Study design and data source
This nationwide, population-based, retrospective cohort study used the Korean NHI claims database (diagnoses based on International Classification of Disease, 10th Revision [ICD-10] codes and procedure history based on Electronic Data Interchange [EDI] codes), which includes all claims data from the Korean NHI program and the Korean Medical Aid program from 2009 until 2016; the data are integrated into the Health Insurance Review and Assessment Service (HIRA) database to include all healthcare utilization data for both inpatients and outpatients. These data contained a de-identification code representing patient age, sex, diagnosis, hospital admissions, dates of visits, and procedure history.[10, 12] Additionally, prescribed drug information containing the generic name, prescription date, and duration of prescription was included. The Institutional Review Board (IRB) of our institution approved the study. Consent was specifically waived by the IRB because all personal identifying information was removed from the database.
Selection of study sample and definitions
The outcomes of interest were incidence rate and risk factors of new-onset postoperative stroke in patients treated with unilateral TKA compared with subjects with bilateral TKA. The study population comprised individuals older than 40 years of age who received TKA (EDI: N2072, N2077) without history of stroke (ICD-10: I60, I61, I62, I63) during the preceding 1 year, as documented by primary diagnosis and first additional diagnosis in the NHI database between January 1, 2009 and December 31, 2016. Patients treated with bilateral TKA were classified into two groups: patients who underwent SiBTKA and had two primary TKA procedure codes entered on the same day and patients who underwent StBTKA and had two primary TKA procedure codes entered without discharge. Similarly, patients treated with unilateral TKA were classified into two groups: patients who underwent only one TKA during the study period and patients who underwent a second TKA more than 6 months after discharge of index TKA. This is because the planned-staged cohort (more than 6 months) does not appear to be directly related to the index TKA for the risk of cardiovascular complication.[6, 13] New-onset postoperative stroke was defined as history of stroke from the date of primary admission or re-admission for stroke in the hospital following TKA. All patients who were deemed to have had a stroke within 12 months after TKA were identified. Patients considered eligible for newly acquired stroke included subjects who received computed tomography (CT) and magnetic resonance imaging (MRI) within one week after admission as well as subjects undergoing relevant surgical procedures, such as burr hole, craniectomy, craniotomy, or thrombectomy. To assess the diagnostic accuracy of the stroke cases registered in the NHI program, we reviewed the image sets and medical records of all registered stroke patients who received TKA at a single medical center. Two neurosurgeons independently investigated whether registered and suspected cases met the diagnostic criteria for strokes released by NHI.
Patient characteristics, comorbidities, and co-medication were considered as confounders in this study. Characteristics were age, sex, location, hospital size, and insurance type. Comorbidities comprised acquired immune deficiency syndrome (AIDS), congestive heart failure (CHF), chronic kidney disease (CKD), chronic obstructive pulmonary disease (COPD), connective tissue disease, transient ischemic attack (TIA), dementia, hemiplegia, myocardial infarction (MI), peptic ulcer disease (PUD), peripheral vascular disease (PVD), liver disease, severe liver disease, malignancy, diabetes, diabetes with complication, atrial fibrillation (AF), valvular heart disease (VHD), carotid artery disease (CAD), and hypothyroidism based on previous diagnoses within one year before the index date. In addition, the Charlson Comorbidities Index was calculated for all patients ; those with no comorbidities received a score of 0 points. Information on the use of drugs was based on a three-month period within one year before the index date because, in South Korea, drugs are generally prescribed for three months and are typically used on a continuous basis. Potent anticoagulants, such as aspirin, vitamin K antagonist, factor Xa inhibitor, and direct thrombin inhibitor, also were selected as confounders because they have been used for thrombophylaxis following TKA. In addition, hospitals were classified into two groups based on size (large: tertiary hospital or general hospital; small or medium: hospital or clinic). In the Korean health care system, the parent category of hospitals includes subcategories of hospitals, general hospitals, and tertiary hospitals, the requirements for whose qualifications are stated by Korean law. As a subcategory, a hospital signifies a small hospital in Korea (30-100 beds). General hospitals are hospitals equipped with more than 100 beds and several specialty departments as designated by law, and tertiary hospitals are large-sized university hospitals selected by the government.
The results of the study should be randomly selected to ensure that there is no difference in characteristics. However, case-control study works on a specific group, so there is no random assignment, and selection bias cannot be avoided.
Propensity score (PS)-based analyses were used to simultaneously control for a large number of covariates and to mimic some of the particular characteristics of a randomized controlled trial; these analyses provide a more robust, less biased estimate when the number of outcome events is low relative to the number of confounders. PS matching is a method of matching the most similar PS. In this study, PS was calculated using logistic regression and performed one-to-one nearest neighbor matching based on the estimated PS. We fit a logistic regression model to estimate the probability of treatment with unilateral TKA versus bilateral TKA, adjusted for all covariates including age category, sex, comorbidities, and co-medication. We evaluated the balance of measured confounders before and after weighting using absolute standardized differences and considered balance as an absolute value less than 0.1, which has been used in the literature as the definition of a negligible difference.[16, 17] We calculated the incidence rate per 1,000,000 person-years by dividing the number of stroke events by the total number of person-years at risk and multiplying the result by 1,000,000. The 95% confidence interval (CI) was calculated assuming a Poisson distribution. Subgroup analysis was conducted based on age category, sex, location, hospital size, insurance type, comorbidities and co-medication. Adjusted hazard ratio (HR) and 95% CI were calculated using multivariate logistic regression modelling after adjusting for age, sex, location, hospital size, co-medication, and comorbidities. In addition, a sensitivity analysis was conducted to assess the influence of residual confounding based on insurance type. All analyses were conducted using SAS Enterprise software version 6.1 (SAS Institute, Cary, NC, USA) and R software version 3.4.1 (R Foundation for Statistical Computing, Vienna, Austria).