Study design

In this retrospective claims data-based cohort study, we identified all patients in South Korea who underwent elective surgical clipping or endovascular coiling of an unruptured intracranial aneurysm from 1 January 2008 to 31 December 2014, and we followed them up to 31 December 2015. The study was approved by the institutional review board of the Ministry of Health and Welfare. The requirement for informed consent was waived because all patients in the database were anonymized.

Data source

The patients’ data included in the study were collected from the claims database of the Korean Health Insurance Review and Assessment Service (HIRA). HIRA is a government agency that examines the accuracy and constructs databases of claims for the National Health Insurance and Medical Aid program. The former is a mandatory social insurance and healthcare provider, and the latter is a service providing public assistance with healthcare costs for low income households, which covers the entire population of South Korea and is based on a fee-for-service model. All clinics and hospitals in South Korea participate in these two programs.

Patients

Patients were eligible if they were older than 18 years, had no history of subarachnoid hemorrhage (I60.X) from 1 January 2007 to the date of index admission, were admitted electively between 1 January 2008 and 31 December 2014, had procedure codes for surgical clipping (S4641, S4642) or endovascular coiling (M1661, M1662), and had a primary diagnosis of unruptured intracranial aneurysm (I67.1).

We excluded all patients with procedure codes for surgical clipping or endovascular coiling between 1 January 2007 and 31 December 2007. Additionally, patients who were admitted to the emergency room or underwent staged treatment were excluded. Staged treatment was defined as endovascular coiling or surgical clipping performed separately during the index admission or elective readmission for endovascular coiling or surgical clipping within 180 days after the index admission. We also excluded those who had the following during admission: intracranial artery bypass operation, resection or endovascular embolization of an arteriovenous malformation, arteriovenous fistula, cavernous malformation, and any tumor located in the brain or dura mater. Detailed lists of procedure codes are provided in online supplementary table 1.

Identification of aneurysm characteristics and coexisting conditions

The size, location, and multiplicity of treated aneurysms were identified based on the description of the procedure codes for surgical clipping, number of clips used for each aneurysm, frequency of general endotracheal anesthesia and procedures during the index admission, claim guidelines for treating multiple aneurysms, and claim guidelines for usage of coils. Procedure codes for diagnostic digital subtraction angiography were also used for identification.

When submitting a claim after surgical clipping, neurosurgeons should enter the proper procedural code for surgical clipping according to whether or not the size of the treated aneurysm was ≥7 mm. Therefore, we categorized the size of the treated aneurysm into <7 or ≥7 mm. Patients who had at least one treated aneurysm of ≥7 mm in size were allocated to the ≥7 mm size variable.

We categorized the location of the treated aneurysm into anterior circulation (internal carotid artery, anterior cerebral artery, or middle cerebral artery) or posterior circulation (vertebrobasilar system) with the use of procedural codes for diagnostic digital subtraction angiography and surgical clipping. If patients underwent surgical clipping or endovascular coiling of posterior circulation aneurysms or had concurrent anterior and posterior circulation aneurysms, they were allocated to the posterior circulation location group.

We determined patients’ coexisting conditions by analyzing the primary and secondary diagnosis of claims data between 1 January 2007 and the index admission. The International Classification of Diseases 10th version of the Elixhauser algorithm from the Healthcare Cost and Utilization Project was used to identify coexisting conditions, except for hemorrhagic stroke, ischemic stroke, and hyperlipidemia in the inpatient and outpatient settings.13 14 To enhance the accuracy of determination, any diagnosis code that was found only once in an outpatient claim was not considered as a coexisting condition.

Detailed descriptions of procedures, the algorithm for identifying the location of aneurysms, claim guidelines, and diagnostic codes for hyperlipidemia, cerebral hemorrhagic and ischemic stroke are provided in online supplementary material figure 1, table 2.

Management of missing values

Missing values were found in 47.6% and 41.9% of the size and location variables, respectively. We applied multiple imputations to manage them. Since the missing pattern was arbitrary and missing variables were binary, we used a fully conditional specification as an imputation algorithm and a logistic regression model as an imputation model.15 Explanatory variables of the logistic regression model for imputation included baseline covariates and outcome variables. One hundred datasets were created and statistically analyzed. Results from each dataset were combined using Rubin’s Rule with appropriate transformation.16

Multiple imputation was performed under the assumption that missing at random (MAR) was the missing mechanism, but it is possible that the missing mechanism is missing not at random (MNAR). To confirm whether our results of primary outcome after multiple imputation are perusable, sensitivity tests were performed under several scenarios with the assumption of MNAR. We used the MNAR multiple imputation procedure provided in the SAS Enterprise Guide Version 6.3 (SAS Institute, Cary, North Carolina, USA), which was based on a pattern-mixture model approach.17

Adjustment for unbalanced covariates

We calculated the propensity score and used it for inverse probability treatment weighting (IPTW) for average treatment effect on the treated (ATT) to control for unbalanced baseline covariates and to estimate ATT.18 The propensity score was the probability that patients would be selected for surgical clipping given baseline covariates. It was estimated with the use of a logistic regression model that included patients’ demographics and coexisting conditions as explanatory variables. After IPTW for ATT, the weight of patients who underwent surgical clipping was fixed to 1, and the weight of patients who underwent endovascular coiling was calculated using the propensity score. This made it possible to compare the outcomes of patients who underwent surgical clipping with the predicted outcomes of those who underwent endovascular coiling.

The standardized difference was estimated to evaluate the balance of baseline covariates before and after weighting. If the standardized difference was less than 10.0%, we considered them appropriately balanced.

Outcomes

The primary outcome was all-cause mortality at 7 years after index admission. Secondary outcomes included periprocedural outcomes, the incidence of aneurysm rupture after treatment, and retreatment for any cause (eg, recanalization of treated aneurysm, de novo formation, or growth of observed aneurysm). Periprocedural outcomes were defined as all-cause mortality and adverse events that were identified from the corresponding procedure or diagnosis code within 30 days after the index admission. We defined aneurysm rupture as emergent readmission with a primary or secondary diagnosis code of I60.x or emergent readmission with a surgical clipping or endovascular coiling code. Retreatment for any cause was defined as elective readmission with a surgical clipping or endovascular coiling code 180 days after index admission. A list of the procedure codes used in the analysis of periprocedural outcomes is provided in online supplementary table 3.

Statistical analysis

We presented categorical variables as percentages and continuous variables as means with SD or medians with interquartile ranges. The primary outcome was estimated using the Kaplan–Meier method, and the log-rank test was performed to compare the groups. In the analysis of the incidence of aneurysm rupture and retreatment, the Kaplan–Meier method and log-rank test were used after censoring death. The 30-day mortality of periprocedural complications was estimated using the Kaplan–Meier method, and comparisons between the treatment groups were performed using the log-rank test. Cox proportional hazard models were used to estimate the hazard ratios (HRs) of the primary outcome, incidence of aneurysm rupture and retreatment, and 30-day mortality. The assumption of proportional hazard was tested and confirmed by including a time-treatment covariate as an interaction term. Categorical variables in periprocedural complications, except 30-day mortality, were compared using the χ² test, and the relative risk (RR) was estimated for treatment effects. For continuous variables related to periprocedural complications (for example, length of hospitalization), the Wilcoxon rank-sum test was performed to compare the median between the treatment groups. All P values were two-sided and if the P value was <0.05 the result was considered statistically significant. All analyses were performed using SAS Enterprise Guide Version 6.3.