Objectives To investigate potential harm and benefits of antiepileptic drugs (AED) given prophylactically to prevent de novo brain tumour-related epilepsy after craniotomy.
Methods Randomised controlled trials (RCT) and retrospective studies published before 27 November 2018 were included. The Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines were applied. Eligible patients were diagnosed with a brain tumour, were seizure naïve and underwent craniotomy. The random effects model was used for quantitative synthesis. The analysis was adjusted for the confounding effect of including patients with a history of seizure prior to study inclusion.
Results A total of 454 patients received prophylactic AED whereas 333 were allocated to placebo or no treatment. Two RCTs and four retrospective studies were identified. The OR was 1.09 (95% CI 0.7 to 1.8, p=0.7, I2=5.6%, χ2 p=0.5), indicating study consistency and no significant differences. An additional two RCTs and one retrospective study combined craniotomy and diagnostic biopsy, and were subgroup analysed—which supported no difference in odds for epilepsy.
Conclusions A prophylactic effect of AED could not be demonstrated (nor rejected statistically). Levetiracetam was associated with less adverse effects than phenytoin. The potential harm of AED was not balanced by the potential prophylactic benefit. This study suggests that prophylactic AED should not be administered to prevent brain tumour-related epilepsy after craniotomy.
- brain tumour-related epilepsy
- brain tumour
- antiepileptic drugs
Statistics from Altmetric.com
- brain tumour-related epilepsy
- brain tumour
- antiepileptic drugs
Tumour-related epilepsy is a frequent manifestation in patients with supratentorial brain tumours. Epilepsy occurs in 50% of patients with brain tumours and is particularly common with slow-growing gliomas, meningiomas located on the convexity of the brain and with metastatic brain tumours. The risk of tumour-related epilepsy is increased after craniotomy and postoperative tumour-related epilepsy is associated with decreased overall survival and delayed neurological recovery.1 2
Antiepileptic drugs (AED) have been administered prophylactically to patients with brain tumours to prevent epilepsy in inoperable patients as well as postoperatively. Phenytoin has been prescribed in management of tumour-related epilepsy for nearly four decades and within the last 20 years levetiracetam has been increasingly used and several studies have investigated its benefits compared with phenytoin.
The evidence for efficacy of prophylactic AED in reducing the risk of tumour-related epilepsy is contradictory with a primary overweight of studies suggesting no benefits. There is an increasing concern that prophylactic AED may cause adverse side effects including delay of neurological recovery and an impairment of the quality of life.3 Results from studies investigating the tolerability of prophylactic AED are inconsistent and evidence-based guidelines remain unestablished. There is a broad consensus on the futility of prescribing AED prophylactically but, in spite of this, AEDs are ubiquitous and massively prescribed to epilepsy-naïve patients during the postoperatively phase—and on follow-up, the majority of these patients have remained on prophylactic AED.4–6
This study presents a comprehensive systematic review and meta-analysis of harm and benefits of prophylactic AED treatment in patients with brain tumour that can serve to support evidence-based guidelines and provide recommendations for further clinical studies. Our main objective was to examine two hypotheses: (1) prophylactic AED treatment prevents de novo brain tumour-related epilepsy after craniotomy in seizure-naïve patients, and (2) administration of AED is associated with adverse and unnecessary side effects that outweigh therapeutic efficacy.
This systematic review and meta-analysis adhered to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines.7
PubMed, Embase, Ovid and Cochrane were searched from 1 January 1966 to 27 November 2018. The search was performed in two steps in order to identify studies specifically examining the prophylactic effect of AED and specifically examining adverse effects of AED. The preliminary search identified 243 publications on PubMed, 762 publications on Embase, 2027 publications on Ovid, and 2 reviews and 19 trials on Cochrane. The search string was based on the terms ‘brain tumor’ or ‘brain neoplasm’ or ‘central nervous system tumor’ or ‘brain cancer’, and ‘seizure’ or ‘epilepsy’, and ‘antiepileptic’ or ‘anticonvulsant’ and ‘prophylactic’ or ‘prophylaxis’. By replacing ‘prophylactic’ or ‘prophylaxis’ with the terms ‘side effects’ or ‘adverse effects’, additionally 360 publications were identified on PubMed, 1313 publications on Embase, 2272 publications on Ovid and 9 publications on Cochrane. We restricted the search to retrospective studies, clinical trials, randomised controlled trials (RCT), systematic reviews and meta-analyses. We applied no language restriction. Reference lists of eligible studies and related meta-analyses were manually evaluated to identify further relevant studies (figure 1).
Eligible studies specifically investigated patients undergoing craniotomy. A total of nine studies were identified, of which three studies had a mixed patient group of patients undergoing craniotomy and biopsies (the ‘mixed surgical intervention’ group). All patients must be investigated in relation to prophylaxis and de novo brain tumour-related epilepsy. Hence, studies that combined seizure-naïve patients with patients experienced epilepsy prior to study onset were excluded from quantitative synthesis. Eligible studies were allocated: first, studies were categorised as either a randomised trial or as a retrospective study; and second, allocation of studies to a subgroup of either craniotomy exclusively or the ‘mixed surgical intervention’ group.
The primary AED was often used in combination with other AED to clinically manage brain tumour-related epilepsy, which was noted and accounted for in the statistical analyses.
All study titles were screened and excluded if irrelevant. Abstracts were read and excluded if in violation of the inclusion criteria. Patient and study characteristics were extracted, including type of AED, medical history of previously prescribed AED prior to study inclusion, medical history of epilepsy prior to study inclusion, tumour histology (benign, malignant or combined), continent, age range and median or mean age of the participants on time of study inclusion (table 1, figure 1).
For all studies investigating the prophylactic effect of AED, all events of seizure(s) were extracted for the treated group and the placebo (or no treatment) group. The events were used to calculate OR for each study with associated 95% CIs.
Data on adverse effects were noted when examining the studies included in this study. Data on adverse effects were only extracted when explicitly stated and have not been self-interpreted by the authors. Studies were allocated to intervals reflecting duration of follow-up: less than a month, between 1 and up to 12 months and exceeding (or equal to) 12 months.
It was difficult to assess adverse effects based on the included studies. There was no common mode to describe and report adverse effects. The reported adverse effects were subject to a subjective clinical assessment. Consequently, our primary objective was not to provide an objective measurement of the risk of adverse effects, but rather to report and describe the observed adverse effects in the included studies.
When examining adverse effects, study characteristics were extracted, including type of AED, drug dosage and length of follow-up time. Studies reported adverse effects as either an adverse effect or a severe adverse effect. Adverse effect was considered as nausea, paraesthesia, nystagmus, headache, somnolence and rash. Severe adverse effects included psychosis, pulmonary embolism, Stevens-Johnsons syndrome and organ failure.
A random effects model, which acknowledges the existence of different effects sizes underlying different studies, was used in this analysis. We adopted the ‘restricted maximum likelihood’ estimator in the random effects model based on meta-analytic studies comparing bias and efficiency of meta-analytic variance estimators in random effects models.8 9
I2 quantifies the proportion of variance in study effect estimates, which is attributable to heterogeneity rather than chance. Thus, an I2 value of 0% correlates with no inconsistency between studies. Heterogeneity was quantified according to Higgins et al, with ‘low’, ‘moderate’ and ‘high’ corresponding to I2 values of 25%, 50% and 75%, respectively.10
The p value for χ2 test was computed to determine whether significant heterogeneity existed.
Statistical analyses were performed in R Studio. This meta-analysis and its graphical content were made by using the ‘metafor’ package.11 Bias was assessed according to the Cochrane Collaboration’s tool for assessing risk of bias in randomised trials.12
In total, 76 publications were eligible for full-text review of which six studies investigated the prophylactic effect of AED in seizure-naïve patients after craniotomy and met the inclusion criteria for quantitative synthesis. Three additional studies did not discriminate between craniotomy and diagnostic surgery, that is, biopsy, and were subgroup analysed concurrently. A total of 15 studies addressed the adverse effects of AED and met the inclusion criteria for quantitative synthesis.
The main studies comprised two randomised trials13 14 and four retrospective studies5 15–17 with a total of 787 eligible patients. Among these, 454 (58%) patients were treated with AED and 333 patients (42%) were given placebo or no treatment. The median follow-up time was 12 months, ranging from 1 to 12 months. The mean follow-up time was 9.2 months (table 1).
The three studies eligible for ‘mixed surgical intervention’ subgroup analysis comprised two randomised trials18 19 and one retrospective study,20 with a total of 286 patients. Among these, 158 patients (55%) received prophylactic AED whereas the remaining 128 (45%) were allocated to placebo or no treatment. The follow-up was 12 months in all studies.
A total of 1073 patients—612 (57%) received prophylactic AED and 461 (43%) received placebo or no treatment—were eligible for quantitative synthesis.
Random effects model based on study design: randomised trials versus retrospective studies
In a random effects model, the overall OR estimate was 1.09 (95% CI 0.69 to 1.75, p=0.7) indicating no significant difference in odds for epilepsy in the AED group versus placebo or no treatment. The overall I2 percentage was 5.6% and considered low. The χ2 p value for all studies combined was 0.44, indicating that no significant heterogeneity was observed. Similarly, the between-studies variance, τ 2, was 0.02 and considered low. Thus, the inconsistency across the studies was low and insignificant (figure 2A).
The estimated OR for randomised trials was 1.02 (95% CI 0.38 to 2.76, p=1), indicating no difference in odds for epilepsy between the group treated with prophylactic AED and the placebo or no treatment group (figure 2A). The estimated odds for epilepsy in the retrospective studies was non-significantly 22% higher for patients treated prophylactically with AED: 1.22 (95% CI 0.69 to 2.17, p=0.5) compared with the placebo or no treatment group.
The I2 percentage was 46.4% and 0% for randomised trials and retrospective studies, respectively, and considered moderate and low. The χ2 p value was 0.17 and 0.45 for randomised trials and retrospective studies, respectively, indicating that no significant heterogeneity was observed. Similarly, the between-studies variance, τ 2, was 0.02 and considered low (figure 2A).
Random effects model based on magnitude of surgical intervention: craniotomy and biopsy
The search identified three additional studies that did not discriminate between surgical procedures and combined patients undergoing craniotomies with patients receiving biopsy exclusively.18–20 The studies comprised 286 patients, of whom 158 received AED and 128 placebo or no treatment.
In a random effects model, the overall OR estimate was 1.03 (95% CI 0.73 to 1.46, p=0.8) indicating no significant difference in odds for epilepsy in the AED group versus placebo or no treatment. Generally, the homogeneity increased when including ‘the mixed surgical intervention’ group compared with the random effects model in figure 2A. The overall I2 percentage decreased to 4.4% and was considered low. The χ2 p value for all studies combined was 0.44, indicating that no significant heterogeneity was observed. Similarly, the between-studies variance, τ 2, was 0.01 and considered low (figure 2B).
The estimated OR for studies including craniotomy exclusively was obviously identical to the overall estimate from the random effects model in figure 2A.
The estimated odds for epilepsy in ‘mixed surgical intervention’ group was non-significantly 5% lower for patients treated prophylactically with AED: 0.95 (95% CI 0.52 to 1.76, p=0.9) compared with the placebo or no treatment group. The I2 percentage was considered low to moderate at 28.2%, but not significant with a χ2 p value of 0.23. Similarly, the between-studies variance, τ 2, was 0.08 and considered low (figure 2B).
A funnel plot of the random effects model was constructed (figure 3). It visualises that the three additional studies of ‘mixed surgical intervention’ were consistent with the ‘craniotomy exclusively’ studies.18–20 Further, a test for model moderators, in essence testing whether the study characteristics (eg, craniotomy vs mixed surgical intervention—table 1) account for any of the heterogeneity, was not overall significant based on the omnibus test (QM=3.0, df=5, p=0.7): hence, the three studies of mixed surgical intervention did not account for any of the heterogeneity but added qualitatively to the model.18–20
Studies that mix seizure-naïve with seizure-experienced patients yield erroneous estimates
The search identified five studies that combined seizure-naïve patients with patients who had a history of seizures prior to study but underwent all craniotomy.2–4 21 22 The seizure-experienced patients were either medicated on AED or not. The five studies were added to the latter random effects model (figure 2B) to test the theoretical impact on effect estimates in meta-regression analysis. Study characteristics equivalent to those listed in table 1 were extracted. Studies including patients with a history of seizures had significantly 6.2 times (95% CI 1.3 to 29.0, p=0.02) higher odds for epilepsy than studies including seizure-naïve patients exclusively.
Assessment of risk of bias
No randomised trials were assessed with ‘low risk of bias’ exclusively, and only two of the four randomised trials were not associated with ‘high-risk of bias’.13 18 The primary cause of ‘high-risk of bias’ was due to none or incomplete blinding of key staff, participants and outcome assessors (performance and detection bias).14 19 The length of follow-up was considered as potential selective reporting bias, for example, one study compared the end outcome between a follow-up of 0.31 month and 30.1 months19 (table 2).
Adverse effects of levetiracetam and phenytoin
A total of 18 studies (three dealt with both levetiracetam and phenytoin) addressed adverse effects of levetiracetam and phenytoin either as a primary objective or as observations during treatment. Because only a few studies addressed the association directly,20 23–30 most data were identified as observations or as remarks in tables in studies investigating other primary objectives.
The follow-up time for patients varied extensively across all 15 studies. Two studies addressed the prophylactic effect on early postoperative epilepsy (within 3 days),3 22 28–30 six studies evaluated an intermediate period (1–6 months)17 24 26 31–33 and seven of the studies evaluated the long-term prophylactic effect of at least 12 months.14 19 20 25 27 34 35 The follow-up time did not significantly moderate study heterogeneity.
A total of 653 patients treated with levetiracetam were observed regarding occurrence of adverse effects. Of these, 49 patients (7.5%) experienced adverse effects: 44 (minor) events (89.8%) and 5 severe adverse effects (10.2%). Four of the five severe adverse effects occurred in patients allocated to a follow-up of minimum 12 months (table 3, figure 4A).
Similarly, a total of 304 patients were treated with phenytoin. Adverse effects were reported in 48 patients (15.5%). Of these, 39 cases (81.2%) were (minor) adverse effects and 9 cases (18.8%) were severe adverse effects. Thus, 2.9% of the patients treated with phenytoin experienced a severe adverse effect. All nine cases occurred in patients allocated to a follow-up of 12 months (table 3, figure 4B).
This study failed to show any superior benefit from prophylactic AED in management of brain tumour-related epilepsy compared with placebo or no treatment, but AED may cause potentially severe adverse effects.
Six studies, two RCTs and four retrospective studies, consistently (I2 of 5.6%) estimated that AED treatment does not exert a prophylactic effect in prevention of de novo brain tumour-related epilepsy after craniotomy. The finding is further supported when adding three studies that combined craniotomy and biopsies equivalently. This reduced the overall heterogeneity (I2 of 4.4%) and the overall estimate dropped from 9% to 3% in favour of the placebo or no treatment group. Although not significant, this might be explained by the fact that surgical resection (in relation to craniotomy) is an independent risk factor of inducing epilepsy.36
Our analysis also indicated that studies including patients with a history of seizure(s) prior to study onset found a significant higher risk of epilepsy compared with studies of seizure-naïve patients exclusively. This finding was explained by the confounding effect of prescribed AED predisposition to epilepsy prior to study inclusion—which was the case for five studies.2–4 21 22 The confounding effect might be explained as follows: patients who were previously medicated with AED have already had epilepsy and were consequently predisposed to recurrent epileptic events. Thus, the prophylactic aim for these patients was to prevent recurrence of epilepsy rather than to prevent de novo epilepsy occurrence. This emphasises the importance of balanced baseline study characteristics, which future studies should implement to avoid erroneous effect size estimates. This was illustrated in a recent meta-analysis by Joiner et al, which concluded a statistically significant reduction in risk of early postoperative seizures in favour of the AED group. However, Joiner et al came to this conclusion by analysing (random effects model) four RCTs of which one (Franceschetti et al 21) mixed seizure-naïve patients with those who had a history of seizure(s) prior to study inclusion.37 The lack of adjusting to this confounder might therefore explain this solitary finding.
The methodically most trustworthy study designs would be of randomised trials, which was the case of four studies.13 14 18 19 However, only the studies by North et al and Glantz et al published in 1983 and 1996, respectively, were not associated with a high risk of bias but were unfortunately outdated.13 18 The studies of Wu et al and Forsyth et al were associated with high risk of bias as the participants nor the key staff were blinded (table 2).14 19
The next step to conclusively close this long-standing, but important, debate is to determine whether a cohort of seizure-naïve patients would benefit from a newer generation AED given the potential risk reduction from the 95% CI bands in a randomised, double-blinded, placebo-controlled trial. It is possible that either a lower frequency of side effects or targeting of a well-defined high-risk subgroup may change the estimates of benefit of AED.
Fixed effects model or random effects model?
The fixed effects model assumes that there is one true effect size that underlies all studies in the analysis. With this assumption, any observed heterogeneity (inconsistency between studies) is considered to reflect sampling error. On the contrary, a random effects model accepts and assumes a distribution of true effects that vary between studies.
Several considerations make the fixed effects model inappropriate. The prophylactic AEDs investigated in this meta-analysis comprise an inconsistent group of AEDs, which was dominated by phenytoin and levetiracetam. However, 9 of the 15 studies used combinations of multiple AEDs to manage tumour-related epilepsy: pregabalin, carbamazepine, lamotrigine, oxcarbazepine and valproate. Further, phenytoin has well-known major complicated pharmacokinetic interactions with multiple drugs including chemotherapeutic agents, which is important as patients with malignant brain tumours may have received chemotherapy concomitantly.38 39
It follows that the fixed effects model is inappropriate for meta-analyses of the various AED studies and that the random effects model should be used. Consequently, the statistical method applied in the recent meta-analyses by Pourzitaki et al 40 and Kong et al 41 is questionable.
Pourzitaki et al applied a fixed effects model meta-analysis on three studies to determine the effect of levetiracetam versus phenytoin on postoperative epilepsy and reported an OR estimate of 0.12 (95% CI 0.03 to 0.42, I2=0%, p=0.4) favouring levetiracetam. Kong et al used a fixed effects model for a meta-analysis on six studies (some included in this study14 21–23 25) and reported an OR estimate of 0.94 (95% CI 0.61 to 1.45, I2=0%, p=0.4).
Adverse effects of AED
Levetiracetam caused fewer adverse effects than phenytoin and has much simpler pharmacokinetics. Levetiracetam is primarily metabolised in the kidneys where CYP enzymes are not involved; hence fewer pharmacological interactions.42 Generally, the studies included in this meta-analysis associated levetiracetam with greater tolerability, fewer adverse effects and without drug–drug interactions. In a systematic review, Usery et al identified 92 drug interactions which complicated treatment with phenytoin. The most interactions included dexamethasone, acetaminophen and fentanyl, which all represent potent CYP450 inducers that significantly affect phenytoin concentrations.26
Studies on the adverse effects of AEDs have exclusively reported the frequency and severity of adverse effects.20 23–27 However, this parameter is not relevant when isolated from the intended benefit; it is reasonable to balance benefit with risk. Adverse effects may be acceptable for a large therapeutical and prophylactic benefit, while minimal or insecure benefit does not justify any level of risk. Evidently, a prophylactic effect of AED was not seen in these patients. Taken together, the adverse effects did not appear to balance any therapeutic benefit.
This meta-analysis, using a statistical random effects model, compared 1073 patients with prophylactic AED treatment (57%) with patients given placebo or no treatment (53%) and found no significant therapeutic benefits of prophylactic treatment with AED in patients with brain tumours. Odds for patients with brain tumour-related epilepsy after craniotomy was ranging from 31% lower to 75% higher in the AED treated group. Adverse effects occurred in 7.5% of patients treated with levetiracetam and 15.5% of patients treated with phenytoin, which is unacceptable since no significant effect could be demonstrated.
Future well-conducted controlled prospective studies are needed to identify risk-specific factors for tumour-related epilepsy and to identify the most appropriate AED if prophylactic treatment is needed.
This project has been reviewed by the Statistical Advisory Service (Section of Biostatistics) at the University of Copenhagen.
Contributors CM led the writing and statistics. MMP assisted in writing and developing the method. The project was supervised by senior physicians TM and AS, who also assisted in writing.
Funding The authors have not declared a specific grant for this research from any funding agency in the public, commercial or not-for-profit sectors.
Competing interests None declared.
Patient consent for publication Not required.
Provenance and peer review Not commissioned; externally peer reviewed.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.