Radiation therapy for melanoma brain metastases: a systematic review

Search results and exclusions are shown in Figure 1. There were 142 publications between 1980–2020, 112 of which were unique studies or were the primary publication of a series of publications, and 30 were duplicates or non-primary publications (Table 1.) Seven studies were published only as abstracts. 57.1% (64/112) of the publications were from the USA, 30.4% from Europe, and 11.6% from other countries. Sample sizes ranged from 7–1304 patients (median 77). While our focus was on RT, most articles (96/112) included patients who had received a variety of other therapies for their brain metastases. Surgery was reported in 79 studies, WBRT in 95 studies, SRS in 84 studies and 64 reports included subsets of patients who received some form of systemic therapy as well as RT. Outcomes for the subsets of patients treated with both RT and any form of contemporary systemic therapy were not analysed. Amongst studies reporting the use of SRS, five reported using both linear accelerator and Gamma Knife methods, 32 used Gamma Knife only, 18 used linear accelerator only and 29 did not report which was used.

Figure 1

Only three studies were prospectively-conducted RCTs; the remainder were retrospective studies of patient cohorts of specific treatment/s (n = 85) or cohorts of patients treated for melanoma brain metastases (n = 24). A comparison of outcomes for different treatment regimens was reported in 97 studies, while 15 studies were non-comparative, reporting outcome data for a single treatment group.

Details of the risk of bias assessment for each study are provided in Supplementary Table 2 and summarised in Figure 2. Of the three RCTs, none reported how their randomisation sequence was generated but all reported complete outcome data and clinically-relevant outcomes. Two of the three trials28, 29 were closed early due to poor accrual and had sample sizes of 7 and 18, greatly limiting the reliability of their results. Baseline characteristics for the different treatment arms were reported and similar for the trial of 18 patients29 but were not reported for the trial of 7 patients.28 The largest trial30 had 215 patients, and while not stratifying for previous treatments, randomisation was effective as previous treatments were well balanced between the two study arms, as were baseline characteristics. This trial provided the most reliable data for identifying the effects of adjuvant WBRT in conjunction with surgery and/or SRS for patients with 1–3 brain metastases.

Figure 2

For the 109 cohort studies, selection bias was a significant concern, as 77 studies (71%) did not provide information specifying how a treatment choice was made. Sixteen studies reported that there were no significant differences in patient characteristics such as age, number of brain metastases and tumour volume between treatment groups, while 11 studies reported significant differences between patients treated using different modalities. The remaining 82 studies did not report similarities or differences in patient characteristics, although in 21 studies the patient characteristics were provided. Fifty-six studies did not report if or how the diagnosis of melanoma brain metastases was verified; this may have resulted in misclassification of disease, although this is probably a relatively inconsequential source of bias.

Many studies reported analyses of one treatment without considering prior and subsequent forms of treatment (e.g. SRS preceded by surgical resection). Our analysis was based on grouping data based on all treatments received for brain metastases.

Five studies (651 patients) included only asymptomatic patients, nine studies (532 patients) included only symptomatic patients, 30 studies (5906 patients) had a mixture of asymptomatic and symptomatic patients and 57 studies (5452 patients) did not report this detail.

Seven of the 95 studies provided specific criteria for choosing WBRT in their patients; four stated that it was used for multiple brain metastases31, 32, 33, 34, three reported its use for progression of brain metastases31, 35, 36, one its use for single, large metastases36 and one its use for symptomatic metastases.37

Nineteen studies reported criteria for choosing SRS. Twelve stated that it was used for small metastases, often <30mm in diameter, nine required good performance status as measured by Karnofsky performance score (KPS), with four of these using a cut-off of KPS ≥ 70. Seven studies used SRS for a small number of brain metastases (usually 1–3). Five studies used SRS when metastases were inaccessible for surgery, four used it in for multiple metastases, but only one specified a number (≤ 9), and three stated that it was used for asymptomatic lesions. Other infrequently used criteria were; expected survival > 3 months, non-life threatening lesions, high risk for surgery, including proximity to the brain stem or optic nerve. A single study32 reported criteria for using a combination of SRS and WBRT, stating that this was used for ≥ 5 lesions.

Fourteen studies provided criteria for surgery; a single metastasis (5 studies), few or <3 metastases (2 studies), accessible metastases (8 studies), symptomatic metastases (4 studies), stable extracranial disease (4 studies), good KPS (1 study), life expectancy > 3 months (1 study), and 2–3 brain metastases if one was life-threatening (2 studies).

Many treatment groupings that included RT were reported but outcomes were not reported for all groups. Given the limited amount of data for clearly-defined treatment groups, we re-grouped data into two additional treatment options; (i) patients treated with WBRT and any of SRS, surgery or non-contemporary systemic therapy, and (ii) patients treated with SRS and any of WBRT, surgery, or non-contemporary systemic therapy.

Patient characteristics within different treatment groups are summarised in Table 2. For all treatment types, there was a predominance of males. Patients treated with WBRT alone were somewhat younger than those receiving SRS and patients undergoing WBRT were less likely to have a single brain metastasis. While the data were sparse, there was considerable overlap in patient characteristics across different treatment modalities, indicating that the choice of treatment was not consistently determined by age, presence of symptoms, number of metastases or control of primary disease.

Ninety-six studies reported median survival for all patients or subsets of patients and there were 49 different treatment groupings.

Within-study comparisons were possible for six treatment groupings (Table 3.). Eleven studies reported median survival for patients treated with WBRT alone or with WBRT and surgery. The median survival in the WBRT alone group was 4.0 months (IQR 3.0–4.0 months), significantly less than for those treated with surgery and WBRT (11.0 months, IQR 8.8–11.8 months) (p = 0.002). Expressing these findings as a median difference between treatments, patients who had WBRT and surgery had a 5.4 month (IQR 4.6–8.0 months) longer survival compared with those treated with WBRT alone. In this group of 11 studies, three reported using surgery in patients with a single brain metastasis37, 38, 39, and two studies reported features for treatment with WBRT, this being good performance status alone in one study40 and multiple lesions, good performance and symptoms in the second study.37 Significant differences in median survival were also apparent between WBRT alone and surgery alone (6.0 months longer for surgery) and median survival for patients treated with WBRT plus SRS was 3.4 months longer than with WBRT alone. In the five studies with groups treated with WBRT alone or surgery alone, three reported that surgery was used for a single or few brain metastases34, 37, 38 and WBRT was used for patients with more than one brain metastasis (1 study)38 and for patients with good performance status (1 study37). None of the four studies reporting patients treated with WBRT alone or WBRT+SRS described features leading to these treatment choices. There were no significant differences in median survival between WBRT alone and SRS alone, or between WBRT alone and WBRT with chemotherapy, or SRS alone compared to WBRT with SRS.

Summarised findings for median survival in all studies are detailed in Table 3 and for other groupings in Supplementary Table 4. The group treated with WBRT alone had the shortest survival; 3.5 months (IQR 2.4–4.0 months). For the group treated with surgery and WBRT, the median survival was 11.0 months (IQR 7.8–12.0 months). Adding chemotherapy to WBRT appeared to provide little benefit, with six studies of 137 patients reporting a median survival of 4.3 months (IQR 2.8–6.0 months) (Supplementary Table 4). The compiled grouping of WBRT with any of SRS, surgery, or non-contemporary systemic therapy had a median survival of 7.2 months (IQR 4.6–9.4 months) across 47 studies with 2230 patients.

Median survival after SRS treatment alone was reported in eight studies, giving a median survival of 7.5 months (IQR 6.7–9.0 months). The median survival for patients treated with Gamma Knife SRS (5 studies, 208 patients) was 7.0 months (IQR 5.6–7.8 months) and for the three studies (980 patients) that did not report which SRS technology was used the median survival was 8.8 months. Eighteen studies reported using linear accelerator SRS but none reported the median survival for patients treated with SRS alone. Adding non-contemporary systemic therapy to SRS treatment did not improve the median survival (7.9 months, IQR 6.1–9.9 months) but the addition of surgery was associated with an increase in median survival of around 5 months (13.0 months, IQR 9.4, 13.5 months). Compiled grouping of SRS with any of WBRT, surgery, or non-contemporary systemic therapy gave a median survival of 8.0 months (IQR 6.2–10.9 months) over 42 studies involving 2702 patients.

In the only completed RCT in patients with melanoma brain metastases30, median survival in the group treated with adjuvant WBRT after definitive local treatment of 1–3 metastases was 16.5 months (95% CI 13–24 months) compared to 13.0 months (95% CI 10–19 months) for those that did not receive adjuvant WBRT (p=0.86).

No studies reported median survival within treatment groups separately for asymptomatic and symptomatic patients. For the 5 studies that included only asymptomatic patients, there were no common treatment groups. In 4 studies of 89 symptomatic patients, the median survival for the treatment group WBRT with surgery was 9.2 months (IQR 5.4–12.8 months) and for WBRT with any other treatment; 5 months (IQR 2.5–10.0 months, 7 studies of 245 patients).

Median survival over time was explored by grouping the data into three time periods based on the first year of patient recruitment within each study (Table 4). Eighty-one studies reported median survival for their whole cohort irrespective of treatment, and these showed increasing survival over the years. There were significant differences in median survival between the pre-1989 group compared with the 1990–2002 (p = 0.017) and 2003–2015 groups (p = 0.002) and also between the groups first treated in 1990–2002 compared with 2003–2015 (p = 0.021).

Median survival within treatment groups over the three time periods showed a trend toward slightly increased survival in more recent years, but none of the differences was statistically significant.

Fifty-six studies reported 1-year survival for all patients or subsets of patients (Table 3). Pooled data from 7 studies of 189 patients treated with WBRT alone gave a 1-year survival rate of 9.0% (IQR 0.0–22.5%) while for SRS alone the 1-year survival was 35.5% (IQR 20.8–47.8%, p = 0.041) in six studies of 330 patients. In the compiled grouping of WBRT with or without any other therapy, the 1-year survival was 21.4% (IQR 13.6%–37.0%). For the SRS grouping with or without any other therapy (WBRT, surgery, non-contemporary systemic therapy), the 1-year survival was 31.0% (IQR 25.0–39.0%) across 35 studies of 2644 patients. In the only completed RCT30, 1-year survival in the group treated with adjuvant WBRT was 58.4% (95%CI 49.6%–68.9%) compared with 54% (95%CI 45.3%–64.3%, p = 0.89) for those treated without WBRT.

Most studies defined local control as a reduction in metastasis size or stability of metastasis size, as determined by follow-up imaging. Fifty-three studies reported local control data, 21 without a defined time frame and for almost all it was reported for the total patient group, not separately for different treatment groupings. The 1-year local control rate was highest for those treated with SRS; 76% (IQR 62.8%–88.5%). The 1-year local control rate after WBRT was 5.5% (IQR 0.0%–12.0%, 1 study, 74 patients). For the 550 patients in 11 studies that treated patients with any combination of WBRT, SRS, surgery and non-contemporary systemic therapy, the 1-year local control rate was 68.0% (IQR 66.072.0%). There was no difference in the 1-year local control rate between Gamma Knife SRS and linear accelerator-based SRS (69% vs. 72.0%).

Thirty-six studies reported rates of new brain lesions developing during the follow-up period. For the 23 studies that reported new brain lesions at 6-months the median rate was 44% (IQR 32.0%– 53.0%) and at 12-months 67% (IQR 62.3%–71.5%, 14 studies). Three studies (189 patients) reported a 6-month new brain lesion rate in patients treated with WBRT and other treatment, giving a median rate of 39% (IQR 34.0%–44.5%) and for SRS and other treatment a rate of 47% (IQR 34.5%–55.5%, 11 studies, 1306 patients). At 12 months, the RCT of patients with 1–3 brain metastases reported a new brain lesion rate of 42% in the adjuvant WBRT group and 50.5% in those who did not receive adjuvant WBRT (p = 0.22). For patients treated with SRS, the proportion who developed a new brain lesion by 12 months was 67% (IQR 57.0%–75.0%; 11 studies, 1278 patients).

Neurologic death was reported in 40 studies but only 11 reported this for a defined treatment group. The definition of neurologic death was variable. Only one study provided a definition that combined an objective measurement with radiological and clinical neurologic changes.41 Other studies used brain lesion progression and/or recurrence (18 studies), brain hemorrhage alone (4 studies), neurologic dysfunction alone (4 studies) or other features (3 studies) as criteria for designating a death as neurologic. The reported proportion of patients with a neurologic cause of death ranged from 0% to 90%. Three studies reported the proportion of patients treated with WBRT with or without surgery who experienced neurologic death (0%, 24%, 83%). Two studies reported neurologic deaths for patients treated with WBRT alone (14%, 88%), two studies reported on SRS alone (41%, 90%), two studies reported neurologic deaths in those treated with SRS and WBRT (50%, 58%) and two studies reported neurologic death in patients treated with WBRT with or without systemic therapy (57%, 75%). Re-grouping the data into SRS with any other treatment (6 studies), gave a median neurologic death rate of 53% (IQR 44.8%–69.4%), while for WBRT and any other treatment (10 studies), the median was 50% (IQR 19.3%–62.3%).

Seventy-six studies assessed whether the number of brain metastases present at diagnosis impacted survival, with 58 reporting significantly improved survival for patients with single lesions while 18 reported no impact. In the RCT30, the number of brain metastases (1 vs. 2–3) did not influence overall survival. Only three studies reported these data within specific treatment groups.42, 43, 44 Two studies43, 44 reported survival in patients treated with WBRT alone comparing those with one metastasis to those with ≥ 2 brain metastases; one study44 reported better survival in the single metastasis group (16 weeks) compared to the multiple metastases group (12 weeks) while the other study43 did not (9 weeks for a single metastases and 11 weeks for multiple metastases).

Adverse effects of RT were reported in 41 studies, but only 17 reported events within treatment groups and these were primarily studies that included systemic therapy. Radiation necrosis (with various radiological and/or pathological definitions) was reported in 13 studies, 11 of which focussed on SRS. The median rate was 8.1% (IQR 3.4%–22.2%). In the four studies using Gamma Knife SRS, the median radiation necrosis rate was 3.4% (IQR 0.47%–5.49%) and for the 4 studies using linear accelerator SRS it was 22.2% (IQR 15.59%–25.66%). Two studies reported this for WBRT, with rates of 1.9% and 3.6%. Eight studies reported intracranial haemorrhage in their patients, with seven studies focussed on SRS, giving a median rate of 14.7% (IQR 0.94–18.8%). Three studies using Gamma Knife SRS reported brain haemorrhage rates with a median rate of 18.8% (IQR 9.86–24.01%) and two studies used linear accelerator SRS, with brain haemorrhage rates of 15% and 16%. Other reported adverse effects included headaches, seizures, skin reactions, fatigue, nausea, alopecia and confusion but because data were sparse and pooled analysis was not possible.

Risk of bias assessment for these studies showed that many of the non-randomised studies included patients who were treated without explanation of how treatment choices were made. In the 30% of studies that did report treatment selection criteria there was considerable variation, reflecting the diversity of clinical practice between and even within individual centres and over the 40-year study period. This selection bias limited our ability to apply results to specific patient groups as we could not be sure in many instances which types of patients received particular treatments. Also, important prognostic factors such as performance status and extent of extracranial disease were rarely reported within treatment groups. Compiling the rather limited patient characteristics data for the different treatment groups showed that there was considerable overlap in the types of patients receiving WBRT and SRS. There was a degree of consistency in offering surgery to patients with a single or few brain metastases, as almost half of the studies that reported criteria for surgery stated this. However, it was not possible to determine survival outcomes for patients who underwent surgery for a single brain metastasis followed by RT as this was not reported. Most studies that analysed the effect on survival of having a single versus multiple brain metastases, irrespective of other treatments, reported improved survival with a single metastasis. This suggests that patients who undergo surgery have a greater likelihood of increased survival at baseline. A valid comparison of different RT modalities should consider or control for factors that have a major impact on survival, an issue not possible to evaluate using the current evidence.

Further difficulties arose in relation to the multitude of different outcomes reported that could not be easily combined. For example, median, 6-month, 1 and 2-year survival rates were often reported but recurrence/regrowth at a treated site versus new lesions at new sites were often not clearly specified within treatment groups or time frames. Similarly, the definitions of neurologic death varied between studies. Only one study provided a robust, measurable definition of this while others relied on less precise features. Definitions of radiation necrosis were also variable, provided in only eight studies, each of which was different; three relied solely on various imaging features, one solely on clinical signs of bleeding and four on combined imaging features and clinical signs. Radiation necrosis and neurologic death are important endpoints being measured in current clinical trials and an assurance of similar definitions and measurements will greatly aid interpretation of these outcomes across studies. A possible solution to the diverse and variably-defined outcomes in studies would be for clinicians, researchers and patients to agree on a minimum required and consistently-defined outcome reporting set, as has been done for other diseases such as rheumatoid arthritis, ulcerative colitis, and lung cancer.45, 46, 47 Researchers have developed a process for selecting outcomes of interest to clinicians and patients, and deciding how these can be implemented in their respective settings.48, 49 A similar strategy for future studies of patients with melanoma brain metastases would be feasible.

Few studies reported whether the treatments resulted in relief of symptoms for symptomatic patients. Australian guidelines suggest that WBRT may be considered in a palliative setting for relief of symptoms, and there are many anecdotal reports of its value in this situation, but we found little reported evidence to support the effectiveness of this option.

Use of treatment groupings was a substantial limitation to interpretation, as many studies grouped together patients who received different treatment combinations. Ideally more uniform treatment groups should be used but this would require studies of much greater size to achieve adequate numbers within each group.