Siobhan West, Sarah J Nolan, Jennifer Cotton, Sacha Gandhi, Jennifer Weston, Ajay Sudan, Roberto Ramirez, Richard Newton

Focal epilepsies are caused by a malfunction of nerve cells localised in one part of one cerebral hemisphere. In studies, estimates of the number of individuals with focal epilepsy who do not become seizure-free despite optimal drug therapy vary according to the age of the participants and which focal epilepsies are included, but have been reported as at least 20% and in some studies up to 70%. If the epileptogenic zone can be located surgical resection offers the chance of a cure with a corresponding increase in quality of life. The primary objective is to assess the overall outcome of epilepsy surgery according to evidence from randomised controlled trials.The secondary objectives are to assess the overall outcome of epilepsy surgery according to non-randomised evidence and to identify the factors that correlate to remission of seizures postoperatively. We searched the Cochrane Epilepsy Group Specialised Register (June 2013), the Cochrane Central Register of Controlled Trials (CENTRAL 2013, Issue 6), MEDLINE (Ovid) (2001 to 4 July 2013), ClinicalTrials.gov and the World Health Organization (WHO) International Clinical Trials Registry Platform (ICTRP) for relevant trials up to 4 July 2013. Eligible studies were randomised controlled trials (RCTs), cohort studies or case series, with either a prospective and/or retrospective design, including at least 30 participants, a well-defined population (age, sex, seizure type/frequency, duration of epilepsy, aetiology, magnetic resonance imaging (MRI) diagnosis, surgical findings), an MRI performed in at least 90% of cases and an expected duration of follow-up of at least one year, and reporting an outcome relating to postoperative seizure control. Three groups of two review authors independently screened all references for eligibility, assessed study quality and risk of bias, and extracted data. Outcomes were proportion of participants achieving a good outcome according to the presence or absence of each prognostic factor of interest. We intended to combine data with risk ratios (RR) and 95% confidence intervals. We identified 177 studies (16,253 participants) investigating the outcome of surgery for epilepsy. Four studies were RCTs (including one that randomised participants to surgery or medical treatment). The risk of bias in the RCTs was unclear or high, limiting our confidence in the evidence that addressed the primary review objective. Most of the remaining 173 non-randomised studies had a retrospective design; they were of variable size, were conducted in a range of countries, recruited a wide demographic range of participants, used a wide range of surgical techniques and used different scales used to measure outcomes. We performed quality assessment using the Effective Public Health Practice Project (EPHPP) tool and determined that most studies provided moderate or weak evidence. For 29 studies reporting multivariate analyses we used the Quality in Prognostic Studies (QUIPS) tool and determined that very few studies were at low risk of bias across the domains.In terms of freedom from seizures, one RCT found surgery to be superior to medical treatment, two RCTs found no statistically significant difference between anterior temporal lobectomy (ATL) with or without corpus callosotomy or between 2.5 cm or 3.5 cm ATL resection, and one RCT found total hippocampectomy to be superior to partial hippocampectomy. We judged the evidence from the four RCTs to be of moderate to very low quality due to the lack of information reported about the randomised trial design and the restricted study populations.Of the 16,253 participants included in this review, 10,518 (65%) achieved a good outcome from surgery; this ranged across studies from 13.5% to 92.5%. Overall, we found the quality of data in relation to the recording of adverse events to be very poor.In total, 118 studies examined between one and eight prognostic factors in univariate analysis. We found the following prognostic factors to be associated with a better post-surgical seizure outcome: an abnormal pre-operative MRI, no use of intracranial monitoring, complete surgical resection, presence of mesial temporal sclerosis, concordance of pre-operative MRI and electroencephalography (EEG), history of febrile seizures, absence of focal cortical dysplasia/malformation of cortical development, presence of tumour, right-sided resection and presence of unilateral interictal spikes. We found no evidence that history of head injury, presence of encephalomalacia, presence of vascular malformation or presence of postoperative discharges were prognostic factors of outcome. We observed variability between studies for many of our analyses, likely due to the small study sizes with unbalanced group sizes, variation in the definition of seizure outcome, definition of the prognostic factor and the influence of the site of surgery, all of which we observed to be related to postoperative seizure outcome. Twenty-nine studies reported multivariable models of prognostic factors and the direction of association of factors with outcome was generally the same as found in the univariate analyses. However, due to the different multivariable analysis approaches and selective reporting of results, meaningful comparison of multivariate analysis with univariate meta-analysis is difficult. The study design issues and limited information presented in the included studies mean that our results provide limited evidence to aid patient selection for surgery and prediction of likely surgical outcome. Future research should be of high quality, have a prospective design, be appropriately powered and focus on specific issues related to diagnostic tools, the site-specific surgical approach and other issues such as the extent of resection. Prognostic factors related to the outcome of surgery should be investigated via multivariable statistical regression modelling, where variables are selected for modelling according to clinical relevance and all numerical results of the prognostic models are fully reported. Protocols should include pre- and postoperative measures of speech and language function, cognition and social functioning along with a mental state assessment. Journal editors should not accept papers where adverse events from a medical intervention are not recorded. Improvements in the development of cancer care over the past three to four decades have been achieved by answering well-defined questions through the conduct of focused RCTs in a step-wise fashion. The same approach to surgery for epilepsy is required.