We developed and adhered to a protocol for population, intervention, comparison and outcome (PICO) search methods, data extraction, quality assessment, meta-analysis and grading of the quality of the evidence for this systematic review.
Articles Published before July 2010 in 16 electronic databases were searched. These databases included three international databases, ten Korean domestic databases and three Japanese domestic databases: (1) Ovid-Medline, (2) Ovid-Embase, (3) Cochrane Library, (4) KISS (
http://kiss.kstudy.com), (5) KMBASE (
http://kmbase.medric.or.kr), (6) KoreaMed (
http://www.koreamed.org), (7) NDSL (
http://www.ndsl.kr), (8) KiSTi (
http://society.kisti.re.kr), (9) J Cerebrovasc Surg (
http://jkcvs.ksevs.org), (10) J Korean Soc Radiol (
http://www.radiology.or.kr), (11) J Korean Neurosurg Soc (
http://jkns.or.kr), (12) Neurointervention (
http://www.ksin.or.kr), (13) Korean J Stroke (
http://www.stroke.or.kr), (14) JAMAS (
http://www.jamas.or.jp), (15) Medical online (
http://www.meteo-intergate.com) and (16) J-stage (
http://www.jstage.jst.go.jp). Various combinations of Mesh headings and keywords were used, such as “intracranial aneurysm”, “subarachnoid haemorrhage”, “embolisation”, “neurosurgical procedures”, “neurosurgery”, “unruptured”, “coil”, “Guglielmi”, “GDC”, and “clip”. The search was limited to human studies without language restrictions.
Study selection, data extraction and quality assessment
Eligible studies were those that included adult (≥18 years) patients with UIA. Patients with mycotic, infectious, dissecting or fusiform aneurysms, arteriovenous malformation or arteriovenous fistula were excluded. Case series, case reports and publications that had not undergone peer review were excluded. In addition, studies were required to have directly comparable outcome measures for endovascular coiling and neurosurgical clipping, such as overall death, in-hospital mortality, disability and complications. Disability was regarded as a score of 1–5 on the Modified Rankin Scale [mRS] and a score of 2–4 on the Glasgow Outcome Scale [GOS]. The mRS, a clinician-reported measure of global disability, is widely applied to evaluate stroke patient outcomes, and scores run from no symptoms at all (score 0) to death (score 6)
. The GOS assesses outcome after severe brain damage and runs from death (score 1) to good recovery (score 5)
. There were various complications, including bleeding or haematoma in the brain, ischemia or infarction, cerebral vessel damage, cerebrospinal fluid fistula, infection, cranial nerve disorder, cognitive impairment, encephalitis, meningitis, embolism, cardiac disease and pulmonary disease.
Four authors (HJL, JEC, JSH and MKH) independently reviewed all searched articles and extracted data using pre-made extraction forms that included study design, follow-up period, inclusion/exclusion criteria, sample size, sex, age, aneurysm location, aneurysm size, race, baseline characteristics, treatment protocol, outcome variables and complications. Disability was calculated based on the percentages of categories other than death/good recovery in mRS/GOS. The numbers of events for outcomes were extracted according to the intention-to-treat principle.
Six authors (HJL, JEC, JHK, JSH, LRL and MKH) independently evaluated the quality of the studies using the Methodological Index for Nonrandomised Studies (MINORS)
. In the case of disagreement, consensus was reached through discussion and negotiations with partners. If a consensus could not be achieved within the group, a third party was involved, and then an agreement was reached by majority rule.
Meta-analysis and grading the quality of the evidence
A meta-analysis was performed to synthesise the outcomes, except for missing data among 24 articles. Binary outcomes were expressed as odds ratios (ORs). The reporting methods for in-hospital mortality were diverse, including the number of patients, unadjusted ORs and adjusted ORs. Therefore, lnOR and standard error were calculated using statistical equations.
Fixed and random effects inverse-variance meta-analysis was used to combine the studies and obtain the average effects and 95% confidence intervals (CI). The subgroup analyses were performed by outcome measurement times (short time, ≤6 months; long time, >6 months) according to medical experts’ opinions.
To assess heterogeneity across studies, funnel plots were visually examined, and Cochran’s Q statistic and the I2 statistic were used. Publication bias was also assessed using funnel plots, Begg and Mazumdar’s rank correlation (Begg’s test) and Egger’s linear regression asymmetry test of the intercept (Egger’s test). Finally, the quality of the body of evidence was graded as “high”, “moderate”, “low”, or “very low”
Meta-analyses were conducted using Comprehensive Meta-analysis 2.0 (Biostat, Englewood, NJ, USA), and the quality of evidence and strength of recommendation were graded using GRADEprofiler 3.2.2 (GRADE Working Group).