Eligibility for inclusion in the meta-analysis include peer-reviewed publications in which full length English language manuscripts were available through electronic indexing comprising: a) clinical studies of patients with temporal lobe epilepsy, b) undergoing open epilepsy surgery as a treatment, or c) undergoing RF-TC, SRS or LITT as a treatment, d) with greater than 10 patients in the intervention arm and e) follow-up duration of ≥12 months. Due to previous meta-analyses of open surgery outcomes (16, 17) only comparative open surgical studies were included to provide the highest levels of evidence. Studies that did not report Engel or ILAE outcomes or similar were also excluded.

Using the PRISMA guidelines (15) a structured search of the PubMed, Embase and Cochrane databases were undertaken. The last date of the search was September 9^th 2020.

Two independent researchers (VNV and FX) applied the search criteria defined following a PICOS (participants, interventions, comparators, outcomes and study design) approach to identify search terms. Participants included human studies reporting surgical treatments for drug resistant focal epilepsy. Surgical interventions included in the search terms were open surgical techniques including selective approaches and the numerous variations, as well as minimally invasive alternatives such as LITT, RF-TC, SRS and high frequency ultrasound. The main comparator for the quantitative analysis was seizure-free outcome.

The search terms can be found in the Supplementary Information.

The reference lists of all studies were searched and cross-references for additional eligible studies considered. Previous systematic reviews and meta-analyses were also screened to capture additional studies (9, 16–18) not identified by the search terms.

During full text review data extraction was performed using a table with a predefined set of criteria including level of evidence, study design, comparison/control group and sample size. At a participant level data relating to side of surgery, patient age, duration of epilepsy, pre-operative MRI findings, post-operative histological findings, follow up duration, seizure outcome grading scale, post-operative seizure outcome, independent predictors of surgical outcome, neuropsychological tests performed, neuropsychological outcome, psychiatric outcome, operative morbidity and operative mortality was recorded. The main outcome for meta-analytic comparison was seizure-free outcome.

Anticipated sources of bias affecting seizure freedom included the seizure outcome grading scale and the duration of follow-up. To mitigate outcome reporting for different grading scales only comparable seizure outcomes were included in the quantitative analysis and where sufficient information was provided outcomes were converted from one grading scale to another; e.g., Engel class 1(A-D) = ILAE class 1, 1a and 2. The issue of overlapping patient cohorts in different papers also required careful analysis of the subjects so that duplicate cohorts were not included in the quantitative analysis. Where duplicate cohorts were unclear, authors were contacted for clarification.

Methodological Index for Non-Randomized Studies (MINORS) (19) and Jadad (20) scores for randomized control trials, respectively, were calculated independently by KM and KD. Low scores suggest less methodologically sound studies.

For each surgery type, pooled estimates of the seizure-free rate per person-year were calculated, together with 95% confidence intervals, using a meta-analysis with inverse variance weighting.

The seizure-free rate per person-year estimates were calculated using the following equation to allow for the varying follow up durations between studies. As all patients are seizure-free immediately following intervention the time at risk of seizure-relapse can be taken as the total follow-up duration of the study with Person-years representing the sum of the total time at risk (total number of patients x follow up duration of study). This approach allows studies with differing follow-up times to be included in the meta-analysis and yields an informative method to compare studies having accounted for follow-up duration. In contrast, restricting the analysis to include only studies where follow-up is of the same length may result in selection bias and loss of information.

A mixed-effects linear regression model was then performed with rate of seizure freedom per person-year and surgery method as a covariate to compare seizure-free rates between surgery types. A random effect for study was included in this model. Statistical analyses were performed using R (version 4.10).

Where provided, patient demographics regarding side of resection, age at operation, and duration of epilepsy were comparable between open surgical interventions (Table 1). Overall, 86.5% of patients had MTS on pre-op MRIs and 79.5% had histologically proven MTS from 14 studies.

Nineteen studies comparing open surgical outcomes were included in the meta-analysis (Table 2). ATLR was compared to transcortical SAH in 63% (12/19), to trans-sylvian SAH in 26% (5/19) and to an unspecified SAH technique in 11% (2/19). Of the studies comparing ATLR with transcortical SAH, 25% (3/12) were prospective whilst the remaining 75% (9/12) were retrospective. One pilot RCT compared ATLR with transcortical SAH and transcortical parahippocampal gyrus resection (22). The study sample size was small and being a pilot study was underpowered to detect differences in Engel 1a outcome between any of the three techniques at one- and five-year follow-up. Despite this, visual field defects were noted to be significantly worse in the ATLR and SAH groups compared to the parahippocampectomy group.

One study, an RCT, reported the SAH group having a significantly lower seizure free outcome (p = 0.013) at 12 months (23). The remaining four studies did not report any significant differences between the two techniques including one paper (24) which was a subset of patients from Mackenzie et al. (25). Two studies did not specify the type of SAH performed so were excluded from the quantitative analysis for SAH (25, 26). One study evaluated a pediatric cohort (mean age 10.6 years) and found that ATL had a significantly better seizure outcome (p = 0.017) after a mean of 5.3 years (27). Studies comparing ATLR with radiosurgery (11) and with radiofrequency (28) were also included in the meta-analysis. In summary, of all the comparative open surgical techniques, two studies provided class one evidence (22, 23), one class two (29) and the remainder class three. A total of 2,183 patients were included in the quantitative synthesis with 1,248 undergoing ATLR and 935 undergoing SAH (transcortical 516, trans-sylvian 354 and unspecified 65).

Meta-analysis revealed a pooled effect size for overall seizure free rate per-person year of 0.70 (95% CI 0.64–0.77) for ATLR (Figure 2A), 0.60 (95% CI 0.49–0.73) for transcortical selective amygdalohippocampectomy (Figure 2B) and 0.72 (95% CI 0.66–0.79) for trans-sylvian selective amygdalohippocampectomy (Figure 2C).

Neuropsychological outcome was reported in 16 studies. Five comparative open surgery studies investigated changes in intellectual status after surgery. Across these studies, intelligence either remained stable (27, 30, 31) or improved in both the visuospatial and verbal domain (32, 33). The improvement occurred as a combined effect of side and type of intervention, with mixed findings across studies. No post-surgical decline in intelligence was found.

Four studies investigated language functions. Language was reported to either remain stable (30, 31) or decline (34, 35) after surgery, but no study reported post-surgical improvement. Decline in language abilities were observed irrespective of side or procedure (35) with greatest declines reported following left-sided resections and ATLR (34).

Nine studies which looked at open surgery only investigated visuospatial memory and 13 investigated verbal memory. Four studies reported no post-operative changes in visuospatial memory whereas for verbal memory, just over half of studies suggested a decline (21, 24, 29, 30, 33, 35, 36). Post-operative improvements in memory are reported for functions subserved by the non-operated side, with improvements in visuospatial memory occurring after left-sided resections (33, 34) and verbal memory gains after right-sided resections (32). Intervention-specific improvement of memory is also observed, with greater improvement after SAH than after ATLR, particularly for visuospatial memory (21, 33, 34). Visuospatial memory also declined after right-sided resection, irrespective of the intervention type (34) or with greater decline after ATLR compared to SAH (33). The factors influencing verbal memory decline varied across studies, with either a decline irrespective of side or type of resection (35), after left-sided resection irrespective of the intervention type (21, 29), or conflicting findings regarding the effect of intervention [i.e. with more decline after SAH (30, 33) or ATLR (24, 29, 36)].

Four studies were included in the quantitative analysis (12, 28, 39, 40). Where reported, the mean age was 33.6 years and the mean duration of epilepsy was 20.3 years, with some studies reporting median ages only. 64 patients underwent left-sided procedures, and 37 patients underwent right-sided procedures. Laterality was not reported in one study (28). There was unclear reporting of the exact number of patients with MTS on MRI, however it was present in the majority. Histopathological confirmation of MTS was not available as this is a lesioning technique.

One study (39) was prospective, one study did not specify (12), and the remainder were retrospective. The range of reported seizure freedom was between 0 and 76.2% seizure free at 12 months (Table 3). An single study which followed an overlapping patient cohort with Vojtech et al. (40), compared RF-TC to ATLR and reported no difference in Engel 1 outcomes of 79.3% (23/29) and 76.5% (13/17) at 5 year follow up, respectively (41). In contrast, a later study comparing RF-TC to ATLR (28), reported none (0/21) of the patients in the RF-TC group were seizure free whereas 75.5% (37/49) of the patients in the ATLR group had an Engel 1a outcome at 12-month follow up (p < 0.0001). Overall, 122 patients from studies equating to level two, three and four evidence were included in the quantitative synthesis. A random-effects meta-analysis yielded a pooled seizure free rate estimate per person-year of 0.38 (95% CI 0.14–1.00) (Figure 2D).

Neuropsychological outcome was reported in one of four studies. This single study suggested better post-operative neuropsychological outcome in the RF-TC group (41) compared with the ATLR group.

Complications were heterogenous across the studies. 4.1% (5/122) patients had haematomas, which were mostly asymptomatic (12, 40). Other complications included transitory anosmia (1/122), upper quadrantanopia (1/122), meningitis (2/122), pulmonary embolism in a patient with hereditary coagulopathy (1/122), and an asymptomatic retention of an electrode fragment (2/122). No complications or mild headache were reported in 36.1% (44/122) patients (40). The most serious early complications were a small intracerebral haematoma and hydrocephalus which resolved with ventricular drainage (40). Reported mortality was 5% (3/61) in the report by Vojtech et al. (40), due to one suicide, an unrelated accident and extracranial malignancy. Mortality was 0% in all other studies. No mortality was directly related to the operative procedure. Psychiatric complications were reported in 6/61 patients (40).

Seven studies (three prospective) comprising 133 patients were included in the meta-analysis (Table 4). Two patients in the paper by Rheims et al. (42) and six patients in the paper by Vojtech et al. (43) overlapped with those of the Marseilles cohort (44) and were excluded from the analysis. The mean duration of epilepsy was similar between studies, with an overall mean of 24.1 years. Between 93 and 100% had MTS on MRI scan. Where laterality was reported, 48 procedures were left-sided and 40 were on the right.

Due to the delay in treatment benefit following SRS, seizure free outcomes were reported from 24 months. All studies reported on the use of the gamma knife (Elekta AB, Sweden). One study compared high (24Gy) and low (20Gy) dose SRS, reporting improved seizure free outcomes but alongside increased minor complications associated with use of the higher dose (45). The highest level of evidence was from the ROSE randomized control trial comparing SRS to ATLR (11). The trial was terminated early due to poor recruitment, but even at this stage revealed that seizure free outcome rates following ATLR were superior to SRS (p = 0.82 at 15% non-inferiority margin). Most studies provided level three and four evidence with two papers providing level two evidence (44, 45) and a single RCT providing level one evidence (11). The pooled seizure-free rate per person-year was 0.50 (95% CI 0.34–0.73) after a treatment delay of 24 months (Figure 2E). Neuropsychological outcome was reported in 5/7 studies. One study directly compared outcome between SRS and open surgery (ATLR) and showed less risk of verbal memory deterioration after SRS, and less risk of visuospatial memory deterioration after ATLR (11). Another neuropsychological study reported decline in verbal memory with high dose (24 Gy) but not low dose (20 Gy) Gamma Knife to the dominant amygdalohippocampal complex (46). Other studies either reported some improvement in a subset of the patients or little post-operative cognitive change.

Two studies (49, 50) reported duplicate patient cohorts and were excluded from the analysis. From the 11 studies reporting LITT outcomes, two were prospective (51, 52) and nine were retrospective (see Table 5). Overall, 520 patients were followed up, ranging between a mean of 12 months (13, 53–56) and 43 months (57). Engel 1 outcomes for the studies ranged between 44% (14) and 67.4% (58). The mean age of patients was 41.9 years, mean duration of epilepsy was reported in nine studies as 23.8 years, an average of 74.2% patients undergoing LITT had MTS on MRI, and the side of resection was reported by nine studies as 263 (left) and 201 (right).

A retrospective case series revealed that trajectories with a medial course through the hippocampal head resulted in less mesial hippocampal head remnant and was associated with improved seizure free outcome (53). Seizure freedom rates were found to be twice as high in patients with MTS compared to non-MTS (13), although a separate study reported similar seizure freedom rates between MTS and non-MTS when stereoelectroencephalography (SEEG) was utilized to prove mesial temporal lobe onset (55). Overall, using data from the 11 LITT studies included in the quantitative analysis a pooled estimate of the seizure-free rate per person-year was 0.59 (95% CI 0.53–0.65) (Figure 2F).

The psychiatric complication rate was 4.4% (23/520), but this was only reported across five studies. Worsening of existing mood disorders was reported in 10 patients, worsening of mood in general was reported in two individuals and transient anxiety was reported in three. Two others needed hospitalization due to psychosis (14) or acute postoperative psychiatric complications (51), and five were diagnosed with a new onset affective disorder.

The median MINORs score for LITT studies was 10/16, indicating a moderate degree of bias in the studies.

The primary aim of this meta-analysis was to review the effectiveness of different interventional techniques for MTLE, by comparing seizure freedom rates for resective, ablative and SRS treatments. We were also interested in stereotactic radiosurgery as no meta-analysis to date has attempted to compare this to LITT, radiofrequency and open surgical methods. We also compared seizure free outcomes between ATLR, trans-sylvian and transcortical SAH. Due to heterogeneity within interventions, comparison within the random-effects model was not possible. However, a mixed-effects linear regression model suggested significant differences between interventions, with LITT, ATLR and SAH demonstrating the largest effects estimate and RF-TC the lowest (Figure 2G).

Between the open surgical methods, our mixed effects model for seizure-free rate per person year estimates were greatest for ATLR, followed by transsylvian SAH and then trans-cortical SAH, but overlap of the 95% confidence intervals prevents determination of comparative superiority. With regard to neuropsychological outcomes, open surgery does not impede on intellectual status and often even leads to post-operative improvement. Language ability, however, is prone to post-surgical decline, but the factors influencing such decline (such as the side and type of surgical intervention) are unclear. Whereas visuospatial memory is likely to either remain stable or improve after surgery, verbal memory is prone to post-surgical decline. In terms of the intervention type, SAH seems more beneficial than ATLR for visuospatial memory. On the other hand, mixed findings are reported regarding the effects of the type of surgical intervention on verbal memory. A meta-analysis of complications following ATLR reported psychiatric complications, visual field defects and cognitive disorders as most common (60). The fact that the efficacy and complications of open surgical techniques are so well-researched mean that it will require clear outcomes from minimally invasive techniques to replace surgical methods such as ATLR that have an established evidence base, including the results from this meta-analysis which provides further evidence for their efficacy in seizure control.

The estimate of the seizure-free rate per person-year for those undergoing radiosurgery was lower than that for open surgical methods such as ATLR and transcortical SAH. Additionally, there were only seven papers (three prospective) included in the meta-analysis for radiosurgery as compared with the 21 papers that looked at ATLR. The highest level of evidence arises from the ROSE randomized control trial which revealed inferiority of SRS compared to ATLR, despite this trial terminating early (11). The benefits of SRS include that it is an outpatient procedure and may be suitable for those patients unable or unwilling to undergo open surgery. Notable drawbacks are a delay in seizure cessation, post-treatment oedema with subsequent need for steroids, and the unexpected late effects of radiation must be considered.

No papers reported significant declines in verbal or visuospatial memory, intelligence or language. Notably, a published meta-analysis of neuropsychological outcomes criticized the methodological quality of SRS studies, calling for better prospective designed trials (61). A meta-analysis considering SRS published in 2016 reported an average of 14 months to seizure cessation. However, this analysis included studies with small sample sizes (<10 patients) as well as other etiologies, including cavernous malformations (62).

This technique had the lowest seizure-free rate in our mixed effects linear regression model. The quality of evidence for this technique was low, with only one prospective study included in our analysis. Nevertheless, advantages associated with this technique are that it can be performed immediately following SEEG, it has a low risk safety profile and unlike gamma knife there is no latency of effect onset (63). Additionally, according to the same study, there was no need for corticosteroid use following RF-TC compared to SRS. RF-TC is unlikely to be successful as a stand-alone treatment for mesial temporal lobe epilepsy but transient reductions in seizure frequency may predict seizure free outcomes following subsequent open surgery. Such studies were not included in the quantitative analysis within the RF-TC arm as the definitive intervention in these studies was open surgery.

The first reported case of LITT was in 2010. In subsequent years, this technique has shown promising seizure effectiveness. Our mixed-effects model demonstrated overlapping 95% confidence intervals with open surgical models. Visual field defects were the most reported complication in line with previous reports (64). Some of these included visual defects that were asymptomatic to the patient, some of which are not identified until formal visual field testing is done. Such asymptomatic patients might fail UK driving criteria once formal perimetry is conducted. The type of perimetry performed can also skew results, with Esterman perimetry (the current DVLA gold standard) being more lenient. We emphasize that it is crucial to conduct formal visual testing after epilepsy surgery, so that visual complications are not underestimated, and patients can be warned about the potential restrictions on driving. Laser ablation did not show a delayed treatment effect like radiosurgery and its evidence base was more reliable than that of radiofrequency, featuring several more studies. This presents this technique as a strong contender to open surgical methods. Dominant-sided verbal memory decline seemed to be the most common neuropsychological complication but not all studies reported a decline, with an improvement in naming function reported. The much shorter hospital stay associated with LITT compared with ATLR also makes it an attractive first line alternative to open surgery (54). LITT may appear beneficial for post-operative neuropsychological outcome when directly compared with open surgery, perhaps due to relative sparing of the parahippocampal gyrus (50). Our comparative analysis suggested ATLR had the largest effect size estimate of all surgical techniques but overlap of the 95% confidence intervals with SAH and LITT prevents firm conclusions regarding superiority. It is possible, however, that LITT effectiveness is overestimated due to a positive reporting bias associated with novel technologies (halo effect), a reluctance to publish poor outcomes, shorter follow up durations and overcoming of learning curves. A systematic review comparing only the minimally invasive techniques LITT, RF-TC and SRS concluded that LITT shows promising seizure efficacy in the short-term (65). Our linear-regression model results provide a much-needed statistical analysis of the data in comparison to open surgical methods to further support LITT as an emerging technique, but long-term follow ups and direct comparisons are needed to form a firm conclusion. Additionally, we excluded overlapping data from our meta-analysis in order to avoid duplicate cohorts in our statistical analysis.