Epileptogenicity of white matter lesions in cerebral small vessel disease: a systematic review and meta-analysis

We systematically searched the Embase and MedLine databases using OVID on February 3, 2023. The full search strategy can be accessed in the supplemental material (Table E1, online only). The literature search included studies published from January 2003 to February 2023, and was restricted to studies in English language. Editorials, case reports, comments, and review articles were not included in this systematic review and meta-analysis. As several research questions were epidemiological, we considered observational studies as eligible. To answer our research questions, all studies needed a control group consisting of patients without seizures/epilepsy (research question 1), without WML (research question 2), or without pharmacological treatment (research question 3).

Duplicates in the search results were removed through a combination of automated and manual methods using EndNote. Two authors (JID and JMH) independently screened titles and abstracts. The same authors then assessed the full-texts of potentially eligible studies to decide on inclusion. Discrepancies in the assessment of eligibility were resolved by consensus between the authors.

In a next step, we extracted data on study characteristics, patient population, exposure, and outcome using a pre-specified template. The following data were extracted: author, year of publication, study design, sample size, size and properties of case groups and control groups, mean age, sex, quantification method of WML, imaging modality, outcome assessment, and outcomes (presence of WML, lesion load and volume with mean and standard deviation, seizure recurrence rate, effect estimates and 95% confidence intervals (CI) where applicable).

Risk of bias for each study was assessed independently by two of the authors (JID and JMH) using the case–control subscale of the Newcastle–Ottawa Scale (NOS) for assessing the quality of nonrandomized studies in meta-analyses [16]. This score ranges from 0 to 9 with a score of ≥ 7 indicating a study of high methodical quality. The NOS comprises the domains ‘selection’, ‘comparability’ and ‘outcome’. In the domain ‘comparability’, a maximum of two points can be awarded. One point is given if the study controls for the most important risk factor as selected by the raters. Another point is awarded if the study controls for any additional factor. We considered the cardiovascular risk profile to be the most important factor to control for. Discrepancies in the assessment of bias were resolved through consensus. The authors then assessed the overall quality of evidence for each research question using the Grading of Recommendations, Assessment, Development, and Evaluation (GRADE) scale. The GRADE scale is a 4-point rating scale, ranging from ‘very low’ (the true effect is probably markedly different from the estimated effect) to ‘high’ (the authors have a lot of confidence that the true effect is similar to the estimated effect). Observational studies are primarily given a certainty rating of ‘low’ with several factors being able to reduce or increase the certainty of evidence [17].

To obtain relevant missing data, we contacted the corresponding authors of six studies through email. Authors of three studies sent additional data.

In our meta-analysis, we calculated summary Odds Ratio (OR) and 95% Confidence Intervals (CI) for the association of seizures or epilepsy and WML as measured by different WML quantification methods.

For the presence of any WML and the presence of relevant WML (Fazekas scale score > 1 or ARWMC scale score > 2), we extracted frequencies in the epilepsy/seizure groups and the control groups from each study and then performed a random effects meta-analysis for binary outcomes.

For the association between WML volume and epilepsy, we extracted effect estimates from studies if they were provided. If these were not given in the studies, we requested original data from the corresponding authors and calculated adjusted effects estimates for each study with z-scores of WML volume (adjustment for age and sex; inclusion method:enter, p < 0.05 [p in], p < 0.1 [p out], iteration 20, and constant included). We then pooled the log-transformed OR and 95% CIs using a random effects model.

To assess heterogeneity, Cochrane Q-statistics were calculated and quantified using the I² values. We considered an I² value > 50% as significant heterogeneity. We assessed publication bias by visual exploration of Funnel plots and Egger’s test. All statistics were performed using SPSS Version 28 (IBM, Chicago, USA).

A total of 2813 records were screened of which 11 studies were finally included in this systematic review. Figure 1 shows the PRISMA flow chart.

The included studies were published between 2006 and 2022. Ten studies reported WML load of patients with first unprovoked seizure and/or epilepsy as compared to control groups (research question 1) [5‐7, 18‐25]. Two studies compared risk of seizure recurrence in patients with and without WML (research question 2) [18, 21]. We could not identify any study on treatment with ASM in patients with CSVD-related WML (research question 3). The characteristics of all included studies including imaging modality, properties of the control groups, and WML quantification method are summarized in Table 1. The studies included in this systematic review comprised a total of 2983 patients; 862 cases and 2121 controls. According to our quality assessment, three studies were of high methodological quality (Table 1, and for more detail: Table E2, online only).

Studies assessing WML load in cases and controls reported either (a) presence of any WML and/or (b) WML load as assessed by a visual rating scale and/or (c) WML volume. One study reported a not further specified degree of white matter changes and was therefore not included in any of the following meta-analyses [24]. A detailed description of the properties of the epilepsy groups as well as individual results regarding lesion load for each study can be found in the supplementary material (Table E3, online only).

We identified two studies analyzing the association between risk of seizure recurrence and WML [18, 21]. These studies included a total of 256 patients with epilepsy of whom 113 had and 143 did not have WML. The characteristics of these studies are summarized in Table 1.

The first study [21] focused on patients with late-onset epilepsy of unknown or structural etiology. Presence of WML was independently associated with an elevated risk of seizure recurrence (RR 1.76, 95% CI 1.01–3.06). The study showed no association between severity of WML and seizure frequency.

The main focus of the second study [18] was to compare lesion load between patients with epilepsy (age ≥ 18 years, exclusion of patients with cardiovascular risk factors) and a control group; therefore, this study is also part of the meta-analysis on presence of WML. In the epilepsy group, presence of WML was not associated with seizure frequency (p = 0.444). However, the assessment of seizure frequency in this study was not further specified, and seizure frequencies in the case and control groups were not reported. Thus, it was impossible to conduct a meta-analysis of these two studies. Additional information on these two studies can be accessed in the supplemental material (Table E4, online only).

In our systematic review, no studies on the treatment of patients with WML of presumed vascular origin and first unprovoked seizure without cortical structural lesions could be identified.

There was a statistically significant association between (a) seizure or epilepsy and presence of WML, (b) seizure or epilepsy and presence of relevant WML as measured by a visual rating scale, but no association between (c) WML volume and epilepsy.

Given the current evidence, it is difficult to determine, whether this discrepancy is due to methodological or pathophysiological reasons. It is important to mention that there are significant differences in the design of the three studies on WML volume and epilepsy [7, 22, 23]. The only study to show a statistically significant difference in WML volume load was on patients with late-onset epilepsy > 50 years [7]. One study had a much younger population (mean age in epilepsy group: 43 years) [23]. The third study did not assess lesion load at onset of epilepsy. Instead, the authors analyzed MRIs that were performed before the occurrence of the respective first seizures, with the time from MRI to first seizure not being reported [22]. After excluding this study from the meta-analysis, the results regarding association between WML volume and epilepsy remained without statistically significant difference. The differences in mean WML volume between these three studies are striking. To our understanding, this could only partly be explained by differences in methods of WML segmentation. Therefore, we confirmed the WML volumes with the corresponding authors of all three studies to ensure the plausibility of the data. This led to the correction of WML volumes in one study. Still, in the epilepsy group, a 16-fold difference in mean WML volume between the study with the highest and the lowest WML volume remained [7, 22]. Reassuringly, all reported WML volumes are within the range of what has been described previously [28, 29]. Therefore, it might be reasonable to assume that in fact, significant WML as detected by a visual rating scale are more relevant with respect to epilepsy than WML volumes. The Fazekas scale as applied by most authors in the studies included in this meta-analysis focuses on deep WML located > 13 mm from the ventricular surface [27, 30]. In contrast, a WML volumetry takes into account WML in the entire brain including periventricular and juxtacortical WML. It is generally accepted that deep WML differ from WML in other locations like periventricular and juxtacortical regarding histopathology and clinical relevance: Deep and confluent white matter alterations are more likely of ischemic origin and are more often associated with clinical symptoms than periventricular and juxtacortical WML [8, 31, 32]. This distinction is better represented by visual scores than by a global quantitative measurement. Results of one study in which presence of deep and periventricular WML was analyzed separately are in line with this notion: Deep WML were more common in the epilepsy group, while periventricular WML were more common in the control group [19]. In a study comparing first-seizure patients to patients with transient ischemic attacks, both periventricular and deep WML were more present in the seizure group [6]. The same study also showed a statistically significant higher prevalence of juxtacortical lesions in the first-seizure group as compared to the control group (80.5% vs 22%, p < 0.001). Finally, in a study that was considered for eligibility in the systematic review but was excluded due to lack of a control group, in patients with epilepsy, WML mainly affected the temporal lobe [13].

Importantly, there is also heterogeneity in the studies on presence of (relevant) WML as assessed by visual rating scales. This is mainly represented in (a) patient populations and (b) differences in visual WML scoring. Regarding patient populations of these studies, four included patients with late-onset seizures [6, 19, 20, 25] while two had no age restrictions [5, 18], and one study did not report the age of its participants [24]. There were also significant differences within comorbidities of the patient populations, especially with regard to cardiovascular risk factors [18, 19]. Another challenge in comparing studies were differences in WML scoring: The most widely used scale for WML was proposed by Fazekas et al. [27] more than 30 years ago. In this scale, both periventricular and deep WML are scored separately on a 4-point rating scale ranging from 0 to 3. To assess CSVD, usually only deep WML are considered [27]. The so-called age-related white matter changes (ARWMC) or Wahlund scale is an updated version of the Fazekas scale. Again, WML are scored on a 4-point rating scale with slight alterations in scoring. More importantly, in this scale, five different brain regions are scored for each hemisphere separately. Thus, a maximum of 30 points theoretically can be given [26].

Even though most studies included in this meta-analysis refer to the paper introducing the Wahlund scale, they in fact mostly provide a “global” or “deep” WML score ranging from 0 to 3, which is more in line with the Fazekas scale in its original form. Out of the seven studies using a visual rating scale, none applied the exact same scoring method. A standardization of WML rating scales to be used in clinical studies would be highly desirable.

So far, this review has interpreted the assumed association between WML and seizures as unidirectional in the sense that WML might cause seizures. However, also the inverse relationship is worthy of discussion. This point is of particular importance, as none of the studies included in this meta-analysis that focused on epilepsy (as opposed to first-seizure patients), stated that it was on new-onset epilepsy patients. Therefore, we do not have certainty on the lesion load at the onset of epilepsy. Thus, given the current data, we cannot exclude that a significant proportion of WML developed after the diagnosis of epilepsy. From a pathophysiological perspective, it seems unlikely that seizures could directly cause WML [33]. Yet, it is possible that epilepsy indirectly causes WML through ASM therapy. Interestingly, in one study included in this review, WML load was highest in patients receiving enzyme-inducing ASM as compared to patients with non-enzyme-inducing ASM and controls without epilepsy [5]. This was interpreted to be caused by negative effects of enzyme-inducing ASM on cardiovascular risk factors [34].

To answer our research questions, only studies with an appropriate control group were eligible for inclusion. It is, however, important to mention that there was heterogeneity in the properties of the control groups in the studies included in this review. We therefore performed sensitivity analyses, including only studies where the control group consisted of healthy participants. Reassuringly, our results stayed robust for all outcome parameters.

We identified two studies on seizure recurrence risk in patients with epilepsy with conflicting results. One study focused on patients with late-onset epilepsy of unknown or structural etiology. Here, presence of WML was independently associated with an elevated risk of seizure recurrence [21]. The other study did not show any statistically significant difference in seizure frequency between patients with and without WML [18]. It is noteworthy, that this study has extensive exclusion criteria: In essence, patients with cardiovascular risk factors, such as hypertension and diabetes mellitus, were not eligible for inclusion [18]. To us, excluding these patients seems counter-intuitive, especially when considering the presumed pathomechanisms behind the suspected epileptogenic properties of WML, namely the hypothesis of WML being a surrogate parameter of cortical microinfarcts [13, 14]. Moreover, it has been demonstrated that cardiovascular risk factors are higher in late-onset epilepsy patients than in controls [35]. It is therefore plausible that this study did underestimate the effect of WML on seizure recurrence.

Assessing the influence of WML on the risk of seizure recurrence is an important milestone in evaluating their epileptogenicity. Both studies identified in this review were on patients with established diagnoses of epilepsy. As these patients are usually treated with ASM, a more desirable study population would be patients with first seizures not yet treated. Only this population would allow assessing the influence of WML on the rate of seizure recurrence without the risk of bias by ASM.

Ultimately, all aspects discussed above lead to the final, clinically most relevant question, whether ASM treatment is justified or even necessary in patients with a first seizure at the presence of WML, even if there are no cortical lesions in the MRI and the EEG is without epileptiform discharges. Our systematic review did not show any prospective studies on ASM treatment in patients with WML with regard to seizure outcome.

The strength of this review lies in its rigorous adherence to the PRISMA guidelines and its meta-analysis using original patient data where possible.

There are limitations to consider. We have already addressed the significant heterogeneity between studies on WML lesion load. Through several sensitivity analyses, we aimed to diminish its effects, but even so, the certainty of evidence remains impaired. This is especially relevant for the fact that all studies included in this review and meta-analysis were observational studies. Thus, the overall quality of evidence included in this review has to be considered as low. In our opinion, a major limitation in the design of most included studies was the lack of adjustment for cardiovascular risk factors (see Table E2, online only).

Initially, we had set out to compare WML load in first-seizure patients and epilepsy patients separately. However, only one study was identified focusing on first-seizure patients [6]. In two studies, a small number of patients with a single seizure were also part of the respective “epilepsy group”, and neither of them performed a separate analysis [20, 24]. A sensitivity analysis without the study only on first-seizure patients did not significantly alter the results of our meta-analysis.

This review explicitly focuses on WML of presumed vascular origin. It is important to mention that CSVD is more than WML. Other signs of CSVD include lacunar infarcts, enlarged perivascular spaces, micro-bleeds, and brain atrophy [8]. These signs were assessed as secondary outcomes by some studies included in this review [6, 7, 19, 20, 22, 25]. To summarize, it was mainly shown that epilepsy was associated with reduced cortical volume and hippocampal or temporal atrophy [6, 7, 22, 25]. This seems plausible from a pathophysiological viewpoint. Also, CSVD in general (including lacunar infarcts, perivascular spaces, and micro-bleeds along with WML) was more prevalent in patients with epilepsy than in controls [19, 20]. The topic of this review on WML of presumed vascular origin also implicates that we did not look into microstructural alterations of white matter as detected in diffusion tensor imaging studies, which were reviewed elsewhere [36].