Scolaris Content Display Scolaris Content Display

Cochrane Database of Systematic Reviews Review - Intervention

Treatment options for progression or recurrence of glioblastoma: a network meta‐analysis

Authors' declarations of interest

Version published: 05 January 2021 Version history

https://doi.org/10.1002/14651858.CD013579.pub2
Collapse all Expand all

Background

Glioblastoma (GBM) is a highly malignant brain tumour that almost inevitably progresses or recurs after first line standard of care. There is no consensus regarding the best treatment/s to offer people upon disease progression or recurrence. For the purposes of this review, progression and recurrence are considered as one entity.

Objectives

To evaluate the effectiveness of further treatment/s for first and subsequent progression or recurrence of glioblastoma (GBM) among people who have received the standard of care (Stupp protocol) for primary treatment of the disease; and to prepare a brief economic commentary on the available evidence.

Search methods

We searched MEDLINE and Embase electronic databases from 2005 to December 2019 and the Cochrane Central Register of Controlled Trials (CENTRAL, in the Cochrane Library; Issue 12, 2019). Economic searches included the National Health Service Economic Evaluation Database (NHS EED) up to 2015 (database closure) and MEDLINE and Embase from 2015 to December 2019.

Selection criteria

Randomised controlled trials (RCTs) and comparative non‐randomised studies (NRSs) evaluating effectiveness of treatments for progressive/recurrent GBM. Eligible studies included people with progressive or recurrent GBM who had received first line radiotherapy with concomitant and adjuvant temozolomide (TMZ).

Data collection and analysis

Two review authors independently selected studies and extracted data to a pre‐designed data extraction form. We conducted network meta‐analyses (NMA) and ranked treatments according to effectiveness for each outcome using the random‐effects model and Stata software (version 15). We rated the certainty of evidence using the GRADE approach.

Main results

We included 42 studies: these comprised 34 randomised controlled trials (RCTs) and 8 non‐randomised studies (NRSs) involving 5236 participants. We judged most RCTs to be at a low risk of bias and NRSs at high risk of bias. Interventions included chemotherapy, re‐operation, re‐irradiation and novel therapies either used alone or in combination. For first recurrence, we included 11 interventions in the network meta‐analysis (NMA) for overall survival (OS), and eight in the NMA for progression‐free survival (PFS). Lomustine (LOM; also known as CCNU) was the most common comparator and was used as the reference treatment. No studies in the NMA evaluated surgery, re‐irradiation, PCV (procarbazine, lomustine, vincristine), TMZ re‐challenge or best supportive care. We could not perform NMA for second or later recurrence due to insufficient data. Quality‐of‐life data were sparse.

First recurrence (NMA findings)

Median OS across included studies in the NMA ranged from 5.5 to 12.6 months and median progression‐free survival (PFS) ranged from 1.5 months to 4.2 months. We found no high‐certainty evidence that any treatments tested were better than lomustine. These treatments included the following.

Bevacizumab plus lomustine: Evidence suggested probably little or no difference in OS between bevacizumab (BEV) combined with lomustine (LOM) and LOM monotherapy (hazard ratio (HR) 0.91, 0.75 to 1.10; moderate‐certainty evidence), although BEV + LOM may improve PFS (HR 0.57, 95% confidence interval (CI) 0.44 to 0.74; low‐certainty evidence).

Bevacizumab monotherapy: Low‐certainty evidence suggested there may be little or no difference in OS (HR 1.22, 95% CI 0.84 to 1.76) and PFS (HR 0.90, 95% CI 0.58 to 1.38; low‐certainty evidence) between BEV and LOM monotherapies; more evidence on BEV is needed.

Regorafenib (REG): REG may improve OS compared with LOM (HR 0.50, 95% CI 0.33 to 0.76; low‐certainty evidence). Evidence on PFS was very low certainty and more evidence on REG is needed.

Temozolomide (TMZ) plus Depatux‐M (ABT414): For OS, low‐certainty evidence suggested that TMZ plus ABT414 may be more effective than LOM (HR 0.66, 95% CI 0.47 to 0.92) and may be more effective than BEV (HR 0.54, 95% CI 0.33 to 0.89; low‐certainty evidence). This may be due to the TMZ component only and more evidence is needed.

Fotemustine (FOM): FOM and LOM may have similar effects on OS (HR 0.89, 95% CI 0.51 to 1.57, low‐certainty evidence).

Bevacizumab and irinotecan (IRI): Evidence on BEV + irinotecan (IRI) versus LOM for both OS and PFS is very uncertain and there is probably little or no difference between BEV + IRI versus BEV monotherapy (OS: HR 0.95, 95% CI 0.70 to 1.30; moderate‐certainty evidence).

When treatments were ranked for OS, FOM ranked first, BEV + LOM second, LOM third, BEV + IRI fourth, and BEV fifth. Ranking does not take into account the certainty of the evidence, which also suggests there may be little or no difference between FOM and LOM.

Other treatments

Three studies evaluated re‐operation versus no re‐operation, with or without re‐irradiation and chemotherapy, and these suggested possible survival advantages with re‐operation within the context of being able to select suitable candidates for re‐operation. A cannabinoid treatment in the early stages of evaluation, in combination with TMZ, merits further evaluation.

Second or later recurrence

Limited evidence from three heterogeneous studies suggested that radiotherapy with or without BEV may have a beneficial effect on survival but more evidence is needed. Evidence was insufficient to draw conclusions about the best radiotherapy dosage. Other evidence suggested that there may be little difference in survival with tumour‐treating fields compared with physician's best choice of treatment. We found no reliable evidence on best supportive care.

Severe adverse events (SAEs)

The BEV+LOM combination was associated with significantly greater risk of SAEs than LOM monotherapy (RR 2.51, 95% CI 1.72 to 3.66, high‐certainty evidence), and ranked joint worst with cediranib + LOM (RR 2.51, 95% CI 1.29 to 4.90; high‐certainty evidence). LOM ranked best and REG ranked second best. Adding novel treatments to BEV was generally associated with a higher risk of severe adverse events compared with BEV alone.

Authors' conclusions

For treatment of first recurrence of GBM, among people previously treated with surgery and standard chemoradiotherapy, the combination treatments evaluated did not improve overall survival compared with LOM monotherapy and were often associated with a higher risk of severe adverse events. Limited evidence suggested that re‐operation with or without re‐irradiation and chemotherapy may be suitable for selected candidates. Evidence on second recurrence is sparse. Re‐irradiation with or without bevacizumab may be of value in selected individuals, but more evidence is needed.

PICOs

Population
Intervention
Comparison
Outcome

The PICO model is widely used and taught in evidence-based health care as a strategy for formulating questions and search strategies and for characterizing clinical studies or meta-analyses. PICO stands for four different potential components of a clinical question: Patient, Population or Problem; Intervention; Comparison; Outcome.

See more on using PICO in the Cochrane Handbook.

Treatment options for people with recurrent and progressive glioblastoma

Why this is important

Glioblastoma multiforme (GBM) is a very aggressive type of brain tumour. Even after treatment involving surgery, radiotherapy and chemotherapy the tumour may continue growing (progress) and almost always comes back (recurs). In this review, we consider progression and recurrence as one entity. A number of different treatments have been tested, but there has been no agreement about the best treatment/s to offer when someone’s tumour progresses or recurs.

Objectives

To evaluate which are the most effective treatments for people with progressive or recurrent GBM who have already received surgery, radiotherapy and chemotherapy. We looked at effectiveness in terms of length of overall survival (OS), survival without disease progression (progression free survival, PFS), serious side effects, and whether treatments affected quality of life. We looked at treatments given when the disease first progressed or recurred (first recurrence) and when it came back again after that (second or subsequent recurrences).

Methods

We searched for relevant research studies comparing the effectiveness of different treatments for recurrent GBM. We used network meta‐analysis (NMA) to compare different treatments. NMA is a statistical method that allows different treatments to be looked at together to decide which is best. This method allows different treatments to be ranked according to their effectiveness, even if treatments have not been directly compared with each other in research studies.

Main results

We included 42 studies (34 randomised controlled trials (RCTs) and 8 non‐randomised studies) involving 5236 people. Interventions included chemotherapy, re‐operation, re‐irradiation and newly developed treatments either used alone or in combination. NMA could not be performed for second or later recurrence due to insufficient data.

Survival outcomes for people with a first recurrence

We found no good evidence that any of the treatments tested were better than lomustine (also known as CCNU). Adding bevacizumab to lomustine (BEV + LOM) did not improve overall survival compared with lomustine alone. Other chemotherapy and novel agents either did not work, or the evidence on them was uncertain. Unfortunately, we did not find any studies on several commonly used treatments, such as PCV (procarbazine, CCNU, vincristine) and TMZ re‐challenge, to include.

Limited evidence suggested that a second operation with or without other treatments may have survival advantages for some individuals with a first recurrence. A small study of a cannabinoid treatment suggests this merits further investigation.

Survival outcomes for people with a second or later recurrence

For second or later recurrence, insufficient evidence meant that we were not able to carry out statistical analysis. Findings suggested that radiotherapy with or without BEV may have some survival advantages but this evidence is uncertain. We found no reliable evidence on best supportive care.

Serious adverse events (SAEs) from treatment

Most treatments caused some serious side effects (SAEs). The BEV + LOM combination was associated with a significantly greater risk of SAEs than lomustine alone. In general, adding treatments to bevacizumab was associated with more SAEs compared with BEV alone.

Authors' conclusions

For treatment of first recurrence of GBM, lomustine appears the most effective chemotherapy treatment and other combination therapies tested had a higher risk of serious side effects. A second operation or radiotherapy, or both, may be of value in selected individuals. For second recurrence, radiotherapy with or without bevacizumab may have a role but more evidence is needed. Several commonly used treatments were not evaluated, such as PCV (lomustine plus procarbazine and vincristine) and temozolomide re‐challenge. More research is needed.

Authors' conclusions

Implications for practice

Based on the available evidence for people with a first recurrence of GBM previously treated with surgery and chemoradiotherapy (Stupp protocol), lomustine (LOM) monotherapy appears to be the best chemotherapy option. Findings suggested that adding bevacizumab to lomustine therapy probably does not improve overall survival and is associated with a significantly greater risk of severe adverse events. We found no evidence on TMZ re‐challenge or PCV, two commonly used salvage regimens, in this context. Re‐operation with or without re‐irradiation and chemotherapy may be suitable for selected candidates.

In the course of conducting the review, we had hoped to answer the JLA question on best treatments for second recurrence; however, although several studies included people with a second recurrence, data were usually not separately analysed and reported for this subgroup. We found little reliable evidence on re‐irradiation with or without bevacizumab and on best supportive care. The role of tumour‐treating fields is also uncertain.

While there is an acceptance of the need to develop and test new agents, combination treatments for recurrent GBM are likely to be associated with a higher risk of severe adverse events than monotherapy and people should be informed of this when deciding whether or not to participate in such clinical trials.

The cost‐effectiveness of different treatment options for recurrent GBM remains unclear as economic studies in this field are lacking.

Implications for research

Findings of this review have highlighted the following research questions in this field. Trial participant randomisation and reporting of findings should be stratified by first and second recurrence. Prognostic scoring to select participants to different trials may be worthwhile.

  • For patients with first or second recurrence, what are the effects of PCV versus lomustine, or lomustine or PCV versus TMZ re‐challenge?

  • For patients with first or second recurrence, what are the effects of radiotherapy or bevacizumab + radiotherapy compared with lomustine?

  • Does delaying BEV to chemorefractory disease extend survival and improve quality of life beyond best supportive care alone?

  • For patients with resectable tumours, what are the effects of re‐operation?

  • What are patients' and health care professionals' experiences and views on treatment of recurrent GBM at first and subsequent recurrences?

We identified several ongoing trials. These include a trial of bevacizumab + dose‐dense TMZ versus bevacizumab alone (JCOG1308C 2019); surgery followed by second‐line therapy versus second‐line therapy alone (NCT02394626 2015); and a phase 3 trial of regorafenib versus lomustine (NCT03970447 2019b). Another ongoing trial is evaluating the effects of early palliative care (STEP care) among patients with various cancers including glioblastoma (ACTRN12617000534381 2017).

Early studies of some oncolytic virus treatments and vaccines have shown promise (e.g. see Ji 2016; Lang 2018; Reardon 2020) and further clinical trials of these therapies are anticipated. Alternative interventions, such as cannabinoid treatments and dietary measures, also need further study. Studies should evaluate economic outcomes.

Summary of findings

Open in table viewer
Summary of findings 1. Summary of overall survival findings

Estimates of effects, certainty assessment and rankings of different treatment options compared with lomustine on overall survival in people with first recurrence of glioblastoma

Patient or population: people with first recurrence of glioblastoma

 Interventions: bevacizumab (BEV), BEV + lomustine (LOM), regorafenib (REG), fotemustine (FOM), ABT414 +  temozolomide (TMZ); BEV + irinotecan (IRI), BEV + onartuzumab (ONA), cediranib (CED), CED + LOM

Comparison: lomustine

Outcome: overall survival
 

All intervention options

Relative effect and 95% CI (network estimate) **

Certainty of the evidence (GRADE)

Ranking*

(9 RCTs; 1734 participants in total)*

LOM
(5 RCTs; 403 participants)
 

Reference comparator

Reference comparator

5.9

REG
(1 RCT; 59 participants)

HR 0.50 (0.33 to 0.76)

⊕⊕⊝⊝
low1

1.3

Depatux‐M (ABT414) + TMZ
(1 RCT; 88 participants)

HR 0.66 (0.47 to 0.92)

⊕⊕⊕⊝
moderate2

2.1

BEV + LOM
(3 RCTs, 401 participants)
 

HR 0.91 (0.75 to 1.10)

⊕⊕⊕⊝
moderate4
 

4.4

FOM
(1 RCT; 32 participants)

HR 0.89 (0.51 to 1.57)

⊕⊕⊝⊝
low3

4.6

ABT414(Depatux‐M)

(1 RCT; 86 participants)

HR 0.96 (0.69 to 1.34)

⊕⊕⊕⊝

low4,6

5.4

CED + LOM

(1 RCT, 129 participants)

HR 1.15 (0.76 to 1.74)

⊕⊕⊕⊝
moderate4

7.2

BEV + IRI
(1 RCT; 82 participants)
 

HR 1.16 (0.71 to 1.88)

⊕⊝⊝⊝
verylow4,5
 

7.4

BEV
(4 RCTs; 259 participants)

HR 1.22 (0.84 to 1.76)

⊕⊕⊝⊝

low4,6

8.1

CED

(1 RCT 131 participants)

HR 1.43 (0.97 to 2.12)

⊕⊕⊕⊝
moderate4

9.5

BEV + ONA

(1 RCT, 64 participants)

HR 1.76 (0.94 to 3.30)

⊕⊝⊝⊝
verylow4
 

10.3
 

Estimates are reported as HR: Hazard Ratio. CI: confidence interval. 

*This refers to the number of studies in the network evaluating the given intervention and the number of participants involved in these studies.

GRADE Working Group grades of evidence
High quality: further research is very unlikely to change our confidence in the estimate of effect.
Moderate quality: further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate.
Low quality: further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate.
Very low quality: we are very uncertain about the estimate.

*We excluded REG and ABT414 on sensitivity analysis, which ranked FOM first, BEV + LOM second, LOM third, BEV+irinotecan (IRI) fourth, and BEV fifth.

1 Downgraded −2 as sparse data from single small open‐label study

2 Downgraded for intransitivity (~ 30% of control arm received TMZ not LOM)

3 Downgraded for imprecision and sparse data from single small study

4 Imprecision

5 No direct evidence and HR for direct effect was estimated from trial report

6 Risk of bias

Open in table viewer
Summary of findings 2. Summary of progression‐free survival findings

Estimates of effects, certainty assessment and rankings of different treatment options compared with lomustine on overall survival in people with first recurrence of glioblastoma

Patient or population: people with first recurrence of glioblastoma 

Interventions: bevacizumab (BEV), BEV + lomustine (LOM), regorafenib (REG), BEV + irinotecan (IRI), BEV + onartuzumab (ONA), cediranib (CED), CED+LOM

Comparison: lomustine 

Outcome: Progression‐free survival
 

All intervention options

Relative effect and 95% CI (network estimate) **

Certainty of the evidence (GRADE)

Ranking*

(7 RCTs; 1383 participants in total)*

LOM
(4 RCTs; 317 participants) 

Reference comparator

Reference comparator

6.2

BEV+LOM
(3 RCTs, 401 participants) 

HR 0.57 (0.44 to 0.74)

⊕⊝⊝⊝
low1,4  

1.6

REG
(1 RCT; 59 participants)

HR 0.65 (0.42 to 1.01)

⊕⊝⊝⊝
very low1,2

2.7

CED + LOM

(1 RCT, 129 participants)

HR 0.76 (0.50 to 1.18)

⊕⊕⊕⊝
moderate2

3.8

BEV+IRI
(1 RCT; 82 participants) 

HR 0.80 (0.44 to 1.45)

⊕⊕⊝⊝
verylow1,3  

4.2

BEV
(4 RCTs; 200 participants)

HR 0.90 (0.58 to 1.38)

⊕⊝⊝⊝
low2,4

5.2

BEV + ONA

(1 RCT, 64 participants)

HR 0.98 (0.51 to 1.87)

⊕⊕⊝⊝
verylow1,4

5.8
 

CED

(1 RCT 131 participants)

HR 1.05 (0.68 to 1.62)

⊕⊕⊕⊝
moderate2

6.4

Estimates are reported as HR: Hazard Ratio. CI: confidence interval. 

*This refers to the number of studies in the network evaluating the given intervention and the number of participants involved in these studies.

GRADE Working Group grades of evidence
High quality: further research is very unlikely to change our confidence in the estimate of effect.
Moderate quality: further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate.
Low quality: further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate.
Very low quality: we are very uncertain about the estimate.

*We excluded REG on sensitivity analysis, which ranked FOM first, BEV + LOM second, LOM third, BEV+irinotecan (IRI) fourth, and BEV fifth.

1 Sparse data from single small open‐label study

2 Imprecision

3 HRs for direct effect estimated from trial report

4 Risk of bias

Open in table viewer
Summary of findings 3. Summary of findings for severe adverse events ‐ 1

Estimates of effects, certainty assessment and rankings of different treatment options compared with lomustine for severe adverse events in people with any recurrence of glioblastoma

Patient or population: people with any recurrence of glioblastoma

Interventions: bevacizumab (BEV) + lomustine (LOM), regorafenib (REG), cediranib (CED), CED + LOM, CED + gefitinib (GET)

Comparison: lomustine

Outcome: severe adverse events

Outcomes

(5 RCTs, 1024 participants)

Illustrative comparative risks* (95% CI)**

Relative effect
(95% CI)

Quality of the evidence
(GRADE)

Ranking

Corresponding risk

LOM
(5 RCTs; 330 participants)

39 per 100*

Reference comparator

N/A

1.7

CED

(2 RCTs; 147 participants)

39 per 100 (21 to 72)

RR 1.00 (0.54 to 1.85)

⊕⊕⊝⊝

moderate1

1.7

REG
(1 RCT; 59 participants)

74 per 100 (36 to 100)

RR 1.90 (0.92 to 3.95)

⊕⊕⊝⊝
low1,2

3.8

CED + GET

(1 RCT; 19 participants)

96 per 100 (18 to 100)

RR 2.46 (0.46 to 13.26)

⊕⊝⊝⊝
very low1, 3

4.3

BEV+ LOM

(2 RCTs, 346 participants)

98 per 100 (67 to 100)

RR 2.51 (1.72 to 3.66)

⊕⊕⊕⊕
high

4.7

CED + LOM

(1 RCT, 123 participants)

98 per 100 (50 to 100)

RR 2.51 (1.29 to 4.90)

⊕⊕⊕⊕
high

4.7

*The basis for thisrisk is the mean risk of SAEs with lomustine across the 5 studies that evaluated lomustine. The corresponding risk (and its 95% confidence interval) is based on this risk in the comparison group and the relative effect of the intervention (and its 95% CI).

** Where the CI exceeded 100 values were truncated (at 100)
CI: Confidence interval; RR: Risk Ratio

GRADE Working Group grades of evidence
High quality: further research is very unlikely to change our confidence in the estimate of effect.
Moderate quality: further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate.
Low quality: further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate.
Very low quality: we are very uncertain about the estimate.

1 Downgraded −1 for imprecision

2 Sparse data from single small open label trial

3 Downgraded −2 for imprecision

Open in table viewer
Summary of findings 4. Summary of findings for severe adverse events ‐ 2

Patient or population: people with any recurrence of glioblastoma

Interventions: bevacizumab (BEV)

Comparison: bevacizumab 9BEV), BEV+carboplatin (CAB), BEV+dasatinib (DAS), BEV+irinotecan (IRI), BEV+onartuzumab (ONA), BEV+TRC105, BEV+VB111, Fotemustine (FOM), BEV+HSPPC96 vaccine

Outcome: severe adverse events

Outcomes

(5 RCTs, 1024 participants)

Illustrative comparative risks* (95% CI)**

Relative effect
(95% CI)

Quality of the evidence
(GRADE)

Ranking

Corresponding risk

BEV
(8 RCTs; 498 participants)

36 per 100*

Reference comparator

N/A

3.1

FOM

(1 RCT, 32 participants)

16 per 100 (4 to 62)

RR 0.44 (0.11 to 1.72)

??? (missing)

1.6

BEV+HSPPC96

1 RCTs; 53 participants)

36 per 100 (12 to 100)

RR 1.01 (0.33 to 3.10)

⊕⊕⊝⊝
low1

3.4

BEV+ONA
(1 RCT; 64 participants)

42 per 100 (21 to 86)

RR 1.17 (0.57 to 2.39)

⊕⊕⊝⊝
low1

4.0

BEV+CAB

(1 RCT; 58 participants)

46 per 100 (22 to 96)

RR 1.27 (0.61 to 2.66)

⊕⊕⊝⊝
low1

4.4

BEV+DAS

(2 RCTs, 83 participants)

19 per 100 (25 to 100)

RR 0.52 (0.69 to 3.34)

⊕⊕⊕⊕
high

5.1

BEV+IRI

(1 RCT, 79 participants)

80 per 100 (43 to 100)

RR 2.22 (1.19 to 4.18)

⊕⊕⊕⊕
high

6.5

BEV+VB111

(1 RCT, 128 participants)

> 100 (92 to 100)

RR 3.77 (2.25 to 6.33)

⊕⊕⊕⊕
high

8.0

BEV+TRC 105

(1 RCT, 49 participants)

> 100 (92 to 100)

RR 6.86 (2.55 to 18.41)

⊕⊕⊕⊕
high

8.8

*The basis for thisrisk is the mean risk of SAEs with lomustine across the 5 studies that evaluated lomustine. The corresponding risk (and its 95% CI) is based on this risk in the comparison group and the relative effect of the intervention (and its 95% CI).

** Where the corresponding risk value and, or CI exceeded 100 values were truncated (at 100)
CI: Confidence interval; RR: Risk Ratio

GRADE Working Group grades of evidence
High quality: further research is very unlikely to change our confidence in the estimate of effect.
Moderate quality: further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate.
Low quality: further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate.
Very low quality: we are very uncertain about the estimate.

Background

Description of the condition

Gliomas are brain tumours that develop from supporting tissue of the brain known as glial cells. The most common and most malignant type of glioma is glioblastoma (GBM). The standard of care (Stupp protocol) for treating GBM in the first instance is surgery (maximal safe resection, which could be biopsy, debulking or resection depending on the tumour's anatomical location) to remove as much of the tumour as possible, followed by radiotherapy (60 Gy in 30 fractions) and chemotherapy (concurrent and adjuvant temozolomide) (NCCN 2018). This initial treatment takes approximately nine months to complete. Chemoradiotherapy has been associated with a median progression‐free survival of 6.5 months and a median overall survival of 14.6 months among reasonably fit people less than 70 years old (Stupp 2005). Approximately 25% of people receiving chemoradiotherapy are likely to be alive two years after diagnosis compared with approximately 10% who receive radiotherapy alone (Stupp 2005). With little improvement in five‐year survival rates over the last 40 years, approximately 12% of people are alive five years after diagnosis (CRUK 2020).

Younger people respond better to first‐line treatment than older people, and those with O⁶‐methylguanine‐DNA methyl‐transferase (MGMT) gene promoter methylation respond better to temozolomide than those with MGMT‐unmethylated status (Malmstrom 2012; Wick 2012). Amongst fitter elderly patients treated with chemoradiotherapy (using a shorter, 3‐week RT regime), MGMT‐methylated status confers a survival advantage, with a median survival of 13.5 months reported for this subgroup in a recent trial (Perry 2017). When GBM is diagnosed among patients who have had lower‐grade gliomas initially treated with radiotherapy only, they are generally treated with temozolomide after surgical confirmation of recurrence as GBM. Not all people receive radiotherapy or chemotherapy (or both) after surgery, however, and best supportive care (palliative care) may be the preferred option, particularly for elderly people and those with poor performance status (NCCN 2018).

After the initial treatment phase, guidelines issued by the National Institute for Health and Care Excellence (NICE) suggest that routine follow‐up by magnetic resonance imaging (MRI) be performed at three‐ to six‐month intervals for the first two years, six‐ to 12‐monthly until five years, and then annually thereafter (NICE 2018). Some tumours that are GBM to start with, after an initial response to treatment or stability in growth, can recur and grow.  In some cases of GBM, there is no period of response or stability and they continue to grow. Lower‐grade tumours (e.g. WHO Grade II or Grade III) can progress to GBM after many years of stability. In all cases, the continued growth is considered 'tumour progression'. Continued GBM growth or recurrence may be detected by these regular surveillance scans or identified upon the development of new symptoms (Thompson 2019). Making a diagnosis of GBM progression or recurrence can, however, be complicated in the first few months after initial treatment by the fact that its appearance on MRI may be indistinguishable from pseudoprogression (NCCN 2018).

As treatment of GBM is not curative, most people who respond to radiotherapy and temozolomide chemotherapy, in combination or sequentially, will experience a recurrence of the disease at some point thereafter, which is usually in the form of local tumour progression (Thon 2013). Following recurrence after chemoradiotherapy, a proportion of people will go on to receive further treatment; however, elderly and frail people are likely to receive best supportive care only.

Description of the intervention

Treatment options for recurrent GBM include the following.

Chemotherapy

This is the most common approach to treating recurrent disease (Thon 2013). The most commonly used chemotherapy regimes are either lomustine (CCNU) given as a single agent or given in combination with procarbazine and vincristine in the regime known as PCV or re‐challenge with temozolomide (NICE 2018; Niyazi 2011). In a chemotherapy‐naive population (i.e. populations that have not received the Stupp protocol) with a first recurrence, single‐agent temozolomide and PCV has been shown to have a similar effect on survival, with a median post‐recurrence survival of approximately seven months (Brada 2010; Parasramka 2017).

Re‐operation

A second surgical resection at recurrence may be possible in up to a quarter of people with recurrent disease depending on the infiltrative nature of the recurrence (Mandl 2008; Niyazi 2011). This also gives the opportunity for molecular analysis, which is helpful in guiding further treatment.

Re‐irradiation

Re‐irradiation in the context of recurrent GBM is usually given as hypofractionated radiotherapy, where the required dose is divided into a number of fractions for larger tumour volumes, with or without chemotherapy (concurrently or adjuvantly, or both), but may also be given as a single high‐fraction dose for small tumour volumes (stereotactic radiosurgery (SRS); Chapman 2019; Niyazi 2011).

Novel agents

There are several novel treatments for GBM recurrence that have been evaluated or are undergoing evaluation in clinical trials but none have been introduced into routine clinical practice. These include anti‐angiogenic therapy, local drug delivery, targeted molecular therapy, vaccines, and electric field therapy (tumour‐treating fields). The most intensively investigated of these alternatives is the anti‐angiogenic agent, bevacizumab. While this agent is currently licensed for use in the USA for treatment of recurrent GBM (Thon 2013), a 2018 review of anti‐angiogenic agents for GBM concluded that there was insufficient evidence to support the use of bevacizumab in recurrent disease (Ameratunga 2018).

Best supportive care

Best supportive (palliative) care only is considered a valid alternative to active treatment of recurrent GBM (Easaw 2011; NICE 2018). The Multinational Association for Supportive Care in Cancer defines supportive care as “the prevention and management of the adverse effects of cancer and its treatment. This includes management of physical and psychological symptoms and side effects across the continuum of the cancer experience, from diagnosis through anti‐cancer treatment to post‐treatment care. Enhancing rehabilitation, secondary cancer prevention, survivorship and end of life care are integral to supportive care” (MASCC 2019). People with GBM experience deteriorating neurological function as well as cancer effects; therefore supportive (palliative) care to improve quality of life and mitigate these effects has an important role to play in the management of this disease from an early stage (EANO 2017).

How the intervention might work

The mechanism of action of the alkylating chemotherapy agents (e.g. temozolomide, nitrosoureas, procarbazine, carboplatin) is to interfere with DNA synthesis by causing cross‐linkage between the strands and DNA breakage, thereby preventing tumour cell division (Drugs.com). Repeated surgical resection aims to reduce the tumour bulk and may only be effective if followed by chemotherapy or radiotherapy (Mandl 2008). Local re‐irradiation aims to deliver targeted radiotherapy to the tumour whilst sparing the surrounding normal tissue (Kim 2019; Niyazi 2011). Bevacizumab, the most common targeted therapy, is a monoclonal antibody that binds to and inhibits vascular endothelial growth factor, interfering with tumour blood supply and inhibiting vessel proliferation (Niyazi 2011). Supportive care in the context of GBM commonly includes the treatment of seizures, steroids (e.g. dexamethasone) to control brain oedema, neurocognitive dysfunction, nausea, and venous thromboembolism (Batchelor 2006).

Why it is important to do this review

There is a general acceptance that the two most effective treatment modalities in GBM are radiotherapy and temozolomide. However, there is no consensus on how to use these and other modalities after initial, first line GBM treatment. The 2015 James Lind Alliance research prioritisation‐setting process highlighted the need for more research guidance on GBM treatment after second recurrence (JLA 2015). In particular, a better understanding of the balance between desirable and undesirable effects associated with active treatment of recurrent GBM is necessary.

There are also significant resource implications associated with the management of GBM. A review by Messali 2014 found that the reported costs of managing GBM ranged from USD 4755 to USD 195,773 across five cost‐of‐illness studies (US dollar (USD) 2013). A greater understanding of the optimum management strategies for GBM will aid in the allocation of future healthcare resources in the most efficient way to maximise patient health. The aim of this review is therefore to identify and evaluate the best evidence on first and subsequent treatment options for when GBM recurs. This should inform conversations between people affected and health professionals, and the effective use of healthcare resources.

Objectives

To evaluate the effectiveness of further treatment/s for first and subsequent progression or recurrence of glioblastoma (GBM) among people who have received the standard of care (Stupp protocol) for primary treatment of the disease; and to prepare a brief economic commentary on the available evidence.

Methods

Criteria for considering studies for this review

Types of studies

Randomised controlled trials (RCTs), quasi‐randomised trials, non‐randomised studies, and controlled before‐and‐after studies that included relevant concurrent comparison groups. We did not expect to find cluster‐randomised trials. In view of the non‐stable nature of the conditions under review we did not include studies using cross‐over designs, nor did we include case‐control studies, or studies without a control group. As many novel interventions evaluated in this field are abandoned after early phase I/II studies due to futility, studies had to include a minimum of 20 participants. We excluded dose‐finding studies.

Types of participants

People aged 16 years of age and older diagnosed with recurrent or progressive disease following primary treatment (surgery and chemoradiotherapy) for glioblastoma (GBM). This included participants whose GBM continued to grow despite standard therapy and those whose disease was initially controlled by standard therapy but which subsequently recurred. Clinical trials included participants with either/both progressive or recurrent disease; definitions were determined by study investigators. For the purposes of this review, these are therefore considered as one entity. Where studies included mixed primary treatments, they were included if at least 80% of participants had received chemoradiotherapy using the standard 6‐week ‘Stupp protocol'. Participants with first and subsequent recurrences were included. Where studies included participants with grades 3 and 4 gliomas, we included them if data were reported separately for the GBM subgroup or if at least 80% of the sample had grade 4 gliomas.

Types of interventions

Any active treatment (chemotherapy, radiotherapy, surgery or another experimental treatment) or treatment combination compared with another active treatment, best supportive (palliative) care or no active treatment.

Types of outcome measures

Primary outcomes

  • Overall survival: survival from study entry until death from all causes, or as reported by investigators

  • Health‐related quality of life (QoL): as measured using a standardised questionnaire, e.g. the European Organisation for Research and Treatment of Cancer (EORTC) QLQ‐C30 or QLQ‐BN20 (specific for brain cancer), or the Functional Assessment of Cancer Therapy scale (FACT‐G (general) or FACT‐Br (specific for brain cancer))

Secondary outcomes

  • Progression‐free survival (survival from study entry to disease relapse, or as defined by investigators)

  • Severe adverse events (grade 3 or higher according to a standardised measurement tool, such as the Common Terminology Criteria for Adverse Events (CTCAE))

Search methods for identification of studies

Electronic searches

For evidence on the effectiveness of interventions, we prepared the search strategies and conducted the searches of the following databases from January 2005 (the threshold for the start of the current standard of care, namely maximal surgical resection followed by chemoradiotherapy) onwards (Appendix 1;Appendix 2; Appendix 3).

  • Cochrane Central Register of Controlled Trials (CENTRAL; 2019, Issue 12), in the Cochrane Library

  • MEDLINE via Ovid (2005 to December week 1 2019)

  • Embase via Ovid (2005 to 2019 week 50)

For economic evidence, we searched the NHS EED database from January 2005 up to the end of December 2014 (when the last records were added to that database); and MEDLINE and Embase from 1 January 2015 to 16 December 2019, as NHS EED already included comprehensive searches of these databases prior to 2015. We also considered relevant grey literature — such as health technology assessments, reports and working papers — for inclusion.

We did not apply language restrictions to any of the searches.

Searching other resources

Study authors searched the following for ongoing trials.

We handsearched the reference lists of included studies to identify newly published articles and additional studies of relevance. We searched neuro‐oncology conference abstracts from 2014 onwards.

Data collection and analysis

Selection of studies

The Information Specialist at the Gynaecological, Neuro‐oncology and Orphan Cancer Group (GNOC) downloaded all titles and abstracts retrieved by electronic searching to EndNote® and removed duplicates and those studies that clearly did not meet the inclusion criteria. A minimum of two reviewers (TL, ER, IL) independently screened the search results, rejecting all clearly irrelevant records and categorising the remaining articles into included studies, excluded studies, ongoing studies and studies awaiting classification. We recorded reasons for exclusion and identified any articles that related to the same study and grouped them. We obtained the full text of potentially eligible articles. We resolved any disagreements about eligibility by discussion with the other review authors.

Data extraction and management

Two reviewers (TL, ER, IL) independently extracted data, including the following items, from eligible studies using a piloted data extraction form.

  • Author contact details

  • Country

  • Setting

  • Dates of participant accrual

  • Trial registration number/identification

  • Funding source

  • Declarations of interest

  • Participant inclusion and exclusion criteria

  • Study design and methodology

  • Study population and baseline characteristics

    • Number of participants enrolled/analysed

    • Age

    • Gender

    • Performance status

    • MGMT‐methylation status

    • Type of primary surgery (biopsy or resection)

    • Details of initial treatment

    • Details of treatment of first recurrence

    • Time from initial diagnosis

  • Intervention details

    • Description of intervention

    • Description of comparator

  • Primary outcome/s of the study

  • Risk of study bias (see below)

  • Review outcomes

    • For time‐to‐event data (survival and disease progression), we extracted the log of the hazard ratio (log(HR)) and its standard error from trial reports. Where they were not explicitly reported, we estimated them from Kaplan‐Meier plots where possible.

    • For dichotomous outcomes, we recorded the number of participants in each treatment arm who experienced the outcome of interest and the number of participants assessed.

    • For continuous outcomes, we recorded the value and standard deviation of the outcome of interest and the number of participants assessed at the relevant time point in each group. We also recorded change‐from‐baseline score data where reported and noted the type of scale used.

We extracted both unadjusted and adjusted statistics where reported. Where possible, we extracted data to allow an intention‐to‐treat analysis, in which we analysed participants in the groups to which they were assigned. We resolved any differences between reviewers by discussion or by appeal to the other review authors.

Assessment of risk of bias in included studies

For randomised trials, we assessed risk of bias using Cochrane's tool and the criteria specified in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2019). This included assessment of:

  • random sequence generation;

  • allocation concealment;

  • blinding of participants and healthcare providers;

  • blinding of outcome assessors;

  • incomplete outcome data (we considered more than 20% missing data to be high risk);

  • selective reporting of outcomes;

  • other possible sources of bias, e.g. lack of a power calculation, baseline differences in group characteristics.

For non‐randomised studies we used the ROBINS‐I tool for assessing risk of bias (Sterne 2016). This included assessment of:

  • bias due to confounding (e.g. baseline differences in prognostic factors, or post‐baseline prognostic factor differences, or switching interventions);

  • bias due to participant selection (both intervention and comparison groups should comprise the same representative group);

  • bias in classification of interventions (e.g. differential misclassification of intervention status that is related to the outcome or the risk of the outcome);

  • bias due to deviations from intended interventions;

  • bias due to missing data (e.g. differential loss to follow‐up that is affected by prognostic factors);

  • bias due to outcome measures (e.g. outcome assessors were aware of intervention status, different methods were used to assess the outcome, or measurement errors were related to intervention status or effects);

  • bias in selection of the reported result.

Two review authors (TL, TD, ER) assessed risk of bias independently and resolved differences by discussion. We summarised judgements in 'Risk of bias' tables along with the characteristics of the included studies. We include both a risk of bias graph and a risk of bias summary. We considered the 'Risk of bias' assessment in our interpretation of the evidence.

Measures of treatment effect

We used the following measures to evaluate treatment effect.

  • For time‐to‐event data (e.g. death or disease progression) we used the hazard ratio (HR) with 95% confidence intervals (CIs).

  • For dichotomous outcomes, we calculated the effect size as a risk ratio (RR) with its 95% CIs.

  • For continuous outcomes measured using the same scale, we reported the mean difference (MD) between treatment groups with 95% CIs. For continuous outcomes (e.g. QoL scores) in which different measurement scales had been used, or if studies report change‐from‐baseline instead of final values, we combined these data using the (unstandardised) mean difference method in Review Manager 5 (RevMan 5) (Review Manager 2014).

Network structure

Where possible, we aimed to compare and rank the following types of interventions.

  • Different chemotherapy agents and regimens (temozolomide, PCV, lomustine/CCNU, etc.)

  • Targeted antiangiogenic agents (e.g. bevacizumab) and other anti‐growth‐factor agents

  • Other immunotherapy, e.g. tumour‐derived vaccines, viral therapy

  • Re‐operation

  • Re‐irradiation

  • Tumour‐treating fields

  • Supportive care

Unit of analysis issues

Two review authors (TL and ER) reviewed any unit‐of‐analysis issues according to Higgins 2019 for each included study and we resolved any differences through discussion. We considered issues such as where there are multiple observations for the same outcome, e.g. repeated measurements with different scales, or outcomes measured at different time points to those stipulated in the review protocol.

Multi‐arm trials

For multi‐arm trials, we treated the multiple comparisons as independent in pairwise meta‐analyses. In the network meta‐analysis, we accounted for the correlation between the effect sizes derived from the same study.

Dealing with missing data

We did not impute missing data. Where missing data were substantial, we took this into consideration in our grading of the evidence.

Assessment of heterogeneity

Assessment of clinical and methodological heterogeneity

We assessed clinical heterogeneity between studies by comparing the studies’ characteristics of included participants, and interventions in each meta‐analysis of each comparison; by visual inspection of forest plots; by estimation of the percentage heterogeneity between trials which cannot be ascribed to sampling variation (Higgins 2003); and by a formal statistical test of the significance of the heterogeneity (Deeks 2001). If there was evidence of substantial heterogeneity, we investigated it and reported the possible reasons for it.

Assessment of consistency across treatment comparisons

We examined the assumption of consistency by assessing the distribution of potential effect modifiers across the pair‐wise comparisons. The assumption held if the following were true.

  • The common treatment used to compare different interventions indirectly is similar when it appears in different trials.

  • All pairwise comparisons do not differ with respect to the distribution of effect modifiers.

The potential treatment modifiers are as follows.

  • Re‐operation

  • MGMT‐methylation status

  • First or subsequent recurrence

  • Time from primary diagnosis

Assessment of statistical heterogeneity and inconsistency
Assumptions when estimating the heterogeneity

We estimated heterogeneity indicators for each pairwise comparison. In network meta‐analysis, we assumed a common estimate for the heterogeneity variance across the different comparisons.

Measures and tests for heterogeneity

We assessed the presence of statistical heterogeneity within the pairwise comparisons using the I² statistic, which is the percentage of variability that cannot be attributed to random error. We based the assessment of statistical heterogeneity in the network on the magnitude of the heterogeneity variance parameter (Tau²) estimated from the network meta‐analysis models.

Assessment of statistical inconsistency

We evaluated the statistical agreement between the various sources of evidence in a network of interventions (consistency) by global and local to complement the evaluation of consistency (Efthimiou 2016).

Assessment of reporting biases

We assessed each paper for the extent and transparency of reporting and for suggestion of reporting bias. We did not find sufficient studies of similar interventions to assess publication bias using funnel plots.

Data synthesis

For effectiveness studies
Methods for direct treatment comparisons

We carried out meta‐analyses in Stata software (version 15), pooling data from studies measuring the same outcomes in similar populations (first recurrence and any recurrence, including mixed populations). Assuming that we found at least two included studies that were sufficiently similar for the findings to be clinically meaningful, we used the random‐effects models with inverse variance weighting for all meta‐analyses. If any studies contributing to a meta‐analysis had multiple intervention groups, we divided the ‘shared’ comparison group into the number of treatment groups and comparisons between each treatment group and treated the split comparison group as independent comparisons. If meta‐analysis was not possible due to the timing of assessment or the type of outcome measure used, we described these data narratively.

Methods for indirect and mixed comparisons

We conducted network meta‐analyses providing that populations of included studies were sufficiently similar to satisfy the assumption of joint randomisation and that the interventions connected, creating a network. This led to two separate networks, one for studies evaluating populations experiencing first recurrence and one for those experiencing any, first and second and subsequent recurrences. The latter populations would be expected to have a worse prognosis than the first recurrence group. We used the random‐effects model in Stata software (version 15) fitting a multivariate network meta‐analysis (White 2015). In 'Summary of findings' tables, we report the value of mean rank for included treatments (Chaimani 2015).

For data where meta‐analysis was not possible, we attempted narrative synthesis but did not grade the evidence. In general, we interpreted the quality of the evidence based on the Cochrane Effective Practice and Organisation of Care (EPOC) Group’s guidance (EPOC 2015).

'Summary of findings' table and results reporting

Based on the methods described in Chapter 11 of the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2019), we prepared a 'Summary of findings' table to present the results of the following outcomes, namely:

  • overall survival;

  • progression free survival; and

  • severe adverse events.

There were insufficient data to present QoL findings. We used the GRADE system to rank the quality of the evidence (Schünemann 2019). Two review authors (TL and ER) independently graded the evidence and resolved any differences by discussion or, if necessary, by involving a third review author. We interpreted the results of the graded evidence based on Cochrane Effective Practice and Organisation of Care guidance (EPOC 2015).

Brief economic commentary

We included a brief economic commentary that summarises the availability and principal findings of the economic evaluations relevant to this review. This includes evaluations alongside trials and model‐based evaluations. The work was performed in line with current guidelines, including a supplementary search to identify economic studies (Shemilt 2019).

Subgroup analysis and investigation of heterogeneity

We analysed data according to studies of populations experiencing a first recurrence and studies with other populations experiencing any (mixed group) or second or subsequent recurrences. We did not conduct subgroup analyses and investigate heterogeneity according to second or subsequent recurrence, MGMT promoter methylation status, and time from primary diagnosis, as data were insufficient for this purpose. We did not find studies specifically of transformed GBM and therefore did not conduct separate analysis of these data.

Sensitivity analysis

In the network meta‐analyses, we explored how the following factors affect the ranking of interventions.

  • Study quality, by excluding studies at high risk of bias to investigate how study quality affected the evidence on effects and the certainty of findings.

  • If the effects from a multi‐arm trial created a single loop in the network (no other loops available), we explored how exclusion of one of the arms affected the NMA findings.

Results

Description of studies

Results of the search

The original search conducted by the CGNOC Information Specialist on 16 December 2019 identified the following records.

Searches for studies of effectiveness

  • CENTRAL Issue 12 2019 – 524 references

  • MEDLINE: 2005 to December week 1 2019 – 1632 references

  • Embase: 2005 to 2019 week 50 – 956 references

  • Preliminary de‐duplication combined total n = 2738 references

Economic searches

  • NHS EED – 9 refs

  • MEDLINE: 2015 to December week 1 2019 – 23 references

  • Embase: 2015 to 2019 week 50 – 58 references

  • Preliminary de‐duplication combined total n = 88 references

For studies of effectiveness, we shortlisted 182 records and obtained the full text of these papers where applicable (several were conference abstracts). Where clinical trial registrations were identified, we visited ClinicalTrials.gov for further trial details. These records were classified as follows.

  • Included: 42 studies with 85 related records (including 35 conference abstracts and 6 clinical trial registrations)

  • Excluded: 57 studies with 69 related records

  • Ongoing: 20 studies with 28 related records

SeeFigure 1.

Figure 1
Flow diagram of searches for studies of effectiveness conducted on 16/12/2019

Flow diagram of searches for studies of effectiveness conducted on 16/12/2019

We identified one new (May 2020) trial report related to an already included study after the review was completed (van den Bent 2018).

Included studies

We included 34 RCTs and 8 non‐RCTs. Most RCTs were conducted in multiple centres across several countries with accrual occurring between 2004 and 2018. All participants had recurrent GBM and the vast majority had received chemoradiotherapy as first line treatment. The treatment of first recurrence was most commonly evaluated (20 studies; Azoulay 2017; Batchelor 2013; Brandes 2016a; Brandes 2016b; Brandes 2018; Brown 2016; Cloughesy 2017; Dresemann 2010; Kunwar 2010; Lombardi 2019; Narita 2019; Omuro 2018; Puduvalli 2018; Reardon 2015b; Scorsetti 2015; Suchorska 2016; Taal 2014; Twelves 2017; van den Bent 2018; Wick 2017). Treatment of first and second recurrences were evaluated in six studies (Friedman 2009; Reardon 2018a; Reardon 2018b; Reardon 2020; Wick 2010; Wick 2014); first, second and third recurrence in one study (Weathers 2016); any recurrence in seven studies (Duerinck 2018; Field 2015; Galanis 2017; Gilbert 2017; Modh 2018; Reardon 2011; Stupp 2012); and in the remainder the number of recurrences was not clear. Data were rarely reported separately for first and subsequent recurrences where populations were mixed.

Nine of the RCTs were phase 3 studies (Batchelor 2013; Cloughesy 2017; Cloughesy 2018; Dresemann 2010; Kunwar 2010; Narita 2019; Stupp 2012; Wick 2010; Wick 2017); the rest were phase 2. Most RCTs recruited patients from Europe and America in multicentre study designs; two RCTs were conducted in Japan (Narita 2019; Omuro 2018).

Sample sizes ranged from 20 to 437 participants, with the total number of participants enrolled to the RCTs numbering 4607 (2573 with first recurrence and 2016 with mixed populations). Participants studied in non‐randomised studies numbered 629, bringing the total number taking part in included studies to 5236 people.

Interventions evaluated in the RCTs

Most interventions were evaluated in single studies leading to 33 different comparisons evaluated in the RCTs alone. (Underlined studies reported hazard ratios (HRs) for survival outcomes; studies that did not report HRs usually reported survival outcomes as median survival). Included RCTs were:

  1. cediranib (CED) + lomustine (LOM) vs lomustine (LOM)(Batchelor 2013);

  2. HSPPC‐96 vaccine + bevacizumab (BEV) vs BEV; (Bloch 2017);

  3. galunisertib (GAL) + LOM vs LOM(Brandes 2016a);

  4. BEV vs fotemustine (FOT) (Brandes 2016b);

  5. BEV + LOM vs LOM(Brandes 2018; Wick 2017);

  6. CED + gefitinib (GET) vs CED (Brown 2016);

  7. onartuzumab (ONA) + BEV vs BEV(Cloughesy 2017);

  8. VB‐111 + BEV vs BEV (Cloughesy 2018);

  9. Imatinib + hydroxyurea (HU) vs HU (Dresemann 2010);

  10. axitinib (AXI) + LOM vs AXI (Duerinck 2018);

  11. BEV + carboplatin (CAB) vs BEV (Field 2015);

  12. BEV + irinotecan (IRI) vs BEV (Friedman 2009);

  13. BEV vs BEV + TRC105 (Galanis 2017);

  14. desatinib + BEV vs BEV (Galanis 2019);

  15. BEV + IRI vs BEV + TMZ (Gilbert 2017);

  16. convection enhanced cintredekin besudotox vs gliadel wafers (Kunwar 2010);

  17. regorafenib (REG) vs LOM (Lombardi 2019);

  18. fractionated stereotactic radiosurgery with BEV vs BEV with chemotherapy (Modh 2018);

  19. personalized peptide vaccination (PPV) vs placebo + best supportive care (Narita 2019);

  20. nivolumab (NIV) vs nivolumab (NIV) + ipilimumab (IPI) (Omuro 2018);

  21. BEV vs BEV + vorinostat (Puduvalli 2018);

  22. metronomic etoposide + BEV vs temozolomide + BEV (Reardon 2011);

  23. afatinib (AFA) vs TMZ vs AFA + TMZ (Reardon 2015b);

  24. rindopepimut vaccine + BEV vs placebo + BEV (Reardon 2020);

  25. pembrolizumab vs PEM + BEV (Reardon 2018b);

  26. tumour‐treating fields (TTF) vs chemotherapy (various)(Stupp 2012);

  27. BEV + LOM vs BEV or LOM (Taal 2014);

  28. hypofractionated radiotherapy + BEV vs BEV (Tsien 2019);

  29. cannabidiol:delta‐9‐tetrahydrocannabinol (CBD:THC) vs placebo (Twelves 2017);

  30. Depatux‐m (ABT414) vs depatux‐m + TMZ vs TMZ or LOM (van den Bent 2018);

  31. BEV vs low dose BEV + LOM (Weathers 2016);

  32. enzastaurin (ENZ) vs LOM (Wick 2010);

  33. asunercept (APG110) + radiotherapy vs radiotherapy (Wick 2014).

Underlined studies reported hazard ratios (HRs) for survival outcomes; studies that did not report HRs, usually reported survival outcomes as median survival. Bloch 2017, Galanis 2017, Modh 2018, Puduvalli 2018, Reardon 2018b,and Tsien 2019 were published as conference abstracts that contained little data. The studies of novel agents imatinib (Dresemann 2010), cediranib (Batchelor 2013), PPV (Narita 2019); nivolumab ± ipilimumab (Omuro 2018), pembrolizumab (Reardon 2018b), enzastaurin (Wick 2010), and afatinib (Reardon 2015b) did not show clinically meaningful survival benefits. Similarly, no survival benefits were noted when onartuzumab (Cloughesy 2017), HSPPC‐96 vaccine (Bloch 2017), carboplatin (Field 2015), irinotecan (Friedman 2009), TRC105 (Galanis 2017), desatinib (Galanis 2019), vorinostat (Puduvalli 2018), or metronomic etoposide or TMZ (Reardon 2011 ) were added to BEV.  

Interventions evaluated in the seven non‐RCTs were the following.

  1. Re‐operation vs no re‐operation (Azoulay 2017; retrospective)

  2. Re‐operation vs no re‐operation (Suchorska 2016; prospective)

  3. BEV vs best supportive care (Cuncannon 2019; prospective);

  4. Gamma Knife surgery (GKS) vs TMZ vs GKS + TMZ, vs re‐operation vs other (Kim 2015; retrospective);

  5. Re‐operation ± radiotherapy + chemotherapy vs chemotherapy (fotemustine + re‐challenge TMZ) (Scorsetti 2015; retrospective)

  6. Trebananib (TNB) vs TNB + BEV (Reardon 2018a)

  7. BEV+CCNU (LOM) vs BEV (Heiland 2016; retrospective study)

  8. Intranasal perillyl alcohol (IPA) + ketogenic diet vs IPA + standard diet (Santos 2018)

For details of individual studies please see Characteristics of included studies.

Excluded studies

Excluded studies numbered 57 and reasons for exclusion and reasons for exclusion of individual studies can be found in the Characteristics of excluded studies section. WE also identified 20 ongoing studies, and details of these can be found in the Characteristics of ongoing studies section.

Risk of bias in included studies

We summarise risk of bias in included studies in Figure 2 and Figure 3. In general, we judged RCTs to be at low or unclear risk of bias, and judged non‐RCTs to be at high risk of bias. We generally judged studies reported as conference abstracts only as being at unclear risk of bias as they contained insufficient information to make judgements (Bloch 2017; Galanis 2017; Modh 2018; Puduvalli 2018; Reardon 2018b; Tsien 2019; Twelves 2017).

Figure 2
Risk of bias of included studies

Risk of bias of included studies

Allocation

Most randomised studies were at an unclear risk of selection bias as the randomisation and treatment allocation process was seldom clearly reported. All non‐randomised studies were at a high risk of selection bias, as patients in Azoulay 2017, Heiland 2016, Kim 2015, Santos 2018, Scorsetti 2015 and Suchorska 2016 were most likely selected for different study treatments based on clinical factors. Cuncannon 2019 selected patients according to willingness to pay for treatment with bevacizumab, which may have been influenced by patient prognosis. Reardon 2018a was a non‐randomised study with little information on how patients were allocated to the different treatment arms.

Blinding

Most studies were open label studies. Less than 25% had blinding of participants and personnel and less than 40% applied assessor blinding to assessments. In grading the findings, however, we assumed a low risk of bias for this criterion with respect to overall survival, which is an objective outcome.

Incomplete outcome data

Most studies were judged to be at low (~ 60%)  or unclear risk (~ 35%) of attrition bias. We judged one non‐randomised study to be at high risk of attrition bias because less than half the participants completed the study (Santos 2018).

Selective reporting

The majority of studies (~ 70%) reported overall survival, progression free survival and toxicity outcomes and we judged them to be at low risk of bias for this criterion; the remainder we judged as having an unclear risk of bias.

Other potential sources of bias

RCTs were usually sponsored by the intervention's manufacturer and affiliated pharmaceutical companies and most had authors with declared interests. The risk of bias implications of these potential sources of bias was judged as unclear in all instances.

Effects of interventions

See: Summary of findings 1 Summary of overall survival findings; Summary of findings 2 Summary of progression‐free survival findings; Summary of findings 3 Summary of findings for severe adverse events ‐ 1; Summary of findings 4 Summary of findings for severe adverse events ‐ 2

We found median survival data for most included studies and these are presented in Table 1, grouped according to the level of recurrence. NMAs for survival outcomes were performed for studies evaluating treatments for first recurrence only. Where studies evaluated further recurrence or mixed recurrence, we did not perform NMA as networks connecting these mainly single phase 2 studies of novel interventions among mixed populations that were mostly shown not to warrant further investigation would produce very low certainty results. Several of the studies in the latter network did not report hazard ratios and were not powered to test efficacy, and most reported no clinically meaningful survival effects or were terminated early for futility.

Open in table viewer
Table 1. Median survival of participants in included studies

Study ID

No. participants

No. of recurrences

Comparison

Median PFS (months)

Median OS

(months)

Study design

Comment

First recurrence 

Azoulay 2017 

183

 1st

I: Re‐operation

C:No re‐operation

 NR

9.8

5.0

NRS

Patients selected for different arms based on clinical factors therefore a high risk of selection bias.

Batchelor 2013

325

1st

I: CED

I: CED+LOM

C: LOM

3.0

4.1

2.7

8.0

9.4

9.8

Phase 3

CED did not significantly improve survival.

Brandes 2016b

91

1st

I: BEV

C: FOT

3.4

3.5

7.3

8.7

Phase 2

Investigators concluded that single‐agent BEV "may have a role'.

Brandes 2018

123

1st

I: BEV+LOM

C: LOM

2.3

1.8

6.4

5.5

Phase 2

Terminated early due to high drop‐out rate during first‐line treatment.

Brandes 2016a

119

1st

I: GAL+LOM

C: LOM

1.8

1.9

6.7

7.5

Phase 2

Investigators concluded that GAL failed to demonstrate activity.

Brown 2016

38

1st

I: CED+GEF

C: CED

3.6

2.8

7.2

5.5

Phase 2

Study was underpowered.

Cloughesy 2017

129

1st

I: ONA+BEV

C: BEV

3.9

2.9

8.8

12.6

Phase 3

No evidence of clinical benefit with addition of ONA

Dresemann 2010

240

1st

I: IMA+HU

C: HU

1.4

1.4

NR

Phase 3

No clinically meaningful differences.

Lombardi 2019

119

1st

I: REG

C: LOM

2.0

1.9

7.4

5.6

Phase 2

Considered by investigators to have potential for further study.

Kunwar 2010 

296

1st

I: cintredekin besudotox

C: gliadel wafers

NR

9.1

8.8

Phase 3

No survival difference but higher risk of pulmonary embolism with cintredekin besudotox (P = 0.014).

Narita 2019

88

1st

I: PPV

C: Placebo

NR

8.4

8.0

Phase 3

Did not meet primary endpoint and PPV shortened the OS in certain patients.

Omuro 2018

20

1st

I: NIV

C:NIV+IPI

1.9

1.5

10.4

9.2

Phase 2

NIV was better tolerated than NIV+IPI

Puduvalli 2018 

74

1st

I: BEV+VOR

C: BEV

3.7

3.9

7.8

9.3

Phase 2

 No clinical benefit with BEV+VOR. Does not warrant further investigation.

Reardon 2015b

78

1st

I: AFA+TMZ

I: AFA

C: TMZ

1.5

1.0

1.9

8.0

9.8

10.6

Phase 2

Investigators concluded that afatinib had limited single agent activity.

Scorsetti 2015 

43

1st

I: Re‐operation

C: No re‐operation

15

5

17

6

NRS

Patients selected for different arms based on clinical factors therefore a high risk of selection bias. 

Suchorska 2016

93

1st
 

I: re‐operation

C: no re‐operation

2.0

1.8

11.4

9.8

NRS
 

Patients selected for different arms based on clinical factors therefore a high risk of selection bias. Complete resection was associated with improved survival compared with incomplete resection. 

Taal 2014

153

1st

I: LOM

I: BEV+LOM

C: BEV

1.0

4.0

3.0

8.0

12.0

8.0

Phase 2

BEV+LOM met investigator criteria for further evaluation in phase 3 studies. Single agent BEV was found to have low activity.

Twelves 2017

21

1st

I: CBD:THC

C: placebo

‐

‐

18.3

12.3

All patients received dose intense TMZ. 1 year survival was 83% and 56% in the CBD:THC and placebo groups, respectively.

van den Bent 2018

260

1st

I: ABT414+TMZ

I: ABT414

C: TMZ or LOM

2.7

1.9

1.9

9.6

7.9

8.2

Phase 2

ABT414 had insufficient single‐agent activity.

Wick 2017

437

1st

I: LOM+BEV

C: LOM

4.2

1.5

9.1

8.6

Phase 3

No OS benefit with combination.

Any, second or subsequent recurrence

Bloch 2017

90

1st or 2nd (% not reported)

I: HSPPC‐96+BEV

C: BEV

NR

7.5

10.7

Phase 2

Terminated for futility after interim analysis.

Cuncannon 2019 

43

2nd or 3rd relapse

I: BEV

C: BSC

NR 

6.0

1.0

NRS

Patients selected for different arms based on patient choice therefore high risk of selection bias.

Friedman 2009

167

1st and 2nd (< 20%)

I: BEV+IRI

C: BEV

5.6

4.2

8.7

9.2

Phase 2

No survival benefit with combination.

Reardon 2018a

80

1st and 2nd (% not reported)

I: PEM

C: PEM+BEV

NR

8.8

10.3

Phase 2

Investigators reported that there was no monotherapy activity.

Reardon 2018b

48

1st and 2nd (33%)

I: TNB

C: TNB+BEV

0.7

3.6

11.4

9.5

NRS

Accrual to TNB was discontinued early due to lack of monotherapy activity.

Cloughesy 2018

256

1st and 2nd (27%)

I: VB111+BEV

C: BEV

3.4

3.7

6.8

7.9

Phase 3

VB111+BEV failed to improve outcomes

Duerinck 2018

79

Any

I: AXI

C: AXI+LOM

2.9

3.0

6.3

6.7

Phase 2

No indication that AXI+LOM improves results.

Field 2015

122

Any (33% 2nd or subsequent)

I: BEV+CAB

C: BEV

3.5

3.5

6.9

7.5

Phase 2

Adding CAB resulted in more toxicity without additional clinical benefit.

Galanis 2017

101

2nd or later

I: TRC105+BEV

C: BEV

2.9

3.2

10.0

7.4

Phase 2

Investigators reported no significant survival difference with TRC105.

Galanis 2019

121

Any (% not reported)

I: DAS+BEV

C: BEV

3.2

3.2

7.3

7.7

Phase 2

DAS+BEV did not significantly improve clinical outcomes.

Gilbert 2017

117

Any (% not reported)

I: BEV+TMZ

C: BEV+IRI

4.7

4.1

9.4

7.7

Phase 2

Both arms surpassed predetermined efficacy thresholds.

Heiland 2016

43

3rd

I: BEV

C: BEV+LOM

2.3

6.1

4.1

6.6

NRS

Investigators concluded that last‐line therapy with BEV/LOM results in a longer PFS and OS compared to BEV only.

Modh 2018

34

"median of 3"

I: FSRS + BEV

C: Chemo+BEV

5.3

1.8

7.1

4.8

RCT

Investigators concluded that FSRS in heavily pretreated patients with recurrent malignant glioma is feasible and improves local control and PFS

Reardon 2011

23

"heavily pre‐treated"

I:ETO+BEV

C:TMZ+BEV

1.9

0.9

4.4

2.9

Phase 2

Investigators concluded that metronomic ETO or TMZ is ineffective administered with BEV in this context.

Santos 2018

32

Unclear

I: IPA + ketogenic diet

C: IPA + regular diet

NR

NR

NRS

Investigators concluded that results on ketogenic diet were encouraging.

Stupp 2012

237

Any 

(88% 2nd or later)

I: TTF

C: BPC

2.2

2.1

6.6

6.0

 Phase 3

No improvement in survival but toxicity and QOL favoured TTF.

Tsien 2019

170

NR

I: BEV+RT

C: BEV

NR

10.1

9.7

Phase 2

BEV +RT was associated with improved 6‐month PFS (54% vs 29%).

Weathers 2016

69

1st (68%), 2nd, and 3rd

I: LOM+BEV (low dose)

C: BEV

4.3

4.1

9.6

8.3

Phase 2

Median PFS for 1st recurrence (47 pts) was 5.0 mths vs 3.2 mths, respectively; median OS was 13.1 mths vs 8.8 mths, respectively.

Wick 2010

266

1st and 2nd (25%)

I: ENZ

C: LOM

1.5

1.6

6.6

7.1

Phase 3

Terminated early due to futility.

Wick 2014

84

1st and 2nd/3rd (29%)

I: APG101+RT

C: RT

4.5

2.5

11.5

11.5

Phase 2

Investigators concluded that APG101 had potential for further clinical development.

BEV = bevacizumab
BPC = Best Physician's Care
BSC = Best supportive care
CAB = carboplatin
CBD:THC = cannabidiol:delta‐9‐tetrahydrocannabinol
CED = cediranib
DAS = desatinib
ENZ = enzastaurine
ETO = etoposide
HU = hydroxyurea
IMA =imatinib
IPA = intranasal perillyl alcohol
IRI = irinotecan
IPI = ipilimumab
LOM = lomustine
NIV = nivolumab
PEM = pembrolizumab
PPV = personalised petide vaccine
RT = radiotherapy
ONA = onartuzumab
TTF = tumour‐treating fields
TNB = trebananib
FSRH = fractionated stereotactic radiotherapy
TMZ = temozolomide

Overall survival (first recurrence)

The NMA findings

Nine RCTs involving the following 11 treatments contributed to this network (Figure 4). Median overall survival estimates across this group of studies ranged from 5.5 months (LOM arm of Brandes 2018) to 12.6 months (BEV arm of Cloughesy 2017) (Table 1).

Figure 4
Network for Overall Survival (first recurrence)

Network for Overall Survival (first recurrence)

Results for this network can be found in the forest plot (Figure 5) and also in the league table showing HRs and 95% CI estimates for all intervention comparisons (Table 2). The global test for inconsistency was not statistically significant (P = 0.15). We found no high‐certainty evidence that any of the treatments evaluated were superior to lomustine monotherapy. Graded pooled network estimates for overall survival of treatments compared with lomustine monotherapy suggest the following.

Figure 5
Forest plot of effects on overall survival of different treatments compared with lomustine

Forest plot of effects on overall survival of different treatments compared with lomustine

Open in table viewer
Table 2. Overall survival: League table showing HR and 95% CI estimates for all intervention comparisons

LOM

ABT414

ABT414_TMZ

BEV

BEV_IRI

BEV_LOM

BEV_ONA

CED

CED_LOM

FOT

REG

LOM

0.96 (0.69,1.34)

0.66 (0.47,0.92)

1.22 (0.84,1.76)

1.16 (0.71,1.88)

0.91 (0.75,1.10)

1.76 (0.94,3.30)

1.43 (0.97,2.12)

1.15 (0.76,1.74)

0.89 (0.51,1.57)

0.50 (0.33,0.76)

1.04 (0.75,1.45)

ABT414

0.68 (0.49,0.95)

1.27 (0.77,2.08)

1.20 (0.67,2.17)

0.95 (0.64,1.39)

1.83 (0.90,3.73)

1.49 (0.89,2.50)

1.20 (0.71,2.03)

0.93 (0.48,1.79)

0.52 (0.31,0.89)

1.52 (1.09,2.12)

1.46 (1.05,2.04)

ABT414_TMZ

1.85 (1.13,3.04)

1.76 (0.98,3.17)

1.38 (0.94,2.03)

2.68 (1.32,5.46)

2.18 (1.30,3.65)

1.75 (1.03,2.97)

1.36 (0.70,2.62)

0.76 (0.45,1.30)

0.82 (0.57,1.19)

0.79 (0.48,1.30)

0.54 (0.33,0.89)

BEV

0.95 (0.70,1.30)

0.75 (0.52,1.08)

1.45 (0.87,2.41)

1.18 (0.69,2.02)

0.95 (0.54,1.64)

0.73 (0.48,1.13)

0.41 (0.24,0.72)

0.86 (0.53,1.40)

0.83 (0.46,1.49)

0.57 (0.32,1.02)

1.05 (0.77,1.44)

BEV_IRI

0.79 (0.48,1.28)

1.52 (0.84,2.77)

1.24 (0.67,2.31)

0.99 (0.53,1.88)

0.77 (0.45,1.31)

0.43 (0.23,0.82)

1.10 (0.91,1.33)

1.06 (0.72,1.55)

0.72 (0.49,1.06)

1.34 (0.92,1.94)

1.27 (0.78,2.07)

BEV_LOM

1.94 (1.03,3.64)

1.58 (1.02,2.44)

1.27 (0.80,1.99)

0.98 (0.56,1.73)

0.55 (0.35,0.87)

0.57 (0.30,1.07)

0.55 (0.27,1.11)

0.37 (0.18,0.76)

0.69 (0.41,1.15)

0.66 (0.36,1.20)

0.52 (0.27,0.97)

BEV_ONA

0.81 (0.39,1.71)

0.65 (0.31,1.39)

0.51 (0.26,0.99)

0.28 (0.13,0.60)

0.70 (0.47,1.03)

0.67 (0.40,1.12)

0.46 (0.27,0.77)

0.85 (0.50,1.45)

0.81 (0.43,1.50)

0.63 (0.41,0.98)

1.23 (0.59,2.58)

CED

0.80 (0.54,1.20)

0.62 (0.31,1.24)

0.35 (0.20,0.62)

0.87 (0.58,1.31)

0.84 (0.49,1.42)

0.57 (0.34,0.97)

1.06 (0.61,1.84)

1.01 (0.53,1.90)

0.79 (0.50,1.24)

1.53 (0.72,3.24)

1.25 (0.83,1.86)

CED_LOM

0.78 (0.38,1.56)

0.44 (0.24,0.78)

1.12 (0.64,1.98)

1.08 (0.56,2.08)

0.74 (0.38,1.42)

1.36 (0.89,2.10)

1.30 (0.76,2.21)

1.02 (0.58,1.80)

1.97 (1.01,3.85)

1.61 (0.81,3.20)

1.29 (0.64,2.60)

FOT

0.56 (0.28,1.13)

1.99 (1.32,3.01)

1.92 (1.13,3.25)

1.31 (0.77,2.22)

2.42 (1.40,4.21)

2.31 (1.22,4.35)

1.81 (1.15,2.85)

3.51 (1.66,7.44)

2.86 (1.62,5.04)

2.29 (1.28,4.10)

1.78 (0.88,3.58)

REG

ABT414 = depatux‐m
BEV = bevacizumab
CED = cediranib
IRI = irinotecan
LOM = lomustine
ONA = onartuzumab

  • There is probably little or no difference between BEV + LOM and LOM only (HR 0.91, 95% CI 0.75 to 1.10; moderate‐certainty evidence).

  • There may be little or no difference between FOM and LOM (HR 0.89, 95% CI 0.51 to 1.57; low‐certainty evidence)

  • There is probably little or no difference between BEV and LOM (HR 1.22, 95% CI 0.84 to 1.76; low‐certainty evidence)

  • REG may be more effective than LOM (HR 0.50, 95% CI 0.33 to 0.76; low‐certainty evidence)

  • ABT414 + TMZ may be more effective than LOM (HR 0.66, 95% CI 0.47 to 0.92; low‐certainty evidence)

  • CED is probably less effective than LOM (HR 1.43, 95% CI 0.97 to 2.12; moderate‐certainty evidence)

  • There is probably little or no difference between CED + LOM and LOM (HR 1.15, 95% CI 0.76 to 1.74; moderate‐certainty evidence)

  • Evidence on BEV + ONA versus LOM (HR 1.76, 95% 0.94 to 3.30) and BEV + IRI versus LOM (HR 1.16, 95% CI 0.71 to 1.88) was very low certainty.

When treatments other than LOM were compared with BEV monotherapy, there was no clear difference in effect between any of the treatments and BEV for this outcome, except for ABT414 + TMZ, which the evidence suggested may be more effective than BEV (HR 0.54, 95% CI 0.33 to 0.89; low‐certainty evidence). The evidence also suggested that there is probably little or no difference between BEV + IRI compared with BEV monotherapy (HR 0.95, 95% CI 0.70 to 1.30; moderate‐certainty evidence).

On sensitivity analysis, when we excluded Lombardi 2019 (a small study of REG) and van den Bent 2018 (a study of ABT414; 73% of control participants received LOM, the others received TMZ), FOT ranked first, BEV + LOM ranked second, and LOM ranked third, BEV+IRI ranked fourth, and BEV ranked fifth. Ranking does not take into account the certainty of the evidence, which indicated that there was little or no difference between BEV + LOM and LOM, probably little or no difference between FOT and LOM, and probably little or no difference between BEV + IRI and BEV. ABT414, CED and ONA were not associated with clinical benefits. See summary of findings Table 1.

Other studies conducted among patients with first recurrence that could not be included in the NMA due to insufficient data did not report encouraging results and were considered not to warrant further investigation in the context of recurrent GBM. We found no studies assessing TMZ re‐challenge (without ABT414) in this context.

Progression‐free survival (first recurrence)

Median PFS across all RCTs reporting this outcome ranged from 1.5 months (LOM arm of Wick 2014) to 4.2 months (BEV + LOM arm of Wick 2017).

The NMA findings

Seven RCTs involving the following eight treatments contributed data to this NMA (Figure 6).

Figure 6
Network for progression free survival (first recurrence)

Network for progression free survival (first recurrence)

Effect estimates for this network can be found in the forest plot (Figure 7) and also in the league table showing HRs and 95% CI estimates for all intervention comparisons (Table 3). The global test for inconsistency was not statistically significant (P = 0.80). Again, we found no high‐certainty evidence that any of the treatments evaluated were superior to lomustine monotherapy. Graded pooled network estimates for progression‐free survival of treatment compared with lomustine monotherapy suggest the following.

Figure 7
Forest plot of  PFS for different treatments compared with lomustine (first recurrence)

Forest plot of  PFS for different treatments compared with lomustine (first recurrence)

Open in table viewer
Table 3. Progression‐free survival: League table showing HR and 95% CI estimates for all intervention comparisons

LOM

BEV

BEV_IRI

BEV_LOM

BEV_ONA

REG

LOM

0.90 (0.58,1.38)

0.80 (0.44,1.45)

0.57 (0.44,0.74)

0.98 (0.51,1.87)

0.65 (0.42,1.01)

1.12 (0.72,1.72)

BEV

0.90 (0.60,1.34)

0.64 (0.41,0.99)

1.09 (0.67,1.77)

0.73 (0.39,1.35)

1.25 (0.69,2.25)

1.12 (0.75,1.67)

BEV_IRI

0.71 (0.39,1.28)

1.22 (0.65,2.28)

0.81 (0.39,1.69)

1.75 (1.36,2.26)

1.57 (1.02,2.43)

1.41 (0.78,2.55)

BEV_LOM

1.71 (0.89,3.29)

1.14 (0.68,1.90)

1.02 (0.53,1.96)

0.92 (0.57,1.49)

0.82 (0.44,1.54)

0.58 (0.30,1.12)

BEV_ONA

0.67 (0.30,1.46)

1.54 (0.99,2.40)

1.38 (0.74,2.57)

1.24 (0.59,2.59)

0.88 (0.53,1.46)

1.50 (0.68,3.30)

REG

BEV = bevacizumab
CED = cediranib
IRI = irinotecan
LOM = lomustine
ONA = onartuzumab
REG = regorafenib

  • BEV + LOM may be more effective than LOM only (HR 0.57, 95% CI 0.44 to 0.74; low‐certainty evidence);

  • There may be little or no difference between BEV and LOM (HR 0.90, 95% CI 0.58 to 1.38; low‐certainty evidence);

  • There is probably little or no difference between CED + LOM and LOM (HR 0.76, 95% CI 0.50 to 1.18; moderate‐certainty evidence);

  • There is probably little or no difference between CED and LOM (HR 1.05; 95% CI 0.68 to 1.62; moderate‐certainty evidence);

  • Evidence on BEV + ONA versus LOM (HR 0.98, 95% CI 0.51 to 1.87) and BEV + IRI versus LOM (HR 0.80, 95% CI 0.44 to 1.45) and REG versus LOM (HR 0.65, 95% CI 0.42 to 1.01) was very low certainty.

When treatments other than LOM were compared with BEV monotherapy there were no clear differences, with the exception of BEV + LOM, the evidence for which suggested that BEV + LOM may be more effective than BEV monotherapy (0.64, 95% CI 0.41 to 0.99; high‐certainty evidence). For BEV + IRI versus BEV, the evidence suggested that there may be little or no difference (0.90, 95% CI 0.60 to 1.34; low‐certainty evidence).

In terms of ranking, BEV + LOM ranked first, REG ranked second, BEV + IRI ranked third, BEV ranked fourth, LOM ranked fifth and BEV + ONA ranked last. See summary of findings Table 2. Ranking does not take into account the certainty of the evidence above.

Other study findings evaluating interventions at first recurrence

The studies of novel agents imatinib (Dresemann 2010), axitinib (Duerinck 2018), personalised peptide vaccination (PPV; Narita 2019), nivolumab with or without ipilimumab (Omuro 2018), pembrolizumab (Reardon 2018b), enzastaurin (Wick 2010), and afatinib (Reardon 2015b) could not be included in the NMA, either due to insufficient data (no HRs reported) or due to no common nodes; however they did not show clinically meaningful survival benefits. Similarly, we could not include studies of HSPPC‐96 vaccine (Bloch 2017), VB111 (Cloughesy 2018), carboplatin (Field 2015), TRC105 (Galanis 2017), desatinib (Galanis 2019), vorinostat (Puduvalli 2018), or metronomic etoposide or TMZ (Reardon 2011) added to BEV in the NMA due to insufficient data or no common nodes; however, we noted no survival benefits with these combinations and published findings suggest that they do not warrant further investigation in the context of recurrent GBM.

One randomised study evaluated the novel intervention cintredekin besudotox compared with gliadel wafers in patients with a first recurrence and no survival differences (median OS ~ 9 months) (Kunwar 2010); however, the risk of pulmonary embolism was increased with cintredekin besudotox (P = 0.014).

A small pilot study evaluated a cannabidiol:delta‐9‐tetrahydrocannabinol (CBD:THC) oro‐mucosal spray among 12 people with recurrent GBM randomised to the intervention and 9 randomised to placebo (Twelves 2017). All participants also received dose‐intense TMZ. The median survival in the CBD:THC group was better than the placebo group (~ 18.3 months vs ~ 12.3 months, respectively). One‐year survival was 83% and 56% in the CBD:THC and placebo groups, respectively.

Re‐operation and re‐irradiation

Three non‐randomised studies evaluated re‐operation among people with first recurrence (Table 1) (Azoulay 2017; Scorsetti 2015; Suchorska 2016). These non‐randomised studies are at a high risk of selection bias and this evidence was not graded. Azoulay 2017 retrospectively compared re‐operation (with or without salvage chemoradiotherapy) with salvage chemoradiotherapy or best supportive care at first recurrence (median time from diagnosis 7.43 months). Sixty‐nine people had repeat surgery and 111 did not: the decision on treatment was made by a multi‐disciplinary team and was based on prognostic factors such as tumour extent and location. The median survival after repeat surgery was 9.8 months compared to 5.0 months for those receiving other treatment (P < 0.0001) (study authors acknowledge a "lack of consistent selection criteria for each treatment modality" as a study limitation).

Scorsetti 2015 retrospectively evaluated overall survival and progression‐free survival in a retrospective study including 21 people receiving re‐resection and/or re‐irradiation plus chemotherapy (combined treatment) and 22 receiving chemotherapy alone. People selected for the different treatment groups had different clinical characteristics at the time of relapse. The median interval from initial diagnosis with glioblastoma was 13 months (6 to 78 months). Median overall and progression‐free survival in the combined treatment group were reported to be 17 and 15 months versus 6 and 5 months in the group receiving chemotherapy alone.

Suchorska 2016 prospectively evaluated re‐operation versus no re‐operation in an exploratory sub‐study of the DIRECTOR trial, which compared different dose‐intense TMZ regimens among 105 people with recurrent GBM. Seventy‐one participants in the cohort underwent re‐operation. There was no significant difference in PFS (2.0 months vs 1.9 months, respectively) or post‐recurrence survival (11.4 months versus 9.8 months, respectively) between those who had surgery and those who did not. However, complete resection was associated with better survival than incomplete resection (9.8 months versus 6.5 months, respectively).

In a retrospective study involving 144 patients at first progression of GBM, Kim 2015 and colleagues evaluated five different treatment options: Gamma Knife (stereotactic) radiosurgery (GKS) (n = 29); temozolomide: either 50 mg/m² daily (metronomic dose) or 150 to 200 mg/m² for 5 days per 4 weeks (n = 31); Gamma Knife radiosurgery + temozolomide: 67.9% received metronomic TMZ chemotherapy (n = 28); re‐operation (n=38); or ‘other treatment’: (n = 18). We have set out results for each arm in the Characteristics of included studies table. The authors concluded that GKS with TMZ was associated with improved overall survival. However, as with the other retrospective studies, it was not clear how patients were selected for the different treatment options (in this study average tumour volume differed across treatment arms).

Evidence on survival outcomes for treatment of second and/or subsequent recurrence

As described above, we could not perform NMAs of second and subsequent recurrence due to insufficient data. Twenty studies evaluated different interventions in mixed populations with first, second and/or subsequent recurrences. Ten of these studies were phase 2 studies that did not show meaningful clinical benefits. Wick 2014 was a phase 2 study of the novel intervention APG101, which the investigators considered to have potential for further development, but only 29% of these participants had second or subsequent recurrence, the majority had first recurrence. Similarly, a phase 2 study of BEV + dose dense TMZ versus BEV + irinotecan reported that both treatment arms passed pre‐specified efficacy thresholds (Gilbert 2017); the proportion of second and subsequent recurrences in this study was unclear.

A few studies have evaluated mainly second or later recurrences. These included Cuncannon 2019, Galanis 2017 (terminated early for futility), Heiland 2016, Modh 2018 and Stupp 2012. Tsien 2019 is also discussed below, although the proportion of second and later recurrences in this study is unclear. Four studies evaluated re‐irradiation but with diversity of line of treatment, fractionation and accompanying systemic therapy. Three studies evaluated BEV with or without radiotherapy.

  • Cuncannon 2019 prospectively evaluated BEV compared with supportive care for chemo‐refractory disease following treatment of relapsed GBM among 48 patients. BEV was offered to the 48 patients at a maximum cost of EUR 12,000; 15 refused for financial reasons and 28 accepted. Most patients were experiencing a second or third relapse and the median survival of patients accepting BEV was 6 months versus 1 month with supportive care only (P < 0.01). Patients in the BEV arm (n = 16) were more likely to receive radiotherapy (35 to 40 Gy in 15 fractions over 3 weeks) than those in the supportive care only arm (n = 0), which authors suggested may have been facilitated by BEV. These findings are at high risk of bias.

  • Modh 2018 compared BEV + fractionated stereotactic radiotherapy (8 Gy × 4 fractions over 2 weeks) with BEV + chemotherapy in an RCT involving 34 heavily pre‐treated participants (median recurrence was 3). The BEV + radiotherapy arm experienced longer progression‐free survival (5.3 months vs 1.8 months) and better local control. Overall survival was 7.1 months vs 4.8 months, respectively. This was reported in a conference abstract only and details were sparse.

  • Tsien 2019, also reported as a conference abstract, was a phase 2 RCT of 170 participants comparing hypofractionated radiotherapy (35 Gy in 10 fractions) plus BEV versus BEV only. The proportion of participants with second or subsequent recurrence among this study sample is not clear as findings are available in a conference abstract only; however, the duration of overall survival (~ 10 months) suggests that the majority of participants had a first recurrence. Investigators reported no significant difference in overall survival; however, significantly more participants were progression‐free at 6 months in the BEV + radiotherapy arm than the BEV arm (54% vs 29%). Authors conclude that the "role of BEV + RT should be limited to small volume recurrences, especially in previously non‐irradiated treatment areas at least 6 months following completion of previous RT." Evidence from these studies is difficult to interpret but suggest that BEV + radiotherapy may have a role in delaying disease progression in second and subsequent recurrence of GBM.

Twenty‐nine per cent of participants in Wick 2014 were experiencing second or third recurrences. This phase 2 RCT evaluated radiotherapy (36 Gy, 2 Gy fractions × 5 per week) plus APG101 (a CD95 inhibitor) compared with radiotherapy alone. Median PFS was 2.5 months (95% CI 2.30 to 3.80) months for radiotherapy and 4.5 (95% CI 3.70 to 5.40) months for radiotherapy + APG101 with a hazard ratio (HR) of 0.49 (95% CI 0.27 to 0.88; P = 0.0162) adjusted for tumour size in favour of radiotherapy plus APG101, with no clear difference in overall survival. Authors reported that the novel agent APG101 warrants further clinical development.

Heiland 2016 retrospectively evaluated "last‐line" therapy in 43 patients at third recurrence. In this study, BEV monotherapy (n = 17) was compared with combined BEV + LOM therapy (n = 18). It was not clear how patients were selected for the two treatment options. Median overall survival after BEV monotherapy was 4.07 months (95% CI 3.02 to 12.98) while in the combined therapy group median overall survival was 6.59 months (95% CI 5.51 to 16.30). Median progression‐free survival was 2.3 months (95% CI 1.87 to 4.39 months) compared with 6.11 months (95% CI 3.41 to 12.98 months) in the combined BEV + LOM group. We considered this study to be at a high risk of bias.

Stupp 2012 was a phase 3 RCT in which more than 80% of 237 participants had failed two or more prior lines of chemotherapy (second recurrence) and 20% of the patients had failed bevacizumab prior to enrolment. Participants were randomised to receive tumour‐treating fields (TTF) or physicians best choice of treatment — most received single agent or a combination chemotherapy regimen containing bevacizumab (31%), or irinotecan (31%), followed by nitrosoureas (25%), carboplatin (13%), temozolomide (11%) or various other agents (5%). Interpretation of findings is difficult because the survival effects of TTF were similar to the control but it is unclear how effective the control arm treatments are, if at all. 

Quality of life

Seven studies reported findings on health‐related quality of life (HRQoL) using the European Organisation for Research and Treatment of Cancer (EORTC) core questionnaire (QLQ‐C30) (Brandes 2016a; Brown 2016; Field 2015; Galanis 2017; Stupp 2012; Suchorska 2016; Taal 2014). All but one of these studies (Taal 2014) also used the EORTC questionnaire relating to brain cancer (BN‐20). In addition, Galanis 2017 used the shorter EORTC QLQ‐C15‐PAL along with the BN‐20 questionnaire.

A multi‐centre trial in the Netherlands (Taal 2014) recruited patients at first recurrence and included three arms: bevacizumab alone (n = 46), lomustine alone (n=45), or combined treatment (n = 47). At baseline, QoL scores were similar in the three groups and, compared with the general population, these patients had impaired scores. Following treatment there were no clear differences between arms for any of the five sub‐scales assessed.

Brandes 2016a also examined bevacizumab for patients with recurrent disease. In this study participants were randomised in a 2:1 ratio to receive bevacizumab (n = 59) or fotemustine (n = 32). QoL was assessed at approximately eight weeks after study drug administration or at disease progression. Follow‐up questionnaires were only completed by 15 (13.56%) in the bevacizumab group and eight (25%) in the fotemustine group. Authors reported an improvement in physical functioning from baseline; although there was no significant difference between groups and there appeared to be variability within groups (SDs were large). For other QLQ C‐30 dimensions the authors reported deteriorations in scores for fatigue, nausea, insomnia and appetite loss for patients in the fotemustine group and in emotional functioning in the bevacizumab group. However, scores for these items were not reported for both groups, and it was not clear whether or not there were any significant differences between the two arms for any of these dimensions of QoL.

In another study (Stupp 2012), patients with recurrent glioblastoma were randomised to TTF (n = 120) versus chemotherapy (“best available” according to physician choice) (n = 117). At three months, post‐treatment QoL scores were available for 27% (n = 63) of patients randomised. Results were set out in graphs and authors report “no meaningful differences” between arms for global health and social functioning. For other QoL dimensions, symptoms appeared to be related to treatment‐associated toxicity in the chemotherapy arm (loss of appetite, diarrhoea, constipation, nausea and vomiting). Authors also reported increased pain and fatigue in the chemotherapy arm. It was not clear whether apparent differences between groups for these symptoms were statistically significant.

A multi‐centre RCT comparing cediranib plus gefitinib with cediranib plus placebo including patients at first progression was terminated early after recruitment of 38 patients (19 in each arm) (Brown 2016). Twenty‐six patients completed questionnaires at six weeks and there were no clear differences between arms for global health status or for any of the sub‐scales. The authors concluded that there was no evidence that the addition of gefitinib resulted in poorer QoL; but the study was most likely underpowered to detect possible differences between groups.

A trial including 122 patients with recurrent GMB compared bevacizumab alone with bevacizumab plus carboplatin (Field 2015). Authors reported change from baseline and there was no significant differences in overall scores detected between groups.

The study by Galanis 2017 compared bevacizumab plus TRC105 with bevacizumab alone. Of 101 patients recruited 65 were included in the main QoL analysis. In terms of overall scores on the EORTC QLQ‐C15‐PAL questionnaire, there was no clear difference between groups (P = 0.19). For the BN20 items, there were no significant differences between groups for any of the dimensions. At four weeks patients were asked whether they thought it had been worthwhile participating in the study and similar proportions in both arms said yes (BEV plus TRC105, 69.4% (25/36), BEV alone 71.9% (23/32).

Suchorska 2016, a non‐randomised study, evaluated QoL among 71 people who underwent re‐operation and 34 who did not at the 8‐week follow‐up visit as part of the DIRECTOR trial. Surgery was associated with better cognitive functioning (P = 0.46). Constipation occurred more commonly in this group (P = 0.039). Complete resection was associated with better global health status compared with incomplete resection (P = 0.008).

Finally, Lombardi 2019 randomised patients with relapsed GMB and compared regorafenib with lomustine. One hundred and fourteen patients completed baseline QoL questionnaires but only 37 were available for follow‐up (24 in the regorafenib group and 13 in the lomustine group). There were no significant differences on any dimensions of either the general or brain tumour questionnaires, other than for appetite loss which was worse in those patients treated with regorafenib: 9 out of 24 receiving regorafenib and none of 13 receiving lomustine had what was described as clinically meaningful worsening appetite (P = 0.0146).

Severe adverse events

Two disconnected networks were constructed from the available data, one around lomustine and the other around bevacizumab.

Network 1: treatments versus lomustine

Five RCTs contributed to the lomustine‐based network involving six different treatments, including:

The network diagram is presented in Figure 8, the league table showing HRs and 95% CI estimates for all intervention comparisons in Table 4 and the forest plot in Figure 9. We interpreted the evidence as follows.

Figure 8
Severe adverse events ‐ network 1

Severe adverse events ‐ network 1

Open in table viewer
Table 4. Severe adverse events for treatments compared with lomustine: League table with effect estimates and 95% CIs

LOM

BEVLOM

CED

CEDGET

CEDLOM

REG

LOM

2.51 (1.72,3.66)

1.00 (0.54,1.85)

2.46 (0.46,13.26)

2.51 (1.29,4.90)

1.90 (0.92,3.95)

0.40 (0.27,0.58)

BEVLOM

0.40 (0.19,0.82)

0.98 (0.17,5.50)

1.00 (0.46,2.15)

0.76 (0.33,1.72)

1.00 (0.54,1.85)

2.51 (1.22,5.17)

CED

2.46 (0.51,11.80)

2.51 (1.43,4.42)

1.90 (0.73,4.94)

0.41 (0.08,2.19)

1.02 (0.18,5.73)

0.41 (0.08,1.95)

CEDGET

1.02 (0.19,5.40)

0.77 (0.12,4.84)

0.40 (0.20,0.78)

1.00 (0.46,2.15)

0.40 (0.23,0.70)

0.98 (0.19,5.18)

CEDLOM

0.76 (0.28,2.03)

0.53 (0.25,1.09)

1.32 (0.58,3.00)

0.53 (0.20,1.36)

1.29 (0.21,8.10)

1.32 (0.49,3.54)

REG

BEV = bevacizumab
CED = cediranib
GET = getitinib
LOM = lomustine
REG = regorafenib

Figure 9
Severe adverse events forest plot for network 1 (treatments vs lomustine)

Severe adverse events forest plot for network 1 (treatments vs lomustine)

  • BEV + LOM is associated with significantly more severe adverse events than LOM (RR 2.51, 95% CI 1.72 to 3.66; high‐certainty evidence)

  • There may be little difference in the risk of severe adverse events between REG and LOM, but the point estimate favours LOM (RR 1.90, 95% CI 0.92 to 3.95; low‐certainty evidence)

  • There may be little difference in risk of severe adverse events between CED and LOM (RR 1.00, 95% CI 0.54 to 1.85; low‐certainty evidence)

  • CED + LOM is associated with significantly more severe adverse events than LOM (RR 2.51, 95% CI 1.29 to 4.90; high‐certainty evidence)

  • The evidence on CED + GET versus LOM is very low certainty.

In terms of ranking, lomustine and cediranib single therapies ranked best with the fewest severe adverse events. REG ranked second although the point estimate of REG suggesting more severe adverse events than lomustine was almost statistically significant. Bevacizumab plus lomustine ranked joint worst with CED + LOM (Figure 9). Ranking does not take into account the certainty of the evidence. See summary of findings Table 3.

Network 2: treatments versus bevacizumab

Eight RCTs contributed data to this network involving nine different treatments, including:

The network diagram is presented in Figure 10, the league table showing HRs and 95% CI estimates for all intervention comparisons in Table 5 and the forest plot in Figure 11. The network comprised mainly novel treatments added to bevacizumab compared with bevacizumab. As expected, pooled network estimates suggested that, compared with bevacizumab, adding treatments to bevacizumab was associated with a higher frequency of severe adverse events. Fotemustine was also compared with bevacizumab and there was no clear difference summary of findings Table 4.

Figure 10
Severe adverse events ‐ network 2

Severe adverse events ‐ network 2

Open in table viewer
Table 5. Severe adverse events for treatments compared with bevacizumab: League table with effect estimates and 95% CIs

BEV

BEVCAB

BEVDAS

BEVIRI

BEVONA

BEVTRC105

BEVVB111

FOM

HSPPCBEV

BEV

1.27 (0.61,2.66)

1.52 (0.69,3.34)

2.22 (1.18,4.18)

1.17 (0.57,2.39)

6.86 (2.55,18.41)

3.77 (2.25,6.33)

0.44 (0.11,1.72)

1.01 (0.33,3.10)

0.79 (0.38,1.64)

BEVCAB

1.19 (0.41,3.51)

1.75 (0.66,4.61)

0.92 (0.33,2.57)

5.39 (1.57,18.47)

2.97 (1.21,7.29)

0.35 (0.07,1.63)

0.79 (0.21,3.03)

0.66 (0.30,1.45)

0.84 (0.29,2.46)

BEVDAS

1.46 (0.53,4.02)

0.77 (0.27,2.23)

4.51 (1.28,15.96)

2.48 (0.97,6.37)

0.29 (0.06,1.40)

0.66 (0.17,2.62)

0.45 (0.24,0.85)

0.57 (0.22,1.51)

0.68 (0.25,1.88)

BEVIRI

0.53 (0.20,1.37)

3.09 (0.95,9.97)

1.70 (0.75,3.84)

0.20 (0.04,0.89)

0.45 (0.13,1.65)

0.85 (0.42,1.75)

1.09 (0.39,3.03)

1.30 (0.45,3.76)

1.90 (0.73,4.93)

BEVONA

5.86 (1.73,19.82)

3.22 (1.33,7.79)

0.38 (0.08,1.75)

0.86 (0.23,3.26)

0.15 (0.05,0.39)

0.19 (0.05,0.64)

0.22 (0.06,0.78)

0.32 (0.10,1.05)

0.17 (0.05,0.58)

BEVTRC105

0.55 (0.18,1.68)

0.06 (0.01,0.35)

0.15 (0.03,0.66)

0.26 (0.16,0.44)

0.34 (0.14,0.83)

0.40 (0.16,1.03)

0.59 (0.26,1.33)

0.31 (0.13,0.75)

1.82 (0.60,5.54)

BEVVB111

0.12 (0.03,0.50)

0.27 (0.08,0.92)

2.26 (0.58,8.80)

2.88 (0.61,13.49)

3.44 (0.72,16.52)

5.03 (1.12,22.49)

2.65 (0.57,12.28)

15.51 (2.89,83.17)

8.54 (2.00,36.51)

FOM

2.28 (0.39,13.29)

0.99 (0.32,3.04)

1.26 (0.33,4.82)

1.51 (0.38,5.93)

2.20 (0.61,7.98)

1.16 (0.31,4.39)

6.80 (1.52,30.30)

3.74 (1.09,12.86)

0.44 (0.08,2.55)

HSPPCBEV

BEV = bevacizumab
CAB = carboplatin
DAS = desatinib
IRI = irinotecan
FOM = fotemustine
HSPCC = HSPCC‐96 vaccine
ONA = onartuzumab
TRC105 = carotuximab

Figure 11
Severe adverse events forest plot for network 2 (treatments vs BEV)

Severe adverse events forest plot for network 2 (treatments vs BEV)

In terms of ranking, fotemustine ranked best with fewest severe adverse events, BEV ranked second, BEV + HSPPV96 vaccine ranked third, BEV + ONA ranked fourth, BEV + CAB ranked fifth, BEV + DAS ranked sixth, BEV + IRI ranked seventh, BEV + VB111 ranked eighth and BEV + TRC105 ranked worst.

Brief economic commentary

For this brief economic commentary, we summarise the results of identified studies based upon what the study authors have said. These studies have not been critically appraised, and the studies may have used methods that are not consistent with accepted practice. For this reason and because the studies are conducted at different times and in different places, we do not attempt to draw any firm or general conclusions regarding the relative costs or efficiency of the different strategies to manage recurrent glioma.

The results of the economic search yielded four economic evaluations that compare the costs and benefits of the management of recurrent glioma. Three of the studies were reported to be cost‐effectiveness analyses (Conen 2017; Ruiz‐Sanchez 2016; Voigt 2016). The other study was reported to be a cost‐effectiveness analysis and cost‐benefit analysis (Roussakow 2017). The studies were all conducted in different countries. One study was conducted in Switzerland (Conen 2017); one in Spain (Ruiz‐Sanchez 2016); one in the USA (Voigt 2016); and one in Germany (Roussakow 2017). Two of the studies assessed the use of bevacizumab (Conen 2017; Ruiz‐Sanchez 2016); one study assessed modulated electrohyperthermia concurrent to dose‐dense temozolomide (Roussakow 2017); and one assessed the value of laser interstitial thermal therapy (LITT) where maximal safe resection may not be feasible (Voigt 2016).

Conen 2017 is a study (reported in a conference abstract) that retrospectively used data from a GBM database over a five‐year period to assess the cost‐effectiveness of the use of bevacizumab in recurrent GBM. The study used a sample of 82 newly diagnosed GBM patients, of which 75 had a first line treatment, 36 had a second line treatment and 14 had a third line therapy. Forty per cent of patients were treated with bevacizumab at first or second recurrence. The authors conclude that bevacizumab treatment increased the overall treatment costs by 1.7 times. The population‐adjusted incremental cost‐effectiveness ratio (ICER) was CHF (Swiss francs) 75,669 per life‐year gained (the price year was not stated). The authors conclude that patients who received bevacizumab treatment for GMB recurrences had longer overall survival and longer quality‐adjusted survival at costs below the accepted threshold of CHF 100,000 per life year gained. The authors state that whether this estimated increase in lifespan is a direct result of bevacizumab treatment or a consequence of a selection bias needs to be addressed prospectively.

Ruiz‐Sanchez 2016 carried out a cost‐effectiveness analysis on the bevacizumab‐irinotecan regimen in GBM recurrences in a retrospective cohort study design with a control group. The intervention group included patients diagnosed with primary GBM between January 2001 and December 2011 in the Principality of Asturias (Spain) and treated in the Central University Hospital of Asturias (HUCA; Oviedo, Spain). The control cohort included all patients treated with TMZ between January 2001 and December 2006. There were 151 patients in the non‐bevacizumab control cohort and 52 in the bevacizumab‐irinotecan cohort. Costs, valued in 2014 EUR, were derived from the study data (including cost of the antineoplastic drug treatment, cost of the antineoplastics, administration and monitoring of administration and premedication). In the cohort with the regimen that included bevacizumab‐irinotecan, the final cost for the 36 patients treated stood at EUR 629,278. The cost in the control cohort was EUR 16,771. In this way, increasing survival by 4.4 months for 36 patients came to EUR 612,506, meaning an additional cost of EUR 46,402 per person for each year of life gained. The authors conclude that bevacizumab‐irinotecan is an effective therapy but it is not cost‑effective. As such, they do not recommend adoptions by their specific local public health system.

The analysis by Voigt 2016 carried out a model‐based analysis that did not primarily focus on recurrent glioma, but did include a sensitivity analysis focusing on the effects of recurrence, and the subsequent impact of that on patient and cost outcomes. A decision tree was developed to evaluate the cost‐effectiveness of using brain laser interstitial thermal therapy (LITT) versus current treatments. The model was purported to adopt the societal perspective; the included costs and benefits are primarily healthcare based, however, and do not include any wider costs. The decision tree evaluated the initial procedure and the resultant outcome (i.e. gross total resection, subtotal resection) using probabilities as identified in the literature. Costs are presented in 2015 US dollars (USD). Patients were followed through the treatment decision tree until they died. The incremental cost per life year gained was USD 48,552 when compared to biopsy. This authors conclude this is higher than is acceptable using an “International Threshold” of USD 32,575/LYG (which is based on Spanish systematic review of Barrios 2012) but acceptable from US threshold of USD 50,000/LYG. A one‐way sensitivity analysis was carried out to assess how GBM recurrence affects this incremental cost‐effectiveness. The authors report that the higher the occurrence of local GBM recurrence (vs. diffuse recurrence), the more likely brain LITT was to be cost‐effective. This was based on willingness to pay (WTP) of USD 2714 per month based on the “International Threshold”.

Roussakow 2017 assessed the efficacy and cost‐effectiveness of modulated electrohyperthermia (mEHT) concurrent to dose‐dense temozolomide (ddTMZ) 21/28 days regimen to five pooled ddTMZ 21/28 days cohorts (114 patients) enrolled between 2008 and 2013. A retrospective clinical and economic evaluation was based on the comparison and effect‐to‐treatment analysis (ETA) of a retrospective, single‐arm study performed in two German centres between 2000 and 2005. The results of the regression analysis show the ddTMZ+mEHT cohort did not significantly improve mean survival time (mST) against the pooled ddTMZ 21/28 days cohorts. Using the effect‐to‐treatment analysis (ETA) suggests that mEHT significantly enhances the efficacy of the ddTMZ 21/28 days regimen, with significantly less toxicity and an estimated maximal attainable median survival time of 10.10 months. The author reported carrying out a cost‐effectiveness analysis with results presented as a cost‐utility ratio. They also reported carrying out a cost‐benefit analysis. Costs were expressed in 2017 US dollars and Euros. Two cost models were used for the cost‐effectiveness analysis: conditionally termed ‘German’ and ‘US’. The first, so‐called German option, is specific for a high‐income country with rigid governmental regulation of the medical market, which leads to relatively low prices for pharmaceuticals with low variance. The second, so‐called US option, is specific for a high‐income country with lower governmental regulation, which leads to relatively high prices for pharmaceuticals with higher variance. The cost‒utility ratio presented found that the ddTMZ+mEHT regimen, both in the German (EUR 19,871 per QALY (95% CI 17,719 to 22,024) and the US (USD 32,704/QALY (95% CI 27,215 to 38,193) models were less than that of the comparator. The sources of the utility values and how these QALY values are derived are not reported. The purported cost‐effectiveness analysis and cost‐benefit analysis only consider costs, not natural units or a monetary valuation of benefits and, thus, are in fact cost analysis, meaning that these elements of the study are only partial economic evaluations.

In summary, the economic evidence identified in this review found conflicting evidence on the use of bevacizumab in recurrent GBM. Of the two studies which evaluated its use, Conen 2017 reported it to be a cost‐effective intervention, whereas Ruiz‐Sanchez 2016 did not. Voigt 2016 estimated the cost effectiveness of the use of brain LITT versus current treatments and concluded that it improves survival at a cost which appears to be of good value to society according to US thresholds for good value. Roussakow 2017 reported evidence on the cost‐effectiveness of modulated electrohyperthermia (mEHT) concurrent to dose‐dense temozolomide (ddTMZ) and concluded that ddTMZ+mEHT is cost‐effective, budget‐saving and profitable, although methods used are not consistent with definitions of the different types of economic evaluation. Economic studies of most treatments evaluated in this review are lacking.

Discussion

Summary of main results

We included 34 RCTs and 8 non‐RCTs involving 5236 patients with progressive/recurrent GBM; 20 studies involved patients with first recurrences and the remainder involved patients with subsequent recurrence or the study sample comprised patients with mixed recurrences (e.g. first, second and/or third recurrences). Several studies lacked suitable data and we could not connect them in the network meta‐analyses (NMA), hence we did not grade evidence from these single studies and non‐randomised studies. We judged most RCTs to be at a low risk of bias and NRSs at high risk of bias. Most interventions were evaluated in single studies and included trials of systemic chemotherapy agents, re‐operation, re‐irradiation, anti‐angiogenic agents, antibody therapies, tumour‐treating fields, and vaccines alone or in combination. For first recurrence, 11 interventions (involving 9 RCTs and 1931 participants) were connected in the network for overall survival (OS), and eight (involving 7 RCTs and 1500 participants) in the network for progression‐free survival (PFS). No studies in the NMA evaluated surgery, re‐irradiation, PCV (procarbazine, lomustine, vincristine), TMZ re‐challenge or best supportive care. We could not perform NMA for second or later recurrence due to insufficient data. Quality of life data were sparse. Only one NRS evaluated best supportive care.

First recurrence (NMA findings)

Median overall survival (OS) across included studies in the NMA ranged from 5.5 to 12.6 months and median PFS across included studies ranged from 1.5 months to 4.2 months. We found no high‐certainty evidence that any treatments tested were better than lomustine, including the following.

Bevacizumab plus lomustine

Evidence suggested that there is probably little or no difference in OS between bevacizumab (BEV) combined with lomustine (LOM) and LOM monotherapy (HR 0.91, 0.75 to 1.10; moderate‐certainty evidence).

Low‐certainty evidence suggested that BEV + LOM may improve PFS compared with LOM monotherapy (HR 0.57, 95% CI 0.44 to 0.74) and more evidence is needed.

Bevacizumab monotherapy

Low‐certainty evidence suggested there may be little or no difference in OS between BEV and LOM monotherapies (HR 1.22, 95% CI 0.84 to 1.76) and that there may be little or no difference in PFS between BEV and LOM monotherapies (HR 0.90, 95% CI 0.58 to 1.38; low‐certainty evidence); more evidence is needed.

Regorafenib (REG)

Evidence suggested that REG may improve OS compared with LOM but more evidence is needed (HR 0.50, 95% CI 0.33 to 0.76; low‐certainty evidence). Evidence on PFS was very low certainty.

Temozolomide (TMZ) plus Depatux‐M (ABT414)

With regard to OS, low‐certainty evidence suggested that TMZ plus ABT414 may be more effective than LOM (HR 0.66, 95% CI 0.47 to 0.92) and may be more effective than BEV (HR 0.54, 95% CI 0.33 to 0.89; low‐certainty evidence) but more evidence is needed.

Fotemustine (FOM)

Evidence suggests that FOM and LOM may have similar effects on OS (HR 0.89, 95% CI 0.51 to 1.57, low‐certainty evidence).

Bevacizumab and irinotecan (IRI)

Evidence on BEV + irinotecan (IRI) versus LOM for both OS and PFS is very uncertain.

Evidence on BEV + IRI versus BEV monotherapy suggested that there is probably little or no difference between these options for OS (HR 0.95, 95% CI 0.70 to 1.30; moderate‐certainty evidence) and PFS (HR 0.95, 95% CI 0.70 to 1.30; moderate‐certainty evidence).

Treatments were ranked according to effectiveness on OS as follows: FOM ranked first, BEV + LOM ranked second, LOM ranked third, BEV + IRI ranked fourth and BEV ranked fifth. It is important to note that ranking does not take into account the certainty of the evidence, which indicated that there was little or no difference between BEV + LOM and LOM, probably little or no difference between FOT and LOM, and probably little or no difference between BEV + IRI and BEV.

Other interventions for first recurrence

Re‐operation with or without re‐irradiation and chemotherapy

Three non‐randomised studies evaluated re‐operation versus no re‐operation with or without re‐irradiation and chemotherapy and these suggested possible survival advantages with re‐operation within the context of being able to select suitable candidates for re‐operation (Azoulay 2017; Scorsetti 2015; Suchorska 2016).

Novel agents

Findings of a small pilot study that evaluated a cannabidiol:delta‐9‐tetrahydrocannabinol (CBD:THC) oro‐mucosal spray compared with placebo suggested that survival may be improved with CBD:THC among people receiving dose‐intense TMZ (Twelves 2017).

Second or later recurrence

Data on second and subsequent recurrence was sparse.

Bevacizumab monotherapy

One non‐randomised study compared BEV with best supportive care for people with chemorefractory disease (second and third recurrence) and showed a survival advantage with BEV; the study was at a high risk of bias, however, as participants selected their treatments and many in the BEV arm also received radiotherapy.

Radiotherapy plus bevacizumab

Limited evidence from three heterogeneous studies (different control groups and populations with different levels of GBM recurrences) reported findings of improved PFS or OS with this intervention for all or for selected candidates. More evidence on radiotherapy with and without BEV is needed.

Tumour‐treating fields

Evidence from one RCT suggested that there may be little difference in effects on PFS or OS of tumour‐treating fields (TTF) compared with physician's choice of treatment in this context.

Severe adverse events (SAEs)

Two distinct networks were constructed around LOM (5 RCTs, 6 interventions, 1024 participants) and BEV (8 RCTs, 9 interventions, 1044 participants). In the LOM network, LOM ranked best and REG second best. BEV + LOM were associated with significantly greater risk of SAEs than LOM monotherapy (RR 2.51, 95% CI 1.72 to 3.66, high‐certainty evidence), and ranked joint worst with CED + LOM.

The BEV network comprised mainly novel treatments added to BEV compared with BEV. In general, adding treatments to bevacizumab was associated with a higher frequency of SAEs compared with BEV monotherapy. FOM ranked best, BEV ranked third and BEV + IRI ranked seventh. Other ranked treatments were clinically ineffective. The SAE network connections did not facilitate ranking lomustine and bevacizumab against each other.

Quality of life (QoL)

Quality of life data reported in seven studies of different interventions were sparse and unreliable, mainly due to high drop‐out rates.

Overall completeness and applicability of evidence

The 2015 JLA priority question was "In second recurrence glioblastoma, what is the effect of further treatment on survival and quality of life, compared with best supportive care?" (JLA 2015). We found little good‐quality evidence that addressed this question as the one study that compared bevacizumab with best supportive care was at a high risk of bias (Cuncannon 2019). Evidence on other interventions, such as radiotherapy, systemic anti‐cancer agents and best supportive care, was lacking. We found only one small study of cannabinoids, which have shown promising anti‐cancer functions in GBM (Dumitru 2018), and the single study evaluating a ketogenic diet was at high risk of bias (Santos 2018); these types of interventions are of interest and high‐quality RCTs on their effects in recurrent GBM are needed (Martin‐McGill 2018).

Other evidence was very incomplete and the best treatment options compared with each other, even for first recurrence, remain uncertain. We found no RCTs comparing the commonly used regimen of PCV (procarbazine, lomustine and vincristine) and lomustine, or TMZ re‐challenge. One trial that employed TMZ re‐challenge used it in combination with a novel agent (Depatux‐M/ABT414) and, although patients in this arm experienced improved survival relative to the control group who received LOM (73% of control participants) or TMZ (27% of control participants), it was not possible to determine whether these effects were due to the TMZ or ABT414 component of the treatment, or both (van den Bent 2018).

With respect to people with second or later recurrence, among included studies the following interventions have shown potential (see Table 1) and, although it was not possible to conduct a NMA for second recurrence, they appear to warrant further investigation, bearing in mind that combination treatments are frequently associated with a higher risk of severe adverse events: radiotherapy; radiotherapy + bevacizumab; radiotherapy + APG101; bevacizumab + temozolomide; and bevacizumab + irinotecan.

Quality of the evidence

Quality of the evidence on lomustine with or without bevacizumab and bevacizumab monotherapy was generally of a moderate to high quality; these interventions had the most data and trials were at low risk of bias. Quality of evidence on other interventions was often from single studies and, therefore, tended to be of a lower quality due to the sparse data and open‐label design of many of the phase 2 trials. Most included studies evaluated novel agents that were compared with or added to lomustine or bevacizumab and many novel agents tested in this patient population have not been investigated further in phase 3 trials because the early phase studies failed. A recent exception is regorafenib, which showed potential survival benefits compared with lomustine in a phase 2 trial. This evidence was of a generally low quality due to sparse data from a single, open‐label study. A phase 3 trial of this intervention among patients with first recurrence of GBM is ongoing (NCT03970447 2019b). Another intervention for which the evidence was generally graded low‐certainty is ABT414 (Depatux‐M) + TMZ, and more evidence on these novel interventions is needed.

We did not grade the evidence on second recurrence because either the novel interventions did not show a clinical benefit, e.g. TRC105 + BEV (Galanis 2017), or narrative findings were based mainly on conference abstracts at unclear risk of bias and/or non‐randomised studies that were at high risk of bias, for example radiotherapy and bevacizumab (Cuncannon 2019; Modh 2018; Tsien 2019).

Potential biases in the review process

We aimed to provide a balanced independent evaluation on this topic across a large number and wide range of interventions assessed since 2005, when the Stupp regimen became the standard of care for treatment of newly diagnosed GBM. We are mindful that the review process itself may introduce bias. We took steps to minimise the potential for such bias by ensuring that at least two members of the review team, working independently, screened titles identified by the search strategy, assessed full texts of reports for potentially eligible studies, extracted data and assessed risk of bias. Where we had any doubt, or where there was discrepancy between reviewers, we consulted the wider review team.

We acknowledge that a potential bias may have been introduced by including the TMZ/LOM arm of van den Bent 2018 in the LOM treatment node of our NMAs. This three‐arm trial evaluated ABT414 (Depatux‐M) versus ABT414 + TMZ versus TMZ or LOM. At the time of the analysis it was unclear what proportion of the participants in the TMZ/LOM arm had received LOM as limited findings were reported in conference abstracts. To investigate the effects of including this trial, we performed a sensitivity analysis by excluding this trial from analyses. This did not change the treatment rankings of LOM or the other treatments, and we concluded that it was unlikely that bias was introduced by including this trial in the NMA. Full details of this trial have since been published showing that 21 out of 77 participants received TMZ and the rest, 56 out of 77 or 73%, received LOM (van den Bent 2020).

Agreements and disagreements with other studies or reviews

A study of patients from eight consecutive phase 2 clinical trials conducted in the USA between 1986 and 1995 showed that the median progression‐free survival time for recurrent GBM was 9 weeks (95% CI 8 to 10 weeks) and the median OS time was 25 weeks (95% 21 to 28 weeks) (Wong 1999). With reference to Table 1, the review findings suggest that median survival for people with recurrent GBM in clinical trials in the post‐Stupp era may not have improved substantially. For people with a first recurrence, median PFS ranged from 1.5 to 4.2 months (6.5 to 18.3 weeks) across included study arms; and median OS ranged from 5.5 months to 12.6 months (23.9 months to 54.8 months).

Bevacizumab

With regard to bevacizumab, our findings are in agreement with other reviews, which have concluded that bevacizumab has little or no effect on overall survival in patients with recurrent GBM (Ameratunga 2018; Lombardi 2017). We found that bevacizumab improved progression‐free survival but not overall survival compared with lomustine monotherapy. Progression‐free survival is a less reliable outcome in this context compared with overall survival because pseudo‐response can occur in which contrast‐enhancing disease may seem improved or stable with bevacizumab when it is not. When bevacizumab was combined with lomustine, we also found an increased risk of severe adverse events; thus our findings suggest that the net clinical effect of bevacizumab added to lomustine may not necessarily be one of overall benefit and such combinations need to be carefully considered. We found no evidence of benefit with the bevacizumab‐irinotecan combination in recurrent GBM and agree with the Abdel‐Rahman 2015 review, that this regimen should be limited to a clinical trial setting until better evidence is available. Evidence on the effectiveness of BEV combined with re‐irradiation is uncertain and needs further study.

TMZ re‐challenge

We included two RCTs that evaluated TMZ in recurrent GBM: one in combination with ABT414 (van den Bent 2018); and the other in combination with bevacizumab (Gilbert 2017). These data were insufficient to draw conclusions on the effectiveness of TMZ re‐challenge, which may be more effective for MGMT methylated tumours compared with other options.

An excluded study, Sun 2013, was an RCT conducted in patients with recurrent GBM or anaplastic astrocytoma comparing TMZ with semustine (Me‐CCNU). Results were not reported separately for the GBM subgroup. Six‐month PFS for the mixed population was 78.9% in the TMZ group compared with 55.9% in the Me‐CCNU group (P < 0.05) and study authors concluded that TMZ was better than Me‐CCNU with mostly mild adverse events.

Weller 2015 was a TMZ re‐challenge study that compared two maintenance TMZ cycles in an RCT of 105 patients with recurrent GBM. It was not suitable for inclusion in this review because it was a dose‐finding study (both arms received TMZ; one week on, one week off, versus three weeks on, one week off). Median time to treatment failure was longer among patients with MGMT‐methylated tumours compared to those with MGMT‐unmethylated tumours (3.2 months vs 1.8 months) but not significantly different between the treatment arms (median of ~ 2 months). Investigators concluded that TMZ re‐challenge should no longer be used in patients with recurrence of MGMT‐unmethylated tumours but that it may be appropriate for those with MGMT‐unmethylated tumours at first recurrence.

Re‐operation and/or re‐irradiation

We found little evidence on the effectiveness of re‐operation and re‐irradiation. Similarly, other reviews have highlighted the need for high‐quality RCTs of re‐irradiation for recurrent GBM to be conducted (Kim 2019; Kazmi 2019). Evaluating limited evidence from 50 non‐comparative studies, Kazmi 2019 found that re‐irradiation may improve survival with relatively low toxicity. However, the best salvage radiotherapy regimen in this context remains to be defined. Kim 2019 highlights that narrow margins should be observed to limit irradiation of normal brain tissue. In addition, it has been suggested that prognostic scoring that makes use of MGMT methylation, age, tumour volume at recurrence and other predictive biomarkers could be used to facilitate selection of patients to different treatment options (Chapman 2019; Kim 2019).

Immunotherapies

Findings from this review show that none of the treatments that have been developed to harness the immune system to target cancer have had demonstrable clinical success in recurrent GBM. Arguably, bevacizumab is the only agent to date that may have an effect equivalent to established systemic chemotherapeutic agents, such as lomustine. It has been suggested that surgical debulking and localised delivery may enhance the effect of immunotherapies in recurrent GBM and more research in this area is anticipated (Brown 2018).

Flow diagram of searches for studies of effectiveness conducted on 16/12/2019

Figures and Tables -
Figure 1

Flow diagram of searches for studies of effectiveness conducted on 16/12/2019

Navigate to figure in ReviewOpen in new tab

Risk of bias of included studies

Figures and Tables -
Figure 2

Risk of bias of included studies

Navigate to figure in ReviewOpen in new tab

original image

Figures and Tables -
Figure 3
Navigate to figure in ReviewOpen in new tab

Network for Overall Survival (first recurrence)

Figures and Tables -
Figure 4

Network for Overall Survival (first recurrence)

Navigate to figure in ReviewOpen in new tab

Forest plot of effects on overall survival of different treatments compared with lomustine

Figures and Tables -
Figure 5

Forest plot of effects on overall survival of different treatments compared with lomustine

Navigate to figure in ReviewOpen in new tab

Network for progression free survival (first recurrence)

Figures and Tables -
Figure 6

Network for progression free survival (first recurrence)

Navigate to figure in ReviewOpen in new tab

Forest plot of  PFS for different treatments compared with lomustine (first recurrence)

Figures and Tables -
Figure 7

Forest plot of  PFS for different treatments compared with lomustine (first recurrence)

Navigate to figure in ReviewOpen in new tab

Severe adverse events ‐ network 1

Figures and Tables -
Figure 8

Severe adverse events ‐ network 1

Navigate to figure in ReviewOpen in new tab

Severe adverse events forest plot for network 1 (treatments vs lomustine)

Figures and Tables -
Figure 9

Severe adverse events forest plot for network 1 (treatments vs lomustine)

Navigate to figure in ReviewOpen in new tab

Severe adverse events ‐ network 2

Figures and Tables -
Figure 10

Severe adverse events ‐ network 2

Navigate to figure in ReviewOpen in new tab

Severe adverse events forest plot for network 2 (treatments vs BEV)

Figures and Tables -
Figure 11

Severe adverse events forest plot for network 2 (treatments vs BEV)

Navigate to figure in ReviewOpen in new tab
Summary of findings 1. Summary of overall survival findings

Estimates of effects, certainty assessment and rankings of different treatment options compared with lomustine on overall survival in people with first recurrence of glioblastoma

Patient or population: people with first recurrence of glioblastoma

 Interventions: bevacizumab (BEV), BEV + lomustine (LOM), regorafenib (REG), fotemustine (FOM), ABT414 +  temozolomide (TMZ); BEV + irinotecan (IRI), BEV + onartuzumab (ONA), cediranib (CED), CED + LOM

Comparison: lomustine

Outcome: overall survival
 

All intervention options

Relative effect and 95% CI (network estimate) **

Certainty of the evidence (GRADE)

Ranking*

(9 RCTs; 1734 participants in total)*

LOM
(5 RCTs; 403 participants)
 

Reference comparator

Reference comparator

5.9

REG
(1 RCT; 59 participants)

HR 0.50 (0.33 to 0.76)

⊕⊕⊝⊝
low1

1.3

Depatux‐M (ABT414) + TMZ
(1 RCT; 88 participants)

HR 0.66 (0.47 to 0.92)

⊕⊕⊕⊝
moderate2

2.1

BEV + LOM
(3 RCTs, 401 participants)
 

HR 0.91 (0.75 to 1.10)

⊕⊕⊕⊝
moderate4
 

4.4

FOM
(1 RCT; 32 participants)

HR 0.89 (0.51 to 1.57)

⊕⊕⊝⊝
low3

4.6

ABT414(Depatux‐M)

(1 RCT; 86 participants)

HR 0.96 (0.69 to 1.34)

⊕⊕⊕⊝

low4,6

5.4

CED + LOM

(1 RCT, 129 participants)

HR 1.15 (0.76 to 1.74)

⊕⊕⊕⊝
moderate4

7.2

BEV + IRI
(1 RCT; 82 participants)
 

HR 1.16 (0.71 to 1.88)

⊕⊝⊝⊝
verylow4,5
 

7.4

BEV
(4 RCTs; 259 participants)

HR 1.22 (0.84 to 1.76)

⊕⊕⊝⊝

low4,6

8.1

CED

(1 RCT 131 participants)

HR 1.43 (0.97 to 2.12)

⊕⊕⊕⊝
moderate4

9.5

BEV + ONA

(1 RCT, 64 participants)

HR 1.76 (0.94 to 3.30)

⊕⊝⊝⊝
verylow4
 

10.3
 

Estimates are reported as HR: Hazard Ratio. CI: confidence interval. 

*This refers to the number of studies in the network evaluating the given intervention and the number of participants involved in these studies.

GRADE Working Group grades of evidence
High quality: further research is very unlikely to change our confidence in the estimate of effect.
Moderate quality: further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate.
Low quality: further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate.
Very low quality: we are very uncertain about the estimate.

*We excluded REG and ABT414 on sensitivity analysis, which ranked FOM first, BEV + LOM second, LOM third, BEV+irinotecan (IRI) fourth, and BEV fifth.

1 Downgraded −2 as sparse data from single small open‐label study

2 Downgraded for intransitivity (~ 30% of control arm received TMZ not LOM)

3 Downgraded for imprecision and sparse data from single small study

4 Imprecision

5 No direct evidence and HR for direct effect was estimated from trial report

6 Risk of bias

Figures and Tables -
Summary of findings 1. Summary of overall survival findings
Navigate to table in Review
Summary of findings 2. Summary of progression‐free survival findings

Estimates of effects, certainty assessment and rankings of different treatment options compared with lomustine on overall survival in people with first recurrence of glioblastoma

Patient or population: people with first recurrence of glioblastoma 

Interventions: bevacizumab (BEV), BEV + lomustine (LOM), regorafenib (REG), BEV + irinotecan (IRI), BEV + onartuzumab (ONA), cediranib (CED), CED+LOM

Comparison: lomustine 

Outcome: Progression‐free survival
 

All intervention options

Relative effect and 95% CI (network estimate) **

Certainty of the evidence (GRADE)

Ranking*

(7 RCTs; 1383 participants in total)*

LOM
(4 RCTs; 317 participants) 

Reference comparator

Reference comparator

6.2

BEV+LOM
(3 RCTs, 401 participants) 

HR 0.57 (0.44 to 0.74)

⊕⊝⊝⊝
low1,4  

1.6

REG
(1 RCT; 59 participants)

HR 0.65 (0.42 to 1.01)

⊕⊝⊝⊝
very low1,2

2.7

CED + LOM

(1 RCT, 129 participants)

HR 0.76 (0.50 to 1.18)

⊕⊕⊕⊝
moderate2

3.8

BEV+IRI
(1 RCT; 82 participants) 

HR 0.80 (0.44 to 1.45)

⊕⊕⊝⊝
verylow1,3  

4.2

BEV
(4 RCTs; 200 participants)

HR 0.90 (0.58 to 1.38)

⊕⊝⊝⊝
low2,4

5.2

BEV + ONA

(1 RCT, 64 participants)

HR 0.98 (0.51 to 1.87)

⊕⊕⊝⊝
verylow1,4

5.8
 

CED

(1 RCT 131 participants)

HR 1.05 (0.68 to 1.62)

⊕⊕⊕⊝
moderate2

6.4

Estimates are reported as HR: Hazard Ratio. CI: confidence interval. 

*This refers to the number of studies in the network evaluating the given intervention and the number of participants involved in these studies.

GRADE Working Group grades of evidence
High quality: further research is very unlikely to change our confidence in the estimate of effect.
Moderate quality: further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate.
Low quality: further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate.
Very low quality: we are very uncertain about the estimate.

*We excluded REG on sensitivity analysis, which ranked FOM first, BEV + LOM second, LOM third, BEV+irinotecan (IRI) fourth, and BEV fifth.

1 Sparse data from single small open‐label study

2 Imprecision

3 HRs for direct effect estimated from trial report

4 Risk of bias

Figures and Tables -
Summary of findings 2. Summary of progression‐free survival findings
Navigate to table in Review
Summary of findings 3. Summary of findings for severe adverse events ‐ 1

Estimates of effects, certainty assessment and rankings of different treatment options compared with lomustine for severe adverse events in people with any recurrence of glioblastoma

Patient or population: people with any recurrence of glioblastoma

Interventions: bevacizumab (BEV) + lomustine (LOM), regorafenib (REG), cediranib (CED), CED + LOM, CED + gefitinib (GET)

Comparison: lomustine

Outcome: severe adverse events

Outcomes

(5 RCTs, 1024 participants)

Illustrative comparative risks* (95% CI)**

Relative effect
(95% CI)

Quality of the evidence
(GRADE)

Ranking

Corresponding risk

LOM
(5 RCTs; 330 participants)

39 per 100*

Reference comparator

N/A

1.7

CED

(2 RCTs; 147 participants)

39 per 100 (21 to 72)

RR 1.00 (0.54 to 1.85)

⊕⊕⊝⊝

moderate1

1.7

REG
(1 RCT; 59 participants)

74 per 100 (36 to 100)

RR 1.90 (0.92 to 3.95)

⊕⊕⊝⊝
low1,2

3.8

CED + GET

(1 RCT; 19 participants)

96 per 100 (18 to 100)

RR 2.46 (0.46 to 13.26)

⊕⊝⊝⊝
very low1, 3

4.3

BEV+ LOM

(2 RCTs, 346 participants)

98 per 100 (67 to 100)

RR 2.51 (1.72 to 3.66)

⊕⊕⊕⊕
high

4.7

CED + LOM

(1 RCT, 123 participants)

98 per 100 (50 to 100)

RR 2.51 (1.29 to 4.90)

⊕⊕⊕⊕
high

4.7

*The basis for thisrisk is the mean risk of SAEs with lomustine across the 5 studies that evaluated lomustine. The corresponding risk (and its 95% confidence interval) is based on this risk in the comparison group and the relative effect of the intervention (and its 95% CI).

** Where the CI exceeded 100 values were truncated (at 100)
CI: Confidence interval; RR: Risk Ratio

GRADE Working Group grades of evidence
High quality: further research is very unlikely to change our confidence in the estimate of effect.
Moderate quality: further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate.
Low quality: further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate.
Very low quality: we are very uncertain about the estimate.

1 Downgraded −1 for imprecision

2 Sparse data from single small open label trial

3 Downgraded −2 for imprecision

Figures and Tables -
Summary of findings 3. Summary of findings for severe adverse events ‐ 1
Navigate to table in Review
Summary of findings 4. Summary of findings for severe adverse events ‐ 2

Patient or population: people with any recurrence of glioblastoma

Interventions: bevacizumab (BEV)

Comparison: bevacizumab 9BEV), BEV+carboplatin (CAB), BEV+dasatinib (DAS), BEV+irinotecan (IRI), BEV+onartuzumab (ONA), BEV+TRC105, BEV+VB111, Fotemustine (FOM), BEV+HSPPC96 vaccine

Outcome: severe adverse events

Outcomes

(5 RCTs, 1024 participants)

Illustrative comparative risks* (95% CI)**

Relative effect
(95% CI)

Quality of the evidence
(GRADE)

Ranking

Corresponding risk

BEV
(8 RCTs; 498 participants)

36 per 100*

Reference comparator

N/A

3.1

FOM

(1 RCT, 32 participants)

16 per 100 (4 to 62)

RR 0.44 (0.11 to 1.72)

??? (missing)

1.6

BEV+HSPPC96

1 RCTs; 53 participants)

36 per 100 (12 to 100)

RR 1.01 (0.33 to 3.10)

⊕⊕⊝⊝
low1

3.4

BEV+ONA
(1 RCT; 64 participants)

42 per 100 (21 to 86)

RR 1.17 (0.57 to 2.39)

⊕⊕⊝⊝
low1

4.0

BEV+CAB

(1 RCT; 58 participants)

46 per 100 (22 to 96)

RR 1.27 (0.61 to 2.66)

⊕⊕⊝⊝
low1

4.4

BEV+DAS

(2 RCTs, 83 participants)

19 per 100 (25 to 100)

RR 0.52 (0.69 to 3.34)

⊕⊕⊕⊕
high

5.1

BEV+IRI

(1 RCT, 79 participants)

80 per 100 (43 to 100)

RR 2.22 (1.19 to 4.18)

⊕⊕⊕⊕
high

6.5

BEV+VB111

(1 RCT, 128 participants)

> 100 (92 to 100)

RR 3.77 (2.25 to 6.33)

⊕⊕⊕⊕
high

8.0

BEV+TRC 105

(1 RCT, 49 participants)

> 100 (92 to 100)

RR 6.86 (2.55 to 18.41)

⊕⊕⊕⊕
high

8.8

*The basis for thisrisk is the mean risk of SAEs with lomustine across the 5 studies that evaluated lomustine. The corresponding risk (and its 95% CI) is based on this risk in the comparison group and the relative effect of the intervention (and its 95% CI).

** Where the corresponding risk value and, or CI exceeded 100 values were truncated (at 100)
CI: Confidence interval; RR: Risk Ratio

GRADE Working Group grades of evidence
High quality: further research is very unlikely to change our confidence in the estimate of effect.
Moderate quality: further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate.
Low quality: further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate.
Very low quality: we are very uncertain about the estimate.

Figures and Tables -
Summary of findings 4. Summary of findings for severe adverse events ‐ 2
Navigate to table in Review
Table 1. Median survival of participants in included studies

Study ID

No. participants

No. of recurrences

Comparison

Median PFS (months)

Median OS

(months)

Study design

Comment

First recurrence 

Azoulay 2017 

183

 1st

I: Re‐operation

C:No re‐operation

 NR

9.8

5.0

NRS

Patients selected for different arms based on clinical factors therefore a high risk of selection bias.

Batchelor 2013

325

1st

I: CED

I: CED+LOM

C: LOM

3.0

4.1

2.7

8.0

9.4

9.8

Phase 3

CED did not significantly improve survival.

Brandes 2016b

91

1st

I: BEV

C: FOT

3.4

3.5

7.3

8.7

Phase 2

Investigators concluded that single‐agent BEV "may have a role'.

Brandes 2018

123

1st

I: BEV+LOM

C: LOM

2.3

1.8

6.4

5.5

Phase 2

Terminated early due to high drop‐out rate during first‐line treatment.

Brandes 2016a

119

1st

I: GAL+LOM

C: LOM

1.8

1.9

6.7

7.5

Phase 2

Investigators concluded that GAL failed to demonstrate activity.

Brown 2016

38

1st

I: CED+GEF

C: CED

3.6

2.8

7.2

5.5

Phase 2

Study was underpowered.

Cloughesy 2017

129

1st

I: ONA+BEV

C: BEV

3.9

2.9

8.8

12.6

Phase 3

No evidence of clinical benefit with addition of ONA

Dresemann 2010

240

1st

I: IMA+HU

C: HU

1.4

1.4

NR

Phase 3

No clinically meaningful differences.

Lombardi 2019

119

1st

I: REG

C: LOM

2.0

1.9

7.4

5.6

Phase 2

Considered by investigators to have potential for further study.

Kunwar 2010 

296

1st

I: cintredekin besudotox

C: gliadel wafers

NR

9.1

8.8

Phase 3

No survival difference but higher risk of pulmonary embolism with cintredekin besudotox (P = 0.014).

Narita 2019

88

1st

I: PPV

C: Placebo

NR

8.4

8.0

Phase 3

Did not meet primary endpoint and PPV shortened the OS in certain patients.

Omuro 2018

20

1st

I: NIV

C:NIV+IPI

1.9

1.5

10.4

9.2

Phase 2

NIV was better tolerated than NIV+IPI

Puduvalli 2018 

74

1st

I: BEV+VOR

C: BEV

3.7

3.9

7.8

9.3

Phase 2

 No clinical benefit with BEV+VOR. Does not warrant further investigation.

Reardon 2015b

78

1st

I: AFA+TMZ

I: AFA

C: TMZ

1.5

1.0

1.9

8.0

9.8

10.6

Phase 2

Investigators concluded that afatinib had limited single agent activity.

Scorsetti 2015 

43

1st

I: Re‐operation

C: No re‐operation

15

5

17

6

NRS

Patients selected for different arms based on clinical factors therefore a high risk of selection bias. 

Suchorska 2016

93

1st
 

I: re‐operation

C: no re‐operation

2.0

1.8

11.4

9.8

NRS
 

Patients selected for different arms based on clinical factors therefore a high risk of selection bias. Complete resection was associated with improved survival compared with incomplete resection. 

Taal 2014

153

1st

I: LOM

I: BEV+LOM

C: BEV

1.0

4.0

3.0

8.0

12.0

8.0

Phase 2

BEV+LOM met investigator criteria for further evaluation in phase 3 studies. Single agent BEV was found to have low activity.

Twelves 2017

21

1st

I: CBD:THC

C: placebo

‐

‐

18.3

12.3

All patients received dose intense TMZ. 1 year survival was 83% and 56% in the CBD:THC and placebo groups, respectively.

van den Bent 2018

260

1st

I: ABT414+TMZ

I: ABT414

C: TMZ or LOM

2.7

1.9

1.9

9.6

7.9

8.2

Phase 2

ABT414 had insufficient single‐agent activity.

Wick 2017

437

1st

I: LOM+BEV

C: LOM

4.2

1.5

9.1

8.6

Phase 3

No OS benefit with combination.

Any, second or subsequent recurrence

Bloch 2017

90

1st or 2nd (% not reported)

I: HSPPC‐96+BEV

C: BEV

NR

7.5

10.7

Phase 2

Terminated for futility after interim analysis.

Cuncannon 2019 

43

2nd or 3rd relapse

I: BEV

C: BSC

NR 

6.0

1.0

NRS

Patients selected for different arms based on patient choice therefore high risk of selection bias.

Friedman 2009

167

1st and 2nd (< 20%)

I: BEV+IRI

C: BEV

5.6

4.2

8.7

9.2

Phase 2

No survival benefit with combination.

Reardon 2018a

80

1st and 2nd (% not reported)

I: PEM

C: PEM+BEV

NR

8.8

10.3

Phase 2

Investigators reported that there was no monotherapy activity.

Reardon 2018b

48

1st and 2nd (33%)

I: TNB

C: TNB+BEV

0.7

3.6

11.4

9.5

NRS

Accrual to TNB was discontinued early due to lack of monotherapy activity.

Cloughesy 2018

256

1st and 2nd (27%)

I: VB111+BEV

C: BEV

3.4

3.7

6.8

7.9

Phase 3

VB111+BEV failed to improve outcomes

Duerinck 2018

79

Any

I: AXI

C: AXI+LOM

2.9

3.0

6.3

6.7

Phase 2

No indication that AXI+LOM improves results.

Field 2015

122

Any (33% 2nd or subsequent)

I: BEV+CAB

C: BEV

3.5

3.5

6.9

7.5

Phase 2

Adding CAB resulted in more toxicity without additional clinical benefit.

Galanis 2017

101

2nd or later

I: TRC105+BEV

C: BEV

2.9

3.2

10.0

7.4

Phase 2

Investigators reported no significant survival difference with TRC105.

Galanis 2019

121

Any (% not reported)

I: DAS+BEV

C: BEV

3.2

3.2

7.3

7.7

Phase 2

DAS+BEV did not significantly improve clinical outcomes.

Gilbert 2017

117

Any (% not reported)

I: BEV+TMZ

C: BEV+IRI

4.7

4.1

9.4

7.7

Phase 2

Both arms surpassed predetermined efficacy thresholds.

Heiland 2016

43

3rd

I: BEV

C: BEV+LOM

2.3

6.1

4.1

6.6

NRS

Investigators concluded that last‐line therapy with BEV/LOM results in a longer PFS and OS compared to BEV only.

Modh 2018

34

"median of 3"

I: FSRS + BEV

C: Chemo+BEV

5.3

1.8

7.1

4.8

RCT

Investigators concluded that FSRS in heavily pretreated patients with recurrent malignant glioma is feasible and improves local control and PFS

Reardon 2011

23

"heavily pre‐treated"

I:ETO+BEV

C:TMZ+BEV

1.9

0.9

4.4

2.9

Phase 2

Investigators concluded that metronomic ETO or TMZ is ineffective administered with BEV in this context.

Santos 2018

32

Unclear

I: IPA + ketogenic diet

C: IPA + regular diet

NR

NR

NRS

Investigators concluded that results on ketogenic diet were encouraging.

Stupp 2012

237

Any 

(88% 2nd or later)

I: TTF

C: BPC

2.2

2.1

6.6

6.0

 Phase 3

No improvement in survival but toxicity and QOL favoured TTF.

Tsien 2019

170

NR

I: BEV+RT

C: BEV

NR

10.1

9.7

Phase 2

BEV +RT was associated with improved 6‐month PFS (54% vs 29%).

Weathers 2016

69

1st (68%), 2nd, and 3rd

I: LOM+BEV (low dose)

C: BEV

4.3

4.1

9.6

8.3

Phase 2

Median PFS for 1st recurrence (47 pts) was 5.0 mths vs 3.2 mths, respectively; median OS was 13.1 mths vs 8.8 mths, respectively.

Wick 2010

266

1st and 2nd (25%)

I: ENZ

C: LOM

1.5

1.6

6.6

7.1

Phase 3

Terminated early due to futility.

Wick 2014

84

1st and 2nd/3rd (29%)

I: APG101+RT

C: RT

4.5

2.5

11.5

11.5

Phase 2

Investigators concluded that APG101 had potential for further clinical development.

BEV = bevacizumab
BPC = Best Physician's Care
BSC = Best supportive care
CAB = carboplatin
CBD:THC = cannabidiol:delta‐9‐tetrahydrocannabinol
CED = cediranib
DAS = desatinib
ENZ = enzastaurine
ETO = etoposide
HU = hydroxyurea
IMA =imatinib
IPA = intranasal perillyl alcohol
IRI = irinotecan
IPI = ipilimumab
LOM = lomustine
NIV = nivolumab
PEM = pembrolizumab
PPV = personalised petide vaccine
RT = radiotherapy
ONA = onartuzumab
TTF = tumour‐treating fields
TNB = trebananib
FSRH = fractionated stereotactic radiotherapy
TMZ = temozolomide

Figures and Tables -
Table 1. Median survival of participants in included studies
Navigate to table in Review
Table 2. Overall survival: League table showing HR and 95% CI estimates for all intervention comparisons

LOM

ABT414

ABT414_TMZ

BEV

BEV_IRI

BEV_LOM

BEV_ONA

CED

CED_LOM

FOT

REG

LOM

0.96 (0.69,1.34)

0.66 (0.47,0.92)

1.22 (0.84,1.76)

1.16 (0.71,1.88)

0.91 (0.75,1.10)

1.76 (0.94,3.30)

1.43 (0.97,2.12)

1.15 (0.76,1.74)

0.89 (0.51,1.57)

0.50 (0.33,0.76)

1.04 (0.75,1.45)

ABT414

0.68 (0.49,0.95)

1.27 (0.77,2.08)

1.20 (0.67,2.17)

0.95 (0.64,1.39)

1.83 (0.90,3.73)

1.49 (0.89,2.50)

1.20 (0.71,2.03)

0.93 (0.48,1.79)

0.52 (0.31,0.89)

1.52 (1.09,2.12)

1.46 (1.05,2.04)

ABT414_TMZ

1.85 (1.13,3.04)

1.76 (0.98,3.17)

1.38 (0.94,2.03)

2.68 (1.32,5.46)

2.18 (1.30,3.65)

1.75 (1.03,2.97)

1.36 (0.70,2.62)

0.76 (0.45,1.30)

0.82 (0.57,1.19)

0.79 (0.48,1.30)

0.54 (0.33,0.89)

BEV

0.95 (0.70,1.30)

0.75 (0.52,1.08)

1.45 (0.87,2.41)

1.18 (0.69,2.02)

0.95 (0.54,1.64)

0.73 (0.48,1.13)

0.41 (0.24,0.72)

0.86 (0.53,1.40)

0.83 (0.46,1.49)

0.57 (0.32,1.02)

1.05 (0.77,1.44)

BEV_IRI

0.79 (0.48,1.28)

1.52 (0.84,2.77)

1.24 (0.67,2.31)

0.99 (0.53,1.88)

0.77 (0.45,1.31)

0.43 (0.23,0.82)

1.10 (0.91,1.33)

1.06 (0.72,1.55)

0.72 (0.49,1.06)

1.34 (0.92,1.94)

1.27 (0.78,2.07)

BEV_LOM

1.94 (1.03,3.64)

1.58 (1.02,2.44)

1.27 (0.80,1.99)

0.98 (0.56,1.73)

0.55 (0.35,0.87)

0.57 (0.30,1.07)

0.55 (0.27,1.11)

0.37 (0.18,0.76)

0.69 (0.41,1.15)

0.66 (0.36,1.20)

0.52 (0.27,0.97)

BEV_ONA

0.81 (0.39,1.71)

0.65 (0.31,1.39)

0.51 (0.26,0.99)

0.28 (0.13,0.60)

0.70 (0.47,1.03)

0.67 (0.40,1.12)

0.46 (0.27,0.77)

0.85 (0.50,1.45)

0.81 (0.43,1.50)

0.63 (0.41,0.98)

1.23 (0.59,2.58)

CED

0.80 (0.54,1.20)

0.62 (0.31,1.24)

0.35 (0.20,0.62)

0.87 (0.58,1.31)

0.84 (0.49,1.42)

0.57 (0.34,0.97)

1.06 (0.61,1.84)

1.01 (0.53,1.90)

0.79 (0.50,1.24)

1.53 (0.72,3.24)

1.25 (0.83,1.86)

CED_LOM

0.78 (0.38,1.56)

0.44 (0.24,0.78)

1.12 (0.64,1.98)

1.08 (0.56,2.08)

0.74 (0.38,1.42)

1.36 (0.89,2.10)

1.30 (0.76,2.21)

1.02 (0.58,1.80)

1.97 (1.01,3.85)

1.61 (0.81,3.20)

1.29 (0.64,2.60)

FOT

0.56 (0.28,1.13)

1.99 (1.32,3.01)

1.92 (1.13,3.25)

1.31 (0.77,2.22)

2.42 (1.40,4.21)

2.31 (1.22,4.35)

1.81 (1.15,2.85)

3.51 (1.66,7.44)

2.86 (1.62,5.04)

2.29 (1.28,4.10)

1.78 (0.88,3.58)

REG

ABT414 = depatux‐m
BEV = bevacizumab
CED = cediranib
IRI = irinotecan
LOM = lomustine
ONA = onartuzumab

Figures and Tables -
Table 2. Overall survival: League table showing HR and 95% CI estimates for all intervention comparisons
Navigate to table in Review
Table 3. Progression‐free survival: League table showing HR and 95% CI estimates for all intervention comparisons

LOM

BEV

BEV_IRI

BEV_LOM

BEV_ONA

REG

LOM

0.90 (0.58,1.38)

0.80 (0.44,1.45)

0.57 (0.44,0.74)

0.98 (0.51,1.87)

0.65 (0.42,1.01)

1.12 (0.72,1.72)

BEV

0.90 (0.60,1.34)

0.64 (0.41,0.99)

1.09 (0.67,1.77)

0.73 (0.39,1.35)

1.25 (0.69,2.25)

1.12 (0.75,1.67)

BEV_IRI

0.71 (0.39,1.28)

1.22 (0.65,2.28)

0.81 (0.39,1.69)

1.75 (1.36,2.26)

1.57 (1.02,2.43)

1.41 (0.78,2.55)

BEV_LOM

1.71 (0.89,3.29)

1.14 (0.68,1.90)

1.02 (0.53,1.96)

0.92 (0.57,1.49)

0.82 (0.44,1.54)

0.58 (0.30,1.12)

BEV_ONA

0.67 (0.30,1.46)

1.54 (0.99,2.40)

1.38 (0.74,2.57)

1.24 (0.59,2.59)

0.88 (0.53,1.46)

1.50 (0.68,3.30)

REG

BEV = bevacizumab
CED = cediranib
IRI = irinotecan
LOM = lomustine
ONA = onartuzumab
REG = regorafenib

Figures and Tables -
Table 3. Progression‐free survival: League table showing HR and 95% CI estimates for all intervention comparisons
Navigate to table in Review
Table 4. Severe adverse events for treatments compared with lomustine: League table with effect estimates and 95% CIs

LOM

BEVLOM

CED

CEDGET

CEDLOM

REG

LOM

2.51 (1.72,3.66)

1.00 (0.54,1.85)

2.46 (0.46,13.26)

2.51 (1.29,4.90)

1.90 (0.92,3.95)

0.40 (0.27,0.58)

BEVLOM

0.40 (0.19,0.82)

0.98 (0.17,5.50)

1.00 (0.46,2.15)

0.76 (0.33,1.72)

1.00 (0.54,1.85)

2.51 (1.22,5.17)

CED

2.46 (0.51,11.80)

2.51 (1.43,4.42)

1.90 (0.73,4.94)

0.41 (0.08,2.19)

1.02 (0.18,5.73)

0.41 (0.08,1.95)

CEDGET

1.02 (0.19,5.40)

0.77 (0.12,4.84)

0.40 (0.20,0.78)

1.00 (0.46,2.15)

0.40 (0.23,0.70)

0.98 (0.19,5.18)

CEDLOM

0.76 (0.28,2.03)

0.53 (0.25,1.09)

1.32 (0.58,3.00)

0.53 (0.20,1.36)

1.29 (0.21,8.10)

1.32 (0.49,3.54)

REG

BEV = bevacizumab
CED = cediranib
GET = getitinib
LOM = lomustine
REG = regorafenib

Figures and Tables -
Table 4. Severe adverse events for treatments compared with lomustine: League table with effect estimates and 95% CIs
Navigate to table in Review
Table 5. Severe adverse events for treatments compared with bevacizumab: League table with effect estimates and 95% CIs

BEV

BEVCAB

BEVDAS

BEVIRI

BEVONA

BEVTRC105

BEVVB111

FOM

HSPPCBEV

BEV

1.27 (0.61,2.66)

1.52 (0.69,3.34)

2.22 (1.18,4.18)

1.17 (0.57,2.39)

6.86 (2.55,18.41)

3.77 (2.25,6.33)

0.44 (0.11,1.72)

1.01 (0.33,3.10)

0.79 (0.38,1.64)

BEVCAB

1.19 (0.41,3.51)

1.75 (0.66,4.61)

0.92 (0.33,2.57)

5.39 (1.57,18.47)

2.97 (1.21,7.29)

0.35 (0.07,1.63)

0.79 (0.21,3.03)

0.66 (0.30,1.45)

0.84 (0.29,2.46)

BEVDAS

1.46 (0.53,4.02)

0.77 (0.27,2.23)

4.51 (1.28,15.96)

2.48 (0.97,6.37)

0.29 (0.06,1.40)

0.66 (0.17,2.62)

0.45 (0.24,0.85)

0.57 (0.22,1.51)

0.68 (0.25,1.88)

BEVIRI

0.53 (0.20,1.37)

3.09 (0.95,9.97)

1.70 (0.75,3.84)

0.20 (0.04,0.89)

0.45 (0.13,1.65)

0.85 (0.42,1.75)

1.09 (0.39,3.03)

1.30 (0.45,3.76)

1.90 (0.73,4.93)

BEVONA

5.86 (1.73,19.82)

3.22 (1.33,7.79)

0.38 (0.08,1.75)

0.86 (0.23,3.26)

0.15 (0.05,0.39)

0.19 (0.05,0.64)

0.22 (0.06,0.78)

0.32 (0.10,1.05)

0.17 (0.05,0.58)

BEVTRC105

0.55 (0.18,1.68)

0.06 (0.01,0.35)

0.15 (0.03,0.66)

0.26 (0.16,0.44)

0.34 (0.14,0.83)

0.40 (0.16,1.03)

0.59 (0.26,1.33)

0.31 (0.13,0.75)

1.82 (0.60,5.54)

BEVVB111

0.12 (0.03,0.50)

0.27 (0.08,0.92)

2.26 (0.58,8.80)

2.88 (0.61,13.49)

3.44 (0.72,16.52)

5.03 (1.12,22.49)

2.65 (0.57,12.28)

15.51 (2.89,83.17)

8.54 (2.00,36.51)

FOM

2.28 (0.39,13.29)

0.99 (0.32,3.04)

1.26 (0.33,4.82)

1.51 (0.38,5.93)

2.20 (0.61,7.98)

1.16 (0.31,4.39)

6.80 (1.52,30.30)

3.74 (1.09,12.86)

0.44 (0.08,2.55)

HSPPCBEV

BEV = bevacizumab
CAB = carboplatin
DAS = desatinib
IRI = irinotecan
FOM = fotemustine
HSPCC = HSPCC‐96 vaccine
ONA = onartuzumab
TRC105 = carotuximab

Figures and Tables -
Table 5. Severe adverse events for treatments compared with bevacizumab: League table with effect estimates and 95% CIs
Navigate to table in Review