Types of studies

Randomised and non‐randomised comparative studies, including cluster‐RCTs and controlled before‐after studies (CBAs) that control for baseline differences. We excluded cross‐over designs, case‐control studies, and studies without a comparison group.

Types of participants

People of any age with a presentation that might suggest a primary brain tumour, specifically focal neurological deficit, headache suspicious of cancer, recent cognitive problems, and late onset seizures. It is accepted that only a small proportion of people would ultimately have a brain tumour, although it would be within the differential diagnosis. We did not plan to exclude participants with a past history of systemic cancer, but had planned to manage these data as a separate subgroup if we found any.

Types of interventions

Any active intervention that may influence the diagnostic pathway, e.g. clinical guidelines, direct access imaging, public health campaigns, educational and other interventions that might lead to early identification of primary brain tumours.

Types of outcome measures

Primary and secondary outcome measures are as follows.

Electronic searches

We searched the following databases from 2000 (This is when the UK National Cancer Plan was introduced by the UK's Department of Health with Referral guidelines for suspected cancer, which has been updated and replaced by NICE 2017) to 13 January 2020:

• Cochrane Central Register of Controlled Trials (CENTRAL; 2020, Issue 1), in the Cochrane Library;

• MEDLINE via Ovid (2000 to December week 4 2019);

• Embase via Ovid (2000 to 2020 week 1).

For economic evidence, we searched the EED database from the end of December 2014 (when the last records were added to that database) to January 2000, and MEDLINE and Embase from 1 January 2015 to January 2020, 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.

Please refer to Appendix 1 for CENTRAL, MEDLINE, and Embase search strategies.

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

Searching other resources

We searched Clinicaltrials.gov on 1 May 2020. We also handsearched conference proceedings from 2014 to 2018 (five years) of conferences of the British Neuro‐oncology Society, the Society for Neuro‐oncology, the European Association of Neuro‐oncology, and the World Federation of Neuro‐oncology Societies to identify other relevant ongoing or unpublished studies.

Selection of studies

After removing duplicates, the Information Specialist at the Cochrane Gynaecological, Neuro‐oncology and Orphan Cancer Group (GNOC) downloaded all titles and abstracts retrieved by electronic searching to Covidence to facilitate study selection. Two review authors (TL, ET) independently screened these records and obtained copies of the full texts of potentially eligible references. At least two review authors (TL, ET, DH, TD) independently assessed each full text for eligibility. Disagreements were resolved by discussion, or by consultation with another reviewer (RG), or the wider group of review authors, if necessary. We have documented reasons for exclusion in the Characteristics of excluded studies tables of the review.

In this version of the review, we did not identify any studies eligible for inclusion. In future versions, if any studies meet the inclusion criteria, we will use the following methods:

Data extraction and management

Three review authors (TL, ET, TD) will independently extract the following data from any eligible studies to a piloted data extraction form. We will resolve discrepancies through discussion, or if required, by consulting another review author (DH or RG). 

• Author contact details

• Country

• Setting

• Dates of participant accrual

• Trial registration number or 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

  • Referral pathway (stroke, epilepsy, brain tumour, self‐referral)

  • Presenting symptoms, signs

  • Type of surgery

  • Other treatment

• Intervention details

  • Type of intervention

  • Type of comparator

• Duration of follow‐up

• Primary outcome(s) of the study

• Review outcomes

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

  • For continuous outcomes, we will extract 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 will also extract change‐from‐baseline score data, where reported, and note the type of scale used

  • We will extract adjusted statistics, where reported

  • Where possible, all data extracted will be those relevant to an intention‐to‐treat analysis, in which participants are analysed in the groups to which they were assigned

  • We will resolve differences between review authors by discussion, or by appeal to the other review authors, when necessary

• Risk of study bias (see below)

Assessment of risk of bias in included studies

For randomised trials, we will assess the risk of bias using Cochrane's tool and the criteria specified in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011). This includes assessment of:

• random sequence generation;

• allocation concealment;

• blinding of participants and healthcare providers;

• blinding of outcome assessors;

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

• selective reporting of outcomes;

• other possible sources of bias, e.g. insufficient number of participants, baseline differences in group characteristics.

For non‐randomised studies (non‐randomised trials and controlled before‐after studies), we will use the ROBINS‐I tool for assessing risk of bias (Sterne 2016). This includes 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 are aware of intervention status, different methods are used to assess the outcome, or measurement errors are related to intervention status or effects);

• bias in selection of the reported result.

Two review authors (TL, ET or TD) will independently assess risk of bias, and resolve differences by discussion or by appeal to another review author (RG). We will summarise judgements in 'Risk of bias' tables, along with the characteristics of any included studies. We will interpret results in light of the 'Risk of bias' assessment. For more details about the assessment of risk of bias, see Appendix 2.

Measures of treatment effect

• For dichotomous outcomes, we will calculate the effect size as a risk ratio (RR) with its 95% confidence interval (CI).

• For continuous outcomes (e.g. QoL scores), in which the same measurement scales were used, we will pool data as a mean difference (MD) with its 95% CI. If studies used different time points and measurement scales, and we consider it clinically meaningful to do so, we will pool data using the standardised mean difference (SMD).

• For time‐to‐event data, we will calculate the effect size as a hazard ratio (HR) with its 95% CI.

Unit of analysis issues

At least two review authors will independently review unit‐of‐analysis issues (TL, TD), as described in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011), for each included study. These include reports where there are multiple observations for the same outcome, e.g. repeated measurements with different scales, or outcomes measured at different time points. When time points differ across studies, or there are multiple observations for the same outcome, we will synthesise the findings narratively.

We will analyse cluster‐randomised trials alongside individually‐randomised trials, and will adjust their sample sizes using the methods described in the Cochrane Handbook for Systematic Reviews of Interventions using an estimate of the intra‐cluster correlation co‐efficient (ICC) derived from the trial (if possible), from a similar trial, or from a study of a similar population, if the authors had not taken clustering into account. We will report the source of the ICC and conduct sensitivity analyses to investigate the effect of variation in the ICC. We consider it reasonable to combine the results from both cluster‐randomised and individually‐randomised study designs if there is little heterogeneity between the study designs, and we consider the interaction between the effect of intervention and the choice of randomisation unit to be unlikely. We will acknowledge heterogeneity in the randomisation unit, and perform subgroup analysis to investigate the effects of the randomisation unit. We will resolve differences by discussion with a third review author (RG).

Dealing with missing data

For included studies, we will note the levels of attrition, but will not impute missing data. In the event of missing data, we will write to study authors to request the data, and describe in the 'Characteristics of included studies' table how we obtained any missing data. We will explore the impact of including studies with high level of missing data in the overall assessment of treatment effect by using sensitivity analysis.

Assessment of heterogeneity

We will assess statistical heterogeneity between studies by visual inspection of forest plots (Higgins 2003), and by using a formal statistical test of the significance of the heterogeneity, assessed using the T², I², and Chi² statistics (Deeks 2001). We will regard heterogeneity as substantial if an I² is greater than 60%, and either T² is greater than zero, or there is a low P value (< 0.10) in the Chi² test for heterogeneity. Where there is evidence of substantial heterogeneity (I² > 60%), we will investigate and report the possible reasons for it, e.g. clinical heterogeneity, high risk of bias studies, etc.

Should we use a different approach to synthesis, which does not support production of a forest plot with effect sizes, it may still be useful to report on heterogeneity in the standardised effect measure used, e.g. effect direction, which is akin to an informal sensitivity analysis, the results of which are speculative, but may be useful for readers. 

Assessment of reporting biases

Where there are 10 or more studies in a meta‐analysis, we will investigate reporting biases, such as publication bias, through visual inspection of funnel plots. If asymmetry is suggested by visual assessment, we will perform exploratory analyses to investigate it.

Data synthesis

We will pool dichotomous data as risk ratios (RRs), and continuous data as mean differences (MDs) or standardised mean differences (SMDs) if different scales have been used. We will use the random‐effects model with inverse variance weighting in Review Manager 2014, because we expect clinical heterogeneity among included studies. We will treat the random‐effects summary as the average range of possible intervention effects, and we will discuss the clinical implications of intervention effects differing between trials. If any trials contributing to a meta‐analysis have multiple intervention groups, we will divide the 'shared' comparison group into the number of treatment groups and comparisons between each treatment group, and treat the split comparison group as independent comparisons.

If different studies report either dichotomous or continuous data for the same outcome, we will attempt to convert continuous data to dichotomous data to facilitate meta‐analysis.

Assuming we find at least two included studies that are sufficiently similar for the findings to be clinically meaningful, we will perform a meta‐analysis of the results. If it is not clinically meaningful to pool data, we will attempt a narrative synthesis of the evidence.

We will synthesise data from non‐randomised studies separately from randomised trials. As different non‐randomised studies may report results in different ways, when found, we may tabulate this sort of evidence and synthesise it narratively.

In any evidence synthesis (meta‐analysis and narrative synthesis), we will subgroup interventions and strategies according to how they might work (see How the intervention might work). If data are very sparse, we may report raw data from individual studies.

Subgroup analysis and investigation of heterogeneity

If it is meaningful to do so, we will synthesise data from different interventions together in the first instance. If we identify substantial heterogeneity, we will use subgroup and sensitivity analyses to investigate it. Where there are sufficient data, we anticipate the following subgroup analysis.

• Type of intervention: e.g. clinical guidelines, direct access imaging, public health campaigns, educational, and other

• Type of referral: referral for suspected brain tumour, or referral for other suspected conditions in which the differential diagnosis includes brain tumour, e.g. epilepsy, stroke, headache

• Age: children younger than 16 years old, young adults (16 to 40 years old), and adults older than 40 years

• Setting: high‐income country and low‐ or middle‐income country settings

We will use formal tests for subgroup differences.

Sensitivity analysis

We plan to perform sensitivity analyses (i) to investigate instances of substantial heterogeneity identified in meta‐analyses of the primary outcomes, and (ii) to investigate how study quality affects the estimate of effect after excluding studies at high risk of bias.

Summary of findings and assessment of the certainty of the evidence

Based on the methods described in Chapter 11 of the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011), we will prepare a 'Summary of findings' table to present the results of the following outcomes:

• time from first symptom to diagnosis;

• time from first presentation to diagnosis;

• proportion of people identified with brain tumours (any type) of those referred with suspicious symptoms.

We will use the GRADE system to rank the certainty of the evidence, with two review authors independently grading the evidence, and resolving differences by discussion, or by involving a third review author (Schünemann 2011). Where the evidence is based on single studies, or where there is no evidence on a specific outcome, we will include the outcome in the 'Summary of findings' table, and grade or explain accordingly. We will provide a rationale for each judgement in the table footnotes. In the absence of a single estimate of effect (when meta‐analysis was not possible), we will rate the certainty of the narrative evidence using the GRADE approach (Murad 2017). We will interpret the results of the graded evidence based on Cochrane Effective Practice and Organisation of Care guidance (EPOC 2017).