Pharmacological treatment of depression in people with a primary brain tumour

Primary brain tumours

The global incidence of primary brain tumours is estimated to be 10.8 per 100,000 person‐years (de Robles 2015), and increasing (Patel 2019). As a group, primary brain tumours differ from metastatic tumours, which spread to the brain from cancer elsewhere in the body. Primary brain tumours can be categorised according to World Health Organization (WHO) criteria reflecting their cell of origin, degree of malignancy, and molecular parameters (Louis 2016). The most common type of malignant primary brain tumour is glioma (Lapointe 2018), and most gliomas (70% to 80%) are 'high‐grade' aggressive tumours. With current treatment, approximately 10% of people with the most aggressive form of glioma (glioblastoma multiforme, GBM) may survive five years after diagnosis (Stupp 2009). A minority of gliomas are slowly growing 'low‐grade' tumours associated with a median survival of approximately seven years (Claus 2015). Low‐grade gliomas are more frequently associated with epileptic seizures than high‐grade tumours (Pallud 2019). Meningiomas are another common type of primary brain tumour, which occurs more frequently in women and the elderly and may be cured by surgery (Wiemels 2010).

Depression

Major depressive disorder (depression) is a clinical syndrome of sadness or loss of interest or/enjoyment, or both; negative beliefs (e.g. feelings of guilt, hopelessness, or low self‐esteem); psychomotor abnormalities (e.g. fatigue and psychomotor retardation); and biological manifestations (e.g. sleep and appetite disturbances) (Gelder 2012). Because sadness, negative thoughts (e.g. about loss of health or death) and physical symptoms can be normal in people with cancer, the gold‐standard method of diagnosing depression is a structured clinical interview conducted by a specialist in mental health (APA 2016). Significant depressive symptoms can also be identified in people with cancer using rating scales (e.g. Zigmond 1983), which use cut‐off scores to estimate the likelihood of depression and, being easier to administer, are often used as a 'proxy' diagnosis in clinical research.

Types of studies

For evidence of efficacy we aimed to include:

• RCTs.

For evidence of adverse effects we also aimed to include:

• controlled clinical trials;

• cohort studies;

• case‐control studies.

We aimed to find all published and unpublished studies, in any language.

Types of participants

Types of interventions

Any drug prescribed to treat depression compared, where relevant, with a control group receiving any other treatment. We anticipated that the most common drug class would be an antidepressant, but any category of prescription drug was eligible. Studies administering only unlicensed herbal remedies (e.g. St John's wort) were excluded.

Types of outcome measures

Electronic searches

For this review update, we searched the following databases up to 30 September 2019:

• the Cochrane Central Register of Controlled Trials (CENTRAL; 2019, Issue 9) in the Cochrane Library;

• MEDLINE via Ovid (October 2012, week 3 to September 2019, week 3);

• Embase via Ovid (2012, week 42 to 2019, week 39);

• PsycINFO (October 2012 to September 2019).

We had previously searched the British Nursing Index, LILACS (Latin American and Caribbean Health Science Information database), PSYNDEX, the NHS National Research Register, the NHS Centre for Reviews and Dissemination's Database of Abstracts of Reviews of Effects (DARE), and Web of Knowledge (covering Science SciSearch, Social Sciences Citation Index, and Biological Abstracts) (to July 2009).

In the original review we constructed individual search strategies for each database (see Appendix 1 to Appendix 2). In the update, we amended search strategies where relevant to remove duplicate terms (Appendix 3 to Appendix 4). Because we aimed to include studies other than RCTs, we conducted topic searches only.

Searching other resources

For the original review we contacted:

• pharmaceutical companies manufacturing all antidepressants listed in the British National Formulary (BNF);

• International Network of Agencies for HTA (INAHTA);

• British Library Document Supply Centre (BLDSC).

This yielded no relevant information and was not repeated for the update.

We handsearched the following journals (up to November 2019) for articles and conference abstracts:

• Journal of Neurology, Neurosurgery and Psychiatry;

• Journal of Neuro‐Oncology;

• Journal of Clinical Oncology;

• Neuro‐Oncology.

We aimed to handsearch:

• reference lists of eligible studies;

• book chapters identified in the reference lists of eligible studies.

Selection of studies

Two review authors (ZB and SH) initially filtered studies based on the title and abstract. We then obtained full copies of potentially relevant studies, which two review authors (ZB and SH) independently assessed for eligibility, supervised by an experienced research fellow (FB).

Data extraction and management

Two review authors (ZB and SH) were to independently extract data from eligible studies using a predefined data extraction form. For the categorical outcome of depression, we planned to extract the number of depressed participants in each arm at final follow‐up in order to estimate risk ratio (RR). For continuous outcomes, we planned to extract the mean and standard deviation (SD) of the outcome of interest in each arm at final follow‐up. We planned to estimate from the means and SDs whether the data were skewed; if they were, we would write to the study authors requesting the mean and SDs of log transformed data. For time‐to‐death data, we planned to extract the hazard ratio (HR) and its variance from trial reports; if these were not presented, we would attempt to either estimate them or extract relevant data from Kaplan‐Meier survival curves. Failing this, we would estimate RR as outlined above. One review author (ZB or SH) would enter data into Review Manager 5, and the other review author would check the data entry (Review Manager 2014).

Assessment of risk of bias in included studies

We intended to assess and report on the methodological risk of bias of included studies in accordance with the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011), which recommends the explicit reporting of RCTs, using the 'Risk of bias' 1 tool, including all of the domains in this tool: random sequence generation; allocation concealment; blinding of participants and personnel; blinding of outcome assessment; incomplete outcome data; selective reporting; and other sources of bias. We would categorise risk of bias as high, low, or unclear, and present this information in a ‘Risk of bias’ table. We would discuss the risk of bias in the included studies, resolving any persisting disagreements between the authors by consulting a third review author.

Measures of treatment effect

When summarising the categorical outcome of depression, we planned to calculate risk ratios (RR) with 95% confidence intervals (CI). When comparing continuous outcome data from studies using different rating scales, we would use the standardised mean difference (SMD); we would calculate the mean difference (MD) for studies using the same scale. Both would be presented with 95% CI for individual outcomes in individual studies. For time‐to‐death data, we planned to pool HRs using the generic inverse variance facility of Review Manager 5 (Review Manager 2014). We planned to analyse secondary outcomes as continuous variables only (MDs or SMDs). Where possible, participants were to be analysed in the groups to which they had been randomised, irrespective of the treatment they actually received.

Unit of analysis issues

We would take different levels of randomisation (e.g. at the level of participants or groups) into account. When there were long‐term follow‐up assessments available within trials, we would analyse outcomes for two different follow‐up categories: short term (i.e. 0 to 6 months); or medium to long term (i.e. 6 months and more). If we identified studies with multiple intervention groups, we would make pair‐wise comparisons between all possible pairs of intervention groups. We would ensure that double‐counting of participants in the analysis did not occur.

Dealing with missing data

We would calculate missing measures of uncertainty (e.g. CIs and SDs) where possible or contact the study authors to request the data. We did not intend to impute missing data.

We aimed to record the number of participants in each intervention arm whose outcomes were not reported at the end of the study, noting if loss to follow‐up was not reported. Controlled non‐randomised studies were to be similarly assessed, with analysis of whether:

• the treatment and comparison groups were comparable at baseline;

• the analysis was adjusted for potential confounding factors.

If as a result of this assessment we considered it likely that the study was biased, we would not include it in meta‐analysis.

Assessment of heterogeneity

We aimed to calculate statistical heterogeneity using the I² statistic, considering an I² value of > 50% as evidence of substantial heterogeneity, and visually inspect forest plots for heterogeneity. As we expected a certain degree of heterogeneity, we planned to use a random‐effects model for meta‐analysis.

Assessment of reporting biases

If at least 10 studies were included, we would draw funnel plots of treatment effect versus precision with the data from all studies (Higgins 2011). We would visually inspect the funnel plots to assess whether there had been selective reporting of outcomes.

Data synthesis

If outcomes differed amongst included studies, we would pool them if measuring the same construct, or report on each outcome systematically for those not measuring the same construct.

We would perform a meta‐analysis if we included two or more RCTs with a low risk of bias in which the study population, intervention, and outcome measures were comparable. We would present this in a ‘Summary of findings’ table, and would include the primary and secondary outcomes listed above. For each outcome, we would report the number of participants, the overall quality of the evidence according to the GRADE approach, and the effect size. If a meta‐analysis was not possible, we would synthesise the findings of the included studies in a table, employing the GRADE levels of evidence (Higgins 2011). We would report the individual effect sizes of the studies and 95% CI. We would use Review Manager 5 for analyses (Review Manager 2014).

Subgroup analysis and investigation of heterogeneity

If we included a sufficient number of studies, that is at least two for each subgroup, we would perform subgroup analysis by antidepressant class, tumour histology, and WHO grade of tumour.

Sensitivity analysis

We aimed to conduct sensitivity analyses by excluding studies that did not report adequate: (i) concealment of allocation or (ii) blinding of the outcome assessor.

Results of the search

A PRISMA diagram displaying records identified and screened is shown in Figure 1.

Included studies

We found no eligible studies for any of our primary or secondary outcomes.

Excluded studies

Studies that were of interest to the research question but did not meet the inclusion criteria of the review are as follows.

Knudsen‐Baas 2018 conducted a cohort study reporting the prevalence and associations of psychotropic medication, including antidepressants, in a Danish population‐based study. The authors did not test the efficacy of any antidepressant interventions, or measure specific side effects.

Boele 2013 tested a psychostimulant (modafinil) in 37 participants with primary brain tumours. In this RCT, the researchers screened for fatigue, which was the primary outcome. Depression was included as a secondary outcome, but participation was not limited to depressed people.

Blackhall 2009 conducted a similar study looking at modafinil to treat cancer‐related fatigue in multiple cancers, and also looked at depression along with neurocognitive function as a secondary outcome. The authors reported some improvement in anxiety and depression, but participation in the study was not limited to depressed people.

Wirz 2010 also conducted an intervention study testing modafinil for reducing fatigue in people with cancer who were undergoing both treatment and pain relief, but did not select people with depression. No people with primary brain tumours were included.

Gehring 2012 compared two psychostimulants (methylphenidate and modafinil) in 24 people with primary brain tumours in an RCT. Depressive symptoms were assessed as a secondary outcome only, and participation was not limited to depressed people.

Maschio 2013 conducted an RCT to determine the effect of pregabalin on seizure control in people with brain tumour. However, depression was a secondary outcome only, and participants did not have to suffer from depression to be included.

Walker 2010 looked at the benefit of using tricyclic antidepressants for mortality reduction in people with glioma or colorectal cancer. They did not investigate management of depression, but found that tricyclic antidepressants did not improve mortality.

Attia 2012 conducted a phase II trial to investigate whether the botanical product Ginkgo biloba had any effect on improving cognitive functioning in irradiated people with brain tumours. Mood (depression) was included as a secondary outcome only, and participants were not selected based on an existing diagnosis of depression.

Gothelf 2005 conducted a pilot study to assess safety and tolerability of a selective serotonin re‐uptake inhibitor (SSRI) antidepressant (fluvoxamine) in children with cancer and comorbid depression or anxiety disorders (Gothelf 2005). The treatment was well tolerated and appeared to reduce depressive symptoms. Participants did not include adults, and only one participant was diagnosed with a brain tumour.

Wellisch 2006 conducted an RCT of a psychostimulant (methylphenidate) in people with brain tumour, with depression as a secondary outcome. However, it was not clear from the published conference abstract whether participation was limited to people with depression, and we were unable to contact the authors for clarification.

Morrow 2003 reported an RCT of an antidepressant (paroxetine) in a mixed population of people with cancer, including those with brain tumours. Although depression was measured as an outcome, the intervention was given primarily to treat fatigue, and participation was not limited to people with depression. Data for the small number of participants with brain tumours were not reported separately.

Moss 2006 conducted an open‐label phase II trial of an antidepressant (bupropion) on fatigue in a mixed population of people with cancer, including those with brain tumour, and measured levels of depression. However, participation was restricted to people with high levels of fatigue, not depression.

Shaw 2006 enrolled 35 participants with brain tumours to an open‐label phase II trial of an acetylcholinesterase inhibitor (donepezil) and measured depression using a rating scale. However, participation was not limited to people with depression.

Meyers 1998 evaluated the effects of a psychostimulant (methylphenidate) in an open‐label phase II trial of 30 participants with high‐grade glioma, measuring depressive symptom severity amongst other outcomes. However, the authors explicitly excluded individuals with major depression at study baseline.

Litofsky 2004 conducted a prospective cohort study of adults with high‐grade glioma, examining the relationship between depression and survival in glioma. However, the authors did not systematically prescribe treatment for depression and did not limit participation to people with depression.

Koval'chuk 2007 conducted a prospective cohort study of 225 participants following operation for brain tumour. The authors measured depression amongst other outcomes, and assessed the influence of different antidepressants on the postoperative normalisation of mood. However, again, participation was not limited to people with depression.

Caudill 2011 conducted a retrospective cohort study of the frequency and toxicity of SSRI prescription in 160 people with glioblastoma presenting to a tertiary neuro‐oncology service over a 10‐year period. The authors reported that being prescribed SSRIs was associated with improved survival at two years postdiagnosis, after controlling for relevant confounders. Participants being prescribed an SSRI reported similar levels of toxicities as those on no SSRI. However, antidepressant prescription was uncontrolled, and the study was not limited to people with depression.

Other studies that were ineligible on methodological grounds were a small uncontrolled case series, Rabey 1985, and a single case study, Oyewumi 1981.

Ongoing studies

We identified one ongoing study. The PASSION trial is testing the effect of ketamine on perioperative depressive symptoms in people undergoing brain tumour resection. Only adults with moderate to severe depressive symptoms will be recruited. The trial was scheduled to finish in February 2019, but to date no results have been published (Zhou 2018). We contacted the authors but have not received a response.