n  = 197)
Total dose / number of fractions
27 Gy * 3fx	103 (52.3)
30 Gy * 5fx	90 (45.7)
25 Gy * 5fx	3 (1.5)
18 Gy * 1fx	1
Number of BM by patient at first SRT
1	88 (67.2)
2	22 (16.8)
3	15 (11.5)
≥ 4	6 (4.6)
Cumulative GTV (cm³)
Mean ± SD; (Min-Max)	5.5 ± 6.8; (0.1–30.9)
Median (IQR)	2.4 (0.5; 8)
Patients with a cumulative intracranial GTV ≥ 8 cm³	29 (22.1)
Previous surgery volume group (cm³)
Mean ± SD	15.2 ± 8.0
No surgery volume group (cm³)
Mean ± SD	3.9 ± 5.3
Concomitant systemic therapy
Yes	99 (75.6.3)
No	29 (22.1)
NR	3 (2.3)
Systemic therapy at the time of first SRT
Multi-agent chemotherapy	34 (25.9)
None	21 (16.2)
Immunotherapy	23 (17.7)
Chemoimmunotherapy	15 (11.5)
Tyrosine kinase inhibitor	15 (11.4)
Single-agent chemotherapy	13 (9.9)
Crizotinib	4 (3.1)
PARPi	1 (0.8)
Anti-HER therapy	2 (1.5)
NR	3 (2.3)

* Notes: Values are reported as n (%). Analyses were primarily performed at the patient level (n = 131), unless otherwise specified (lesion-level analyses by brain metastasis). Continuous variables are mean ± SD or median interquartile range (IQR). PDL1 positive: Concomitant (peri-SRT) therapy was defined as treatment delivered within 30 days before or after SRT. Cardiovascular risk as defined by the 2021 European Society of Cardiology

BM: brain metastasis; FSRT: fractionated stereotactic radiotherapy; Fx: fraction; GPA score: histology-specific graded prognostic assessment score (Non-squamous CLC GPA, Breast GPA, Melanoma GPA, Renal cell carcinoma GPA, Gastrointestinal GPA, as applicable); KPS: Karnofsky Performance Status; NA: not applicable; NR: not reported; PARPi: poly(ADP-ribose) polymerase inhibitor; RT: radiotherapy; SD: standard deviation; SRT: stereotactic radiotherapy; WBRT: whole-brain radiotherapy

Local control and brain progression

At 12 months, LC probability was 77.8%. Overall, 29 of 131 patients experienced at least one LF. Crude local control over the entire follow-up was 85.3% (168/197 treated targets). Lesions with LF had significantly larger GTV than controlled lesions (median GTV 5.3 cm³ [1.4–12.4] vs. 1.9 cm³ [0.4–5.9]; p = 0.008). In contrast, histology, prior WBRT, prior surgery, and dose regimen were not significantly associated with LF on univariable analysis (Table 2).

Table 2.

Univariable comparison between patients with and without local failure

Variable	No local Failure (n = 84)	Local Failure (n = 33)	P- value
Primary tumor histology			0.146¹
Pulmonary adenocarcinoma	43 (51.2%)	16 (48.5%)
Pulmonary squamous cell carcinoma	7 (8.3%)	3 (9.1%)
Ovarian carcinoma	2 (2.4%)	0
Invasive breast carcinoma	2 (2.4%)	2 (6.1%)
Clear cell renal cell carcinoma	13 (15.5%)	2 (6.1%)
Prostate adenocarcinoma	1	0
Digestive carcinome	6 (2.4%)	6 (9.1%)
Small-cell lung carcinoma	10 (11.9%)	2 (6.1%)
Sarcoma	0	2 (6.1%)
PD-L1 expression (%), median [IQR]	2 [0–60]	20 [0–60]	0.4693
Oligometastatic disease	57 (69.5%)	20 (60.6%)	0.484²
Extracranial disease controlled	40 (47.6%)	22 (66.7%)	0.099²
Prior WBRT	8 (9.5%)	1 (3.1%)	0.441¹
Prior surgery	12 (14.5%)	7 (21.9%)	0.497²
Total SRT dose (Gy)			0.1021
18 Gy	1	0
27 Gy	55(65.5%)	18 (54.5%)
30 Gy	28 (33.3%)	15 (45.5%)
GTV (cm³), median [IQR]	1.9 [0.4–5.9]	5.3 [1.4–12.4]	0.008³
Time between MRI and SRT (days), median [IQR]	16.3 [7.8–26.9]	14.2 [8.4–21.7]	0.509³

¹ Fisher’s exact test; ² chi-square test; ³ Wilcoxon–Mann–Whitney test (patient-level analyses; n = 131). Gy: gray; GTV: gross tumor volume; IQR: interquartile range; RN: radiation necrosis; SRT: stereotactic radiotherapy; WBRT: whole brain radiotherapy

On multivariable Cox regression, GTV remained independently associated with an increased risk of LF (HR per cm³ = 1.03; 95% CI 1.01–1.05; p = 0.017) (Table 3). BPFS at 1 and 2 years was 36.8% and 12.8%, respectively, with a median of 9.79 months (95% CI 7.4;13.6) (Fig. 1). Brain progression was mainly driven by the occurrence of new BM. On univariate analysis, shorter bPFS was associated with a higher number of BM at baseline (p = 0.048) (Table 4). This effect remained robust in multivariable analysis (HR 1.41; 95% CI 1.10–1.80; p = 0.006). In 43.2% of patients, brain progression was associated with concomitant extracranial progression. At 18 months, 52 patients (64%) presented with brain progression involving fewer than three metastases; among them, 32 received salvage SRT, 11 underwent surgical resection, and 17 were treated with salvage WBRT. Overall, 93% of patients experienced one episode of brain progression before the end of follow-up or death. LMD occurred in 12 patients (9.2%), including 6 patients (31.6%) in the postoperative group (n = 19) compared with 6 (5.4%) in the non-surgical group (Fisher’s exact test, p = 0.002) (Table 5).

Table 3.

Multivariate analysis

Endpoint	Variable	HR (95% CI)	P-value
Local failure free survival	GTV (per cm³)	1.03 (1.01–1.05)	0.017
Brain progression free survival	Number of BM (per lesion)	1.41 (1.10–1.80)	0.006
Brain radionecrosis	Prior surgery (yes against no)	2.12 (1.05–4.33)	0.037
Brain radionecrosis	GTV (cm³)	1.58 (1.04–2.49)	0.048

Cox proportional hazards models

Notes: Multivariable Cox models include variables with p < 0.05 in univariate analysis and relevant clinical covariates; proportional hazards assumption verified. BM: brain metastase; GTV: gross tumor volume

Fig. 1.

Kaplan–Meier survival curves after stereotactic radiotherapy for brain metastases. (A) Overall survival; (B) Brain progression-free survival; (C) Radionecrosis-free survival; (D) Local failure-free survival

Table 4.

Univariable comparison between patients with and without brain progression

Variable	No brain PD (n = 42)	Brain PD (n = 84)	P-value
Primary tumor histology			0. 938¹
Pulmonary adenocarcinoma	22 (52.4%)	44 (52.4%)
Pulmonary squamous cell carcinoma	3 (7.1%)	6 (7.1%)
Ovarian carcinoma	0	2 (2.4%)
Invasive breast carcinoma	2 (4.8%)	2 (2.4%)
Digestive carcinoma	12 (28.5%)	24 (28.6%)
Clear cell renal cell carcinoma	5 (11.9%)	11 (13.1%)
Prostate adenocarcinoma	0	1
Small-cell lung carcinoma	3 (7.1%)	3 (3.6%)
Sarcoma	0	2 (2.4%)
PD-L1 expression (%), median [IQR]	15 [0–60]	1 [0–60]	0.8143
Oligometastatic	28 (68.3%)	57 (68.7%)	1²
Uncontrolled extracerebral disease	22 (53.7%)	45 (53.6%)	1²
Prior WBRT	3 (4.1%)	7 (12.5%)	0.7921
Prior surgery	8 (19%)	11 (13.4%)	0.575²
Number of brain metastases, median [IQR]	1 [1–1]	1 [1–2]	0.048³
GTV (cm³), median [IQR]	2.8 [0.4–8.9]	2.2 [0.5–7.7]	0.817³
Total SRT dose (Gy)			0.7771
18 Gy	0	1
27 Gy	29 (69.0%)	57 (67.9%)
30 Gy	13 (31.0%)	26 (31.0%)
Time between MRI and SRT (days), median [IQR]	19 [13–27]	18 [12–26]	0.6193
Concomitant systemic therapy	26 (61.1%)	43 (51.2%)	0.653¹

¹ Fisher’s exact test; ² chi-square test; ³ Wilcoxon–Mann–Whitney test (patient-level analyses; n = 131). Gy: gray; GTV: gross tumor volume; IQR: interquartile range; PD: progression; RN: radiation necrosis; SRT: stereotactic radiotherapy; WBRT: whole brain radiotherapy

Table 5.

Association between prior surgery and leptomeningeal disease

Variable	leptomeningeal disease (n)	No leptomeningeal disease (n)	Total	P-value
Prior surgery	6	13	19	0.0021
No prior surgery	6	106	112
Total	12	119	131

Notes: leptomeningeal disease defined by institutional imaging/clinical criteria. N = number of patients. ¹ Fisher’s exact test

Radionecrosis

RN was diagnosed in 42 of 131 patients (32.1%). RN after SRT was histologically confirmed in three BM while the remaining 97.7% were diagnosed based on imaging findings combined with clinical arguments: reflecting real-world practice. The median time to RN diagnosis was 12 months (range: 6–24 months). RN was CTCAE v5.0 grade 1 (asymptomatic) in 23 patients (17.5%), grade 2 in 15 (11.4%), and grade 3 in 4 (3.1%). No grade 4–5 events were observed. In univariable comparisons, RN was more frequent in patients with prior surgery (n = 6 (8.2%) versus n = 13 (23.2%), p = 0.037) and median GTV (median: 1.7 [0.3–5.9] versus 4.0 [0.8–9.4], p = 0.044), while no significant association was observed with dose regimen, number of BM, or prior WBRT (Table 6). In multivariate analysis, both GTV size (HR 1.58; (1.04–2.49); p = 0.048) and previous neurosurgical resection (HR 2.12; IC95 (1.05–4.29); p = 0.037) were independent predictors of RN (Table 3). RN cases were commonly located in the cerebellum (n = 12) or parietal (n = 9) lobes. RnFS rates at 1 and 2 years were 41.2% and 16.8%, respectively; median RnFS was not reached within follow-up (Fig. 1). Fifteen symptomatic patients requiring prolonged corticosteroid therapy (≥ 1 year in 70% of cases). Four patients (3.1%) benefited from bevacizumab; one underwent surgical resection for symptomatic RN with histologic proof. Four cases of grade 3 RN were observed. Three occurred in patients with lung adenocarcinoma (single lesion in two cases and three lesions in one case), and one in a patient with SCLC. Treatment regimens were 30 Gy in 5 fx (n = 2) and 27 Gy in 3 fx (n = 2). Prior WBRT was documented in one patient (SCLC case), and prior surgery in one patient (adenocarcinoma with three lesions, including one resected lesion). Brain minus PTV constraints were met in three of the four cases; in the remaining case (30 Gy in 5 fx), V24 reached 21.2 cm³, with two PTV separated by less than 1 cm. Lesion volumes at RN diagnosis ranged from 1.4 to 18.3 cm³.

Table 6.

Univariable comparison between patients with and without radionecrosis

Variable	No brain RN (n = 74)	Brain RN (n = 55)	P-value
Sex (male versus female)	44 (59.5%)	34 (61.8%)	0.929²
High cardiovascular risk	25 (33.3%)	14 (25%)	0.402²
Primary tumor histology			0.091¹
Pulmonary adenocarcinoma	44 (58.7%)	24 (42.9%)
Pulmonary squamous cell carcinoma	7 (9.3%)	4 (7.1%)
Ovarian carcinomaInvasive breast carcinoma	0 0	2 (3.6%) 4 (7.1%)
Clear cell renal cell carcinoma	6 (8%)	10 (17.9%)
Prostate adenocarcinoma	1 (1.3%)	0
Digestive carcinoma	8 (18.2%)	7 (20.6%)
Small-cell lung carcinoma	7 (9.3%)	5 (8.9%)
Sarcoma	2 (2.7%)	0
PD-L1 expression (%), median [IQR]	10 [0–60]	20 [0–65]	0.782³
Prior WBRT	3 (4.1%)	7 (12.5%)	0.101¹
Prior surgery	6 (8.2%)	13 (23.2%)	0.033²
Total SRT dose (Gy)			0.2971
18 Gy	1	0
27 Gy	49 (65.3%)	38 (67.9%)
30 Gy	25 (33.4%)	18 (32.1%)
Time between MRI and SRT (days), median [IQR]	9 [7–11]	9 [7–12]	0.358³
Number of brain metastases, median [IQR]	1 [1–2]	1 [1–2]	0.698³
GTV (cm³), median [IQR]	1.7 [0.3–5.9]	4.0 [0.8–9.4]	0.044³
Concomitant systemic therapy			0.653¹
None	11 (14.9%)	10 (17.9%)
Immunotherapy	14 (18.9%)	9 (16.1%)
Chemoimmunotherapy	7 (9.5%)	8 (14.3%)
Osimertinib	3 (4.1%)	4 (7.1%)
Crizotinib	4 (5.4%)	0
PARPi	0	1
Multi-agent chemotherapy	30 (40.5%)	19 (33.9%)
Anti-HER therapy	0	2 (3.6%)
Tyrosine kinase inhibitors	3 (4.1%)	2 (3.6%)

¹ Fisher’s exact test; ² chi-square test; ³ Wilcoxon–Mann–Whitney test (patient-level analyses; n = 131). Gy: gray; GTV: gross tumor volume; IQR: interquartile range; PARPi: poly(ADP-ribose) polymerase inhibitor; RN: radiation necrosis; SRT: stereotactic radiotherapy; WBRT: whole brain radiotherapy

Survival

OS rates at 1 and 2 years were 66.6% and 39.2%, respectively. Median OS was 18.3 months (95% CI 15.6–23.9). Median progression-free survival (PFS) was 4.9 months (95% CI,3.7–6.3) (Figs. 1). Death was attributed to neurological causes in 49 patients (37.4%). Among the 119 patients with a calculable DS-GPA, 90% lived beyond the median survival predicted [18–20]. Observed median OS generally followed the expected prognostic gradient across increasing GPA strata. At the group level, observed medians were broadly consistent with the diagnosis-specific GPA–predicted median survival values, while substantial within-stratum heterogeneity and small subgroup sizes accounted for variability in observed-to-predicted ratios. These findings support use of diagnosis-specific GPA as a population-level stratification framework rather than an individual survival prediction tool.

Discussion

FSRT is widely used for BM, particularly for larger lesions, postoperative cavities, and targets close to critical structures, where the intent is to preserve LC while mitigating the risk of RN [9, 10, 21–23]. Despite consistently favourable LC in contemporary series, RN remains clinically meaningful and is strongly influenced by treated volume and dose–volume exposure, while post-treatment imaging uncertainty between RN and tumour progression further complicates outcome assessment [22–24]. Asymptomatic RN is rarely captured systematically in routine follow up and is likely underreported, contributing to heterogeneity across published FSRT series [21, 25, 26]. In this context, real world cohorts with consistent planning practices and longitudinal follow-up remain informative to refine pragmatic fractionation strategies and interpret toxicity signals.

Brain control

For the moment, there are no randomized trials directly comparing LC or RN between SRS and FSRT; current evidence remains largely retrospective and heterogeneous (Table 7). Across series, per-fraction doses > 6 Gy and total doses ≥ 30 Gy in 5 fx are generally associated with better LC, whereas schedules < 30 Gy in 5 fx appears suboptimal for larger lesions. In particular, 27 Gy in 3 fx is frequently reported as an effective option for BM ≥ 2 cm, with improved LC and a lower RN signal than SRS [21, 22, 27–29]. Still, cross-study comparisons require caution given differences in target margins, dose calculation algorithms, prescription isodoses and machine used [25, 30]. In our cohort, most lesions received 27 Gy in 3-fx or 30 Gy in 5-fx regimens considered acceptable for 2–4 cm BM in contemporary guidance [9]. At 1 year, LC and brain control were 77.8% and 75.6%, with 14.7% LF. Consistent with prior reports, BM volume was the main driver of LF failure, whereas brain progression was primarily influenced by the number of BM. These findings reinforce an individualized approach in which dose and fractionation are tailored to tumor volume and disease burden rather than applied uniformly. In FSRT series, Murai et al. using 5-fx (25 Gy) regimen to BM > 4 cm and 3-fx (27 Gy) to BM between 2.5 cm and 4 cm, LC was 78% and 65% at 1-year with no grade 3 toxicities. With 27 Gy in 3 fx, Minitti et al. (n = 138 patients) and Mengue et al. (n = 427 patients) reported a 1-year LC of 73% of LC (BM> 3 cm) and 76.5% respectively [28, 31, 32]. Last HyTEC review, cited here for contextual comparison only, reported that 30 Gy in 3 fx was effective and safe, bearing in mind the report’s limitations (variation in LC definition between series, technique, prescription isodose, PTV expansions, methodology to estimate LC) [30].

Table 7.

Main post-operative and intact-lesion FSRT series (2015–2025): local control and radionecrosis

Study	Design	Dose / Fx	N (patient / BM)	1-year Local control	Overall RN	Symptomatic RN
Navarria et al. [33] (2016)	Single BM > 2 cm; FSRT. Retrospective. Single-center.	27 Gy/3 (2.1–3.0 cm); 32 Gy/4 (3.1–5.0 cm)	102 / 102	96%	5.8%	NR
Minniti et al. [28] (2016)	BM > 2 cm; FSRT vs. SRS. Retrospective. Single-center.	27 Gy/3 and SRS: 14–20 Gy	289 / 343	FSRT : 91% vs. SRS : 77%	FSRT 8% vs. SRS 20% (p = 0.004)	NR
Lesueur et al. [34] (2018)	Melanoma/RCC; BM < 3 cm; FSRT vs. SRS; CK. Retrospective. Single-center.	30 Gy/3; 36 Gy/6 vs. SRS: 20–22 Gy/1	60 /193	FSRT: 72% vs. SRS: 79%	FSRT 7.1% vs. SRS 9.6% (p = 0.55)	NR
Martinage et al. [35] (2019)	1 to 3 BM; post-op cavities; FSRT. Retrospective. Two centers.	24 Gy/3–30 Gy/5	160 / 167	88% (95% CI 81–93)	8.9%, and 57% of reirradiated cavities	2 cases of grade 3
Navarria et al. [36] (2020)	adjuvant FSRT, 1 BM. Prospective, phase II. Single-center.	30 Gy/3	101 / 101	98.9%	NR	26 cases of grade 2–3
Putz et al. [37] (2020)	Intact BM; FSRT vs. SRS. Restrospective. Single-center.	2–3 fx (total dose variable) and SRS: 16–20 Gy	120/ 190	FSRT: 68.6% vs. SRS: 65.4%, (HR 0.47, p = 0.015)	FSRT: 0% vs. SRS: 9.6%, (HR 0.18, p = 0.045)	FSRT: 0% vs. SRS: 3.9% of grade 4, p = 0.041
Eitz et al. [38] (2020)	Post-op cavities; FSRT. Restrospective. Multicentre international.	Mostly 30 Gy/5	558 / 581	84%	8.6%	NR
Faruqi et al. [39] (2020)	Post-op cavities and intact BM; LINAC. Retrospective. Single-center.	20–35 Gy / 5	187/ 118 (with 132 intact BM)	80–90%	21.2%	10.8%
Mengue et al. [32] (2020)	Intact and post-op BM. CK. Retrospective. Single-center.	27 Gy/3; 30 Gy/5; 35 Gy/5	389 / 400	Overall: 76.5% (better if < 2.5 cm); For post-op arm: 87.0%	5%	5%
Loo et al. [40]  (2020)	FSRT vs. SRS 1–3 BM. Critical review.	23.1 Gy / 3 and SRS: 14 Gy	179 / 287	FSRT: 78.4% vs. SRS: 88.1%	SRS: 0.5% vs. FSRT: 5%, associated with GTV	NR
Telentschak et al. [41](2021)	FSRT, reirradiation, large BM, deep BM; CK. Retrospective Single-center.	27 Gy/3	34/ 57 (including 27 reirradiation)	78.6%	6%	1 case of grade 3
Andruska et al. [42] (2021)	Intact and post-op BM. LINAC. Retrospective. Multi-institutional.	25–40 Gy/5 (median 30 Gy)	83 / 117	NR	NR	15%
Yan et al. [43] (2022)	Intact BM (no prior cranial RT). Retrospective. Single-center.	18–32.5 Gy / 3–5	60 / 133	82.2%	7.1% RN at 1 year (13.2% if BM > 1 cm; 3.2% if BM ≤ 1 cm)	NR
Myrehaug et al. [21] (2022)	Intact BM; LINAC. Retrospective. Single-center.	30–35 Gy/ 5	220 / 334	85%	15–16% ARE	9.5%
Kawai et al. [44] (2023)	BM ≤ 2 cm; Retrospective. Single-center.	30 Gy/5	49 / 179	93%	One patient	0
Gruber et al. [45] (2023)	Intact BM. Retrospective. Single-center.	30 Gy/6	73 / 103	68.7%	4.8%	NR
Barillaro et al. [46] (2023)	Intact BM. Retrospective. Single-center.	18–36 Gy / 3–5 and SRS: 15–21 Gy	46 / 127	87.2%	30%	20% of grade 2
Hahnemann et al. [47] (2024)	Post-op cavities; CK. Retrospective. Single-center.	35 Gy/7	105 / NR	98–99%	20% of RICE (including RN)	5–7%
Pan et al. [48] (2024)	Intact and post-op BM; FSRT. Retrospective. Single-center.	25–40 Gy / 5	40 / 70	84%	33% ARE/RICE global	RN
ESTRON El Shafie et al. [49] (2025)	FSRT post-cavities vs. WBRT after BM surgery. Prospective, phase II randomized. Single-center.	SRT: 30–35 Gy/5–7 vs. WBRT: 30 Gy /10	54 (27 per arm) /	FSRT: 100% vs. WBRT: 91.5%	FSRT arm: 4 cases	FSRT: 2 cases of grade 2; No grade ≥ 3

RN: radionecrosis; SRT: stereotactic radiotherapy; GTV: gross tumor volume; Gy: gray; CK: Cyberknife; BM: brain metastasis; RT: radiotherapy; Fx: fraction; FSRT: fractionated SRT; SRS: single-fraction radiosurgery; WBRT: whole brain radiotherapy; ARE: adverse radiation effects; RICE: radiation-induced contrast enhancements; RCC: renal cell carcinoma; NR: not reported. VS: versus

For patients with multiple lesions, the balance between SRT and WBRT remains patient-specific: randomized trials show better distant brain control with WBRT plus SRS versus SRS alone (1-year brain tumor recurrence rate was 46.8% and 76.4% respectively (P<0.001)) [6]. But at the cost of worse neurocognition and no survival gain, supporting focal strategies in carefully selected patients. In BM > 2.5–3 cm, surgery may improve LC and reduce neurologic death compared with radiation alone; in the post-operative setting, SRT to the cavity is favored over WBRT for neurocognitive preservation, but with a higher risk of LMD recurrences [4, 7, 50] In published data, LMD rates after FSRT and SRS ranged from 13 to 19% and 7–31% respectively. In our cohort, despite postoperative FSRT, LMD remained substantial in the surgical subgroup (31.6%, 6 patients on 19 vs. 5.4% without surgery), suggesting meticulous cavity contouring, dose selection, and the potential role of neoadjuvant SRS scheduling; ongoing trials should refine patient selection, dosing, and peri-operative sequencing [23, 24].

Radionecrosis

For larger BM, retrospective comparisons suggest FSRT may lower RN risk relative to single-fraction SRS, though prospective comparative data are still absent [51]. RN frequencies vary widely (4.5–30%), reflecting differences in diagnostic criteria (radiologic versus metabolic versus symptomatic versus histologic) and MRI follow-up [26, 52]. In Barillaro et al. study including 127 BM, rate of RN was 30% with 13.7% of grade 2. Minniti et al. and Mengue et al. reported respectively 8% and 5% of RN, but they did not report asymptomatic cases [28, 46]. In our cohort, any-grade RN reached 32.1%, while symptomatic RN was 14.5%. The median time to RN was 12 months (range 6–24 months), consistently with prior series [27, 28, 42, 46]. Larger BM and prior surgery increased RN risk. Volume and dose-volume exposure is the primary modifiable driver of RN. Treating multiple spatially close targets may increase the overlap of intermediate-to-high dose regions and thereby contribute to RN risk. In our descriptive analysis of symptomatic RN cases, several grade ≥ 2 events occurred in the setting of short inter-PTV distances (< 2 cm), suggesting that spatial clustering of lesions is a clinically relevant factor in plan optimization. When targets are in proximity, intermediate-to-high dose regions overlap, leading to a disproportionate increase in the volume of normal brain exposed to biologically relevant dose levels, even when individual target constraints are respected. This phenomenon has been reported in prior multi-lesion SRT and FSRT series, where composite normal-brain dose-volume metrics, such as V12Gy in SRS or V20 and V30Gy in FSRT regimens, were more predictive of RN than per-lesion parameters. In the HyTEC analyses and in contemporary FSRT cohorts, these composite metrics were shown to capture the cumulative effect of dose overlap from adjacent targets, particularly in patients treated for multiple or closely spaced metastases [21, 30]. Prior WBRT was not significantly associated with RN (Fisher’s exact test, p = 0.101); however, the small number of WBRT-exposed patients precluded multivariable analysis. From a dosimetric standpoint, the use of brain minus PTV dose–volume metrics may underestimate exposure of the peritumoral brain, which is the region most susceptible to RN. By excluding the target volume from normal brain DVH calculations, steep dose gradients and perilesional hotspots may not be fully captured by standard constraints, particularly in the presence of multiple or closely spaced lesions. For BM ≥ 2–3 cm, retrospective comparisons generally show comparable or improved LC and lower RN with FSRT regimens relative to SRS, supporting the emphasis on volume-adapted fractionation, although prospective head-to-head evidence remains limited [40, 46]. Those remarks remain a consideration as indications for SRT extend to patients with ≥ 10 BM [53].

Signals linking histology, oncogenic drivers, or concurrent systemic therapy to RN are inconsistent in other published series. Some authors have reported as risk factors: lung adenocarcinoma, renal histology and ALK mutation lesions whereas others do not, [21, 54, 55]. Similarly, we found no association with systemic therapies (including immunotherapy and TKIs) on univariate analysis [56, 57]. However, given the limited number of RN events and the absence of multivariable adjustment for treatment class, these results should be interpreted cautiously. Moreover, immune-related imaging changes can mimic RN; making structured frameworks (iRANO) and advanced MRI (perfusion, spectroscopy) valuable for adjudication [58, 59]. The dose–volume dependency of RN further argues for careful constraint reporting [29].

Neurological death

Brain control remains a concern as we observed a 37.4% rate of neurological death, high but consistent ranges for SRT cohorts (20–50%). Approximately 90% of patients lived longer than the DS-GPA median estimate for their stratum; however, these analyses should be interpreted in the context of a retrospective, single-center cohort and are intended to provide descriptive real-world benchmarking rather than formal model validation. It supports multidisciplinary, contextual decision-making with DS-GPA as a contributory tool rather than a rigid exclusion cut-off. Patients with very limited life expectancy should be referred to best supportive care rather than local therapy; better selection tools remain a priority.

Strengths, limitations, and implications

This single-center experience reflects routine practice with consistent planning, standardized fractionated CyberKnife^® protocols, and systematic follow up over eight years, enabling characterization of failure patterns and clinically meaningful endpoints. Several limitations should be acknowledged. First, the retrospective design exposes the study to selection and information biases, and some variables were unavailable in a proportion of patients, including molecular biomarkers. Second, RN versus tumor recurrence was primarily determined using clinical and imaging criteria in real world conditions. Although this reflects routine practice, histological confirmation and advanced imaging were not uniformly available, which may have led to misclassification in a minority of cases. Third, neurocognitive outcomes, quality of life measures, and patient reported outcomes were not collected, limiting the functional interpretation of toxicity beyond CTCAE grading. Fourth, while dosimetric parameters were reviewed on a case-by-case basis in patients who developed RN, with attention to dose volume objectives, reirradiation contexts, and lesions proximity, this qualitative assessment does not replace a comprehensive cohort wide dosimetric analysis. As a result, the present study cannot formally identify independent dose volume predictors of RN or support dosimetric modeling. In addition, normal brain dose volume metrics were derived from a brain minus PTV structure as part of institutional practice, which may limit direct comparison with published constraints that frequently use whole brain volumes including the target. Moreover, we did not perform shell-based dosimetric analyses (the region between the GTV and the PTV or fixed thickness rings around the target), which may reflect the biologically relevant peritumoral brain at risk of RN. Despite these limitations, our findings support volume adapted fractionation and strict dose volume constraints to balance local control and RN risk, and they reinforce the importance of optimizing peri operative SRT workflows.

Future directions

Prospective studies should clarify dose–fractionation for large, irregular, or closely spaced BM and standardized LC and RN definitions. Peri-operative strategies including neoadjuvant SRT against post-operative SRT need randomized evaluation focusing on LC, LMD, RN and neurocognition. As focal approaches extend to patients with ≥ 10 BM, trials and registries should stratify by total intracranial tumor volume and optimize planning across all targets. Finally, AI-driven radiomics and dosiomics should deliver calibrated tools for individualized assessment of LC, RN risk, and early classification distinguishing RN from relapse [60, 61]. Consensus indicates that clinicians increasingly defer brain RT or surgery when patients are receiving intracranial-active systemic therapy, particularly in NSCLC, breast cancer, and melanoma [9].

Conclusion

FSRT achieved durable LC with acceptable rates of symptomatic RN in this cohort. Tumor volume and prior surgery were associated with both LC and RN, supporting individualized dose and fractionation strategies based on lesion size and clinical context. Further prospective data are needed to refine dose constraints and integrate evolving therapies.

Abbreviations

18F-FDG PET/CT

Fluorodeoxyglucose positron emission tomography/computed tomography

ADC

Antibody–drug conjugate

ALK

Anaplastic lymphoma kinase

bPFS

Brain progression-free survival

BED

Biologically effective dose

BC

Breast cancer

BM

Brain metastasis

CT-AP/TAP

Chest–abdomen–pelvis

CE-T1w

Contrast-enhanced T1-weighted (MRI sequence)

CI

Confidence interval

CK

CyberKnife^®

CTV

Clinical target volume

CTCAE v5.0

Common Terminology Criteria for Adverse Events, version 5.0

DWI

Diffusion-weighted imaging

EGFR

Epidermal growth factor receptor

EORTC

European Organisation for Research and Treatment of Cancer

EANO

European Association of Neuro-Oncology

ESMO

European Society for Medical Oncology

ESTRO

European Society for Radiotherapy and Oncology

FLAIR

Fluid-attenuated inversion recovery (MRI sequence)

FSRT

Fractionated stereotactic radiotherapy

Fx

Fraction

GPA

Graded Prognostic Assessment (histology-specific)

GI

Gastrointestinal

GTV

Gross tumor volume

HER2

Human epidermal growth factor receptor 2

HR

Hazard ratio

HSRT

Hypofractionated stereotactic radiotherapy

HyTEC

Hypofractionated Treatment Effects in the Clinic

IC

Immune cells (for PD-L1 immune-cell scoring; use only if reported)

iRANO

Immunotherapy Response Assessment in Neuro-Oncology

IQR

Interquartile range

KPS

Karnofsky Performance Status

LC

Local control

LFFS

Local failure-free survival

LMD

Leptomeningeal disease

MET

MET proto-oncogene (hepatocyte growth factor receptor)

MRI

Magnetic resonance imaging

NSCLC

Non-small-cell lung cancer

OAR

Organ at risk

OS

Overall survival

PD-L1

Programmed death-ligand 1

PET

Positron emission tomography

PS

Performance status

PTV

Planning target volume

RANO-BM

Response Assessment in Neuro-Oncology for Brain Metastases

RCC

Renal cell carcinoma

RN

Radionecrosis

ROS1

c-ros oncogene 1 (receptor tyrosine kinase)

RT

Radiotherapy

SCLC

Small-cell lung cancer

SD

Standard deviation

SRS

Stereotactic radiosurgery (single-fraction)

SRT

Stereotactic radiotherapy

TKI

Tyrosine-kinase inhibitor

TPS / CPS

Tumor proportion score / Combined positive score (PD-L1)

V12 / V20 / V24 / V28.8

Normal-brain volume receiving ≥ 12/20/24/28.8 Gy

WBRT

Whole-brain radiotherapy

Author contributions

Conceptualization: Catherine Durdux, Aurélia Alati. Investigation: Aurélia Alati. Ressources: Catherine Durdux, Philippe Giraud, Aurélia Alati, Jean-Emmanuel Bibault, Sarah Kreps, Emmanuelle Fabiano. Supervision: Catherine Durdux. Writing – original draft: Aurélia Alati. Writing – review & editing: Aurélia Alati, Catherine Durdux. Funding acquisition: None. Statement: All authors read and approved the final manuscript.

Data availability

De-identified data underlying this article are available from the corresponding author upon reasonable request. The data are not publicly shareable due to patient privacy.

Declarations

Competing interests

The authors declare no competing interests.