2.1 Ethics statement
Data resource was based on the National Health Insurance Research Database (NHIRD). [22] Researchers can access NHIRD data after ethical and scientific review processes. Prior to applying, the study has been approved by the ethical review board of National Taiwan University Hospital (NTUH-REC No.201406018W). There are 27 institutional review boards capable of issuing approvals, and all are supervised and regulated by the Taiwan Ministry of Health and Welfare. To protect individuals’ confidentiality, all datasets in the Data Science Centre are pseudonymized. Personal ID, birth date, and names are encrypted, and this de-identification process was approved by an independent third party. We performed data analysis in the branches of the Data Science Centre. The analyzed results were also examined by the Data Science Centre before exporting. The Institutional Review Board had verified the anonymity of data analysis performed in this study. All research procedures followed the directives of the Declaration of Helsinki.
2.2 Study Design
This is an observational study to compare the cardiovascular events of two cohorts of the newly treated (incident) PD patients before and after the relaxation of NHI’s erythropoietin payment criteria. Cohort 1 included dialysis patients who started to receive maintenance PD treatments during a specified period of 28 months before the relaxation of NHI’s erythropoietin payment criteria. To ensure an adequate observation period, this cohort was followed up for additional 14 months after the month of the cohort’s last patient enrollment. Cohort 2 included incident dialysis patients who started to receive maintenance PD treatments within a 28-month time interval after the relaxation of NHI’s erythropoietin payment criteria. Additional 14-month follow-up observations were also made after the month of the cohort’s last patient enrollment. We set a 6-month time lag between the initiation of relaxing erythropoietin payment criteria and the time that the first enrollment of patients in Cohort 2 in order to accommodate possible adaptation of physician prescribing practices to the new policy.
Because the potential imbalances in the distributions of many measured and unmeasured baseline covariates exist between the two cohorts, researchers have to employ the propensity score (PS) analysis, developed by Rosenbaum et al. [23] Thus, the influence of any potential enrollment biases between these two cohorts was attenuated through a PS matching approach and identification of patients with comparable characteristics in the two cohorts. This study defined PS as the probability of a patient having experienced a cardiovascular event. Patients in Cohort 1 and 2 were matched with PS scores estimated by age, sex and a comorbidity index with the nearest neighbor-greedy approach. The comorbidity index was developed by Liu et al. [24] specifically for U.S. Medicare dialysis population and had been validated in Taiwanese dialysis patients [25].
After matching with PS, patients were followed up till experiencing either one of the following three events: 1) the occurrence of cardiovascular endpoints, or 2) change to hemodialysis, or 3) data cut-off point (October 31, 2006 for Cohort 1; October 31, 2010 for Cohort 2), whichever occurred earlier. Survival analysis models were then employed to investigate the differences in the risk of cardiovascular events between the two cohorts of incident PD patients. Baseline demographics and comorbid conditions were used as covariates in the statistical analyses. Monthly erythropoietin doses administered to patients of Cohort 1 and Cohort 2 during the follow-up period were compared to examine whether there existed a difference in the monthly erythropoietin dosage administered between the two cohorts of incident PD patients. In calculation of erythropoietin dosage, while erythropoietin alfa and erythropoietin beta were considered equivalent, darbepoietin alfa was converted into erythropoietin alfa with 1ug of darbepoietin alfa equal to 200U of erythropoietin alfa [26].
Cardiovascular risk could be affected by treatments with concomitant medications related to cardiovascular comorbidities. Patients taking medications related to cardiovascular comorbidities during the follow-up period in two cohorts were also examined. The concomitant medications related to cardiovascular comorbidities were identified by corresponding ATC code, including acetylsalicylic acid (B01AC06) or clopidogrel (B01AC04), angiotensin converting enzyme inhibitors (C09A) or angiotensin receptor blockers (C09C), beta blockers (C07), calcium channel blockers (C08) and statins (C10AA). A patient received the medication for any three months during the follow-up period would be considered under the treatments of concomitant medications related to cardiovascular comorbidities.
Finally, in addition to administering erythropoietin, because the patient Hct level could also be affected by the usages of iron and red-cell transfusion, iron and red-cell transfusion for patients in the two cohorts were examined to determine if differences in the usages of iron and red-cell transfusion existed between the two cohorts.
2.3 Patient selection
Incident PD patients were identified from the claim data of entire beneficiaries covered by the NHI system from 2003 to 2010. Collection and analysis of the NHI claimed data were approved by the National Taiwan University Hospital Human Research Ethics Committee. The analyses were performed on de-identified data extracted from the NHI research database compiled by Taiwan National Health Research Institutes. A patient receiving over 90-day consecutive dialysis treatments and with PD on the day 90th and thereafter is considered as an incident PD patient in this study. Cohort 1 recruited patients with the 90th dialysis day between May 1, 2003 and August 31, 2005, and Cohort 2 recruited patients with the 90th dialysis day between May 1, 2007 and August 31, 2009. Young patients under 20 were excluded because the comorbidities were different between pediatrics and adults. There were 1,759 patients in Cohort 1 and 2,981 patients in Cohort 2. After the PS matching, each cohort contained only 1,754 patients.
2.4 Statistical Analyses
The primary outcome measure is a composite cardiovascular endpoint, defined as myocardial infarction, heart failure hospitalization, stroke or death. Myocardial infarction was defined by International Classification of Diseases, Ninth Revision (ICD-9) code 410, 411 in the hospital discharge diagnosis. Heart failure hospitalization was defined by ICD-9 hospital discharge diagnosis codes 398.91, 422, 425, 428, 402.x1, 404.x1, 404.x3, and V42.1. Stroke was defined by ICD-9 hospital discharge diagnosis codes 433, 434, 436, 437.0 and 437.1. For primary outcome measure, all patients in both cohorts were followed up to the occurrence of myocardial infarction, heart failure hospitalization, stroke or death, whichever occurred earlier. Secondary outcomes include myocardial infarction, heart failure hospitalization, stroke and death. Each patient was followed up to the occurrence of each cardiovascular event. Data on patients who did not have an event were censored at the data cut-off point or transition to hemodialysis, whichever occurred earlier.
The selection and analyses of primary and secondary endpoints of cardiovascular risk in our study were the same as previous large-scale studies [1-3]. In addition to cardiovascular events, death was also considered an important clinical endpoint in the evaluation of cardiovascular risk because reducing mortality is an ultimate goal of reducing cardiovascular risk. Using a composite primary endpoint with each component evaluated as the secondary endpoint analysis is commonly adopted by many clinicians [2, 3], such as pivotal studies of new drug applications. This allows a thorough evaluation of contribution of each component of composite primary endpoint and avoiding any biases introduced by one of the dominating component.
Cox proportional hazards model was employed to estimate the cardiovascular risk between the two cohorts. Estimated hazard ratios (HRs) for Cohort 2 relative to Cohort 1 and 95% confidence intervals (CIs) were calculated. In order to obtain more insightful results, patients were further stratified by diabetic status. Cox regression analyses for diabetic and non-diabetic patients were performed separately. All analyses were performed using SAS software, version 9.1.