The Korean government uses the National Health Insurance Service (NHIS) database, which covers 97% of all citizens (almost 50 million people) in the Republic of Korea. All hospitals in Korea send information about inpatient and outpatient visitations, procedures, prescriptions, and national health examination data to the NHIS. The NHIS then assigns diagnosis codes based on the International Classification of Disease (ICD), 10th edition. These data resources are widely validated and used for epidemiologic studies [16]. The NHIS provides information from claims data for research purposes and includes mortality records with the cause and date of death, which are retrieved from the Statistics Korea database ¹ . Data are available with the approval and oversight of the NHIS (NHIS-2019-1-448) through the Korean National Health Insurance Sharing Service ² . The specific codes used to define every diagnosis, procedure, and drug in this study are shown in Supplementary Tables S1, S2.

This study used NHIS data of patients newly diagnosed with ESKD (defined as requiring hemodialysis, peritoneal dialysis, and/or KT) between January 2007 and December 2018. As KT was usually performed after a period of dialysis treatment (except for cases of pre-emptive KT), comparing KT and dialysis based on the date of ESKD diagnosis would inevitably lead to immortal time bias of patients receiving KT, thereby resulting in an overestimation of the survival of these patients [17]. To minimize this bias, we applied a “prevalent new user design,” which has been used in pharmacoepidemiology. Treating ESKD as a “disease,” dialysis as a “former drug,” and KT as a “new drug,” in accordance with the components of a prevalent new user design, we established separate time-based exposure cohorts for dialysis and KT [18, 19]. The time interval of ±3 months surrounding the date of KT was used to select the dialysis control patients (Supplementary Figure S1). The cohort entry date was defined as the KT date for the KT cohort and the corresponding date of dialysis prescription for the dialysis cohort. When patients included in the dialysis cohort at certain cohort entry dates subsequently underwent KT, they were censored and reused as KT subjects based on the date of KT. This provides an intention-to-treat approach for comparing the effects of proceeding with KT versus continuing on dialysis alone or waiting for further KT at the given entry date. Baseline characteristics, including prior AMI and exclusion criteria, were based on the cohort entry date of each subject. Prior AMI was defined as the first diagnosis of AMI with a hospital admission duration of >2 days.

In this study, we included only patients with a prior AMI within 5 years before each cohort entry date. We excluded patients who were <19 or >75 years of age at the time of cohort entry. Patients diagnosed with cancer (because of its effects on KT eligibility) and those diagnosed with stroke, valvular heart disease, and/or cardiac conduction abnormality (because of the effects of these non-AMI CVDs on KT accessibility and outcomes) within 5 years before cohort entry were also excluded. In addition, patients receiving dialysis for >10 years before KT were excluded to eliminate individuals in excellent medical condition while on dialysis, who then received KT.

The KT and dialysis cohorts were matched according to these steps: 1) the dialysis date corresponding to the KT date was set as the cohort entry date in the dialysis cohort, 2) exclusion criteria were applied based on the cohort entry date, 3) only patients with an AMI within 5 years before cohort entry were selected, and 4) dialysis patients were matched to KT patients based on time-conditional propensity scores calculated using conditional logistic regression stratified by dialysis cohort or KT cohort [20]. The covariates used for generating the time-conditional propensity scores were age, sex, diabetes mellitus, calendar year of ESKD diagnosis, calendar year of cohort entry date, interval from ESKD to cohort entry date, interval from AMI to cohort entry date, type of AMI treatment (percutaneous coronary intervention [PCI], coronary artery bypass graft [CABG], or medication only), and secondary prevention drugs after AMI (angiotensin converting enzyme inhibitors or angiotensin receptor blockers, beta blockers, statins, antiplatelet agents (aspirin, clopidogrel, cilostazole, ticlopidine, prasugrel, ticagrelor, or triflusal) or calcium channel blockers). Use of a drug was defined as being prescribed the drug >2 times during outpatient visits within 1 year before cohort entry.

Patients with underlying conditions were matched according to the Charlson Comorbidity Index (CCI) calculated using data from the 5 years period before cohort entry [13, 14]. Diabetes mellitus and congestive heart failure (CHF) were matched separately from CCI because of their prominent effects on CVD and survival in patients with ESKD. Matching was performed with a 1:3 ratio, without replacement, and in chronological order. If a matched dialysis subject underwent KT during follow-up, the patient was censored at the time of KT, then included in the KT cohort and matched with other patients in the dialysis cohort based on the newly designated entry date (KT date).

The primary outcome of this study was all-cause mortality and MACE, which was a composite of cardiovascular mortality, recurrent AMI, and stroke. The secondary outcomes were each component of MACE and a coronary revascularization procedure (PCI or CABG). Cardiovascular mortality was defined as any death with an ICD-10 code of I00–I99, as confirmed in the Statistics Korea database. Recurrent AMI was defined as hospitalization for the AMI diagnosis code and/or coronary revascularization. The study population was followed from each cohort entry date until the date of death, 31 December 2018, or the date of subsequent KT (in the dialysis cohort), whichever came first.

Matching on time-conditional propensity scores was performed with greedy (nearest neighbor) matching techniques [21]. Covariate balances were considered adequate when standardized mean differences after matching were <0.1 [22]. Baseline characteristics were compared between the KT and matched dialysis groups using the t-test or chi-squared test, as appropriate. Continuous variables were expressed as mean ± standard deviation, and categorical variables were expressed as number (percentage). Kaplan–Meier survival curves with the log-rank test were used to compare cumulative outcome incidences. Hazard ratios for each outcome were obtained before and after being adjusted for baseline characteristics using Cox proportional-hazard regression analysis. Death from causes other than CVD and loss to follow-up were considered as competing risks when comparing MACE and each component. Moreover, regression analyses were performed by Fine and Gray’s model for those outcomes. Sensitivity analyses were performed in various subgroups for all-cause mortality and MACE: age (<65 vs. ≥65 years), sex, AMI treatment method (PCI/CABG vs. medication alone), interval from AMI to cohort entry date (<6 vs. 6–12 vs. ≥12 months), year of cohort entry date (2007–2012 vs. 2013–2018), interval from ESKD to cohort entry date (<1 vs. 2–5 vs. 5–10 years), CCI (<9 vs. ≥9), and CHF (presence or absence). The sensitivity of the effect of KT was analyzed by creating an interaction of the p-value between KT versus non-KT and each subgroup. Furthermore, to confirm whether KT adversely affected outcomes during the early post-KT period, we performed several independent analyses (<3, <6, and <12 months after cohort entry), where administrative censoring was applied to the maximum time point (or earlier if the patient was lost to follow-up).

All p values were two-sided, and p values <0.05 were considered significant. Analyses were performed using the statistical package SAS 9.4 (SAS Institute, Cary, NC, United States) and R version 4.2.0 for Windows ³ .

The Institutional Review Board (IRB) of the Yonsei University Wonju College of Medicine (Wonju, Korea) approved this study (IRB number: CR319308). Informed consent was waived because anonymous and de-identified information was used for the analyses. This trial was registered with the Clinical Research Information Service, Republic of Korea (KCT0005759).

Of the 331,994 first diagnosed with ESKD during the study period, 325,785 were in the dialysis cohort and 13,428 were in the KT cohort (Figure 1). From these, a 1:3 matched ESKD population with prior AMI were included in the comparative analyses: 1,269 dialysis patients were matched to 423 KT patients based on time-conditional propensity scores with appropriate balance (Supplementary Figure S2). Baseline characteristics are shown in Table 1. The mean age was 52.3 ± 10.8 years for the dialysis group and 53.3 ± 11.1 years for the KT group (p = 0.979). Men were more frequent in both groups (73.8% in the dialysis group vs. 73.5% in the KT group; p = 0.924). Year of first ESKD diagnosis and cohort entry date were similar between the two groups. As a result of chronologic matching, the interval from ESKD diagnosis to cohort entry date was similar between the two groups, not only when stratified by <1 year, 1–5 years and 5–10 years, but also when mean values were compared (30.6 ± 29.1 months vs. 33.1 ± 29.7 months; p = 0.482). Mean interval from AMI to cohort entry date was 25.0 ± 18.1 months in the dialysis group and 23.9 ± 18.8 months in the KT group. Additionally, the two groups had similar treatment modalities for their prior AMI (CABG [7.0% vs. 8.0%], PCI [44.7% vs. 40.0%], and medical treatment alone [48.3% vs. 52.0%]; p = 0.226), which were consistent with AMI treatment distributions previously reported in patients with chronic kidney disease [23, 24]. Secondary prevention drugs after AMI were used in similar percentages of patients at cohort entry in both groups. The mean CCI value was more than 8 and similar in both groups (8.7 ± 2.6 vs. 8.6 ± 2.4; p = 0.600). The frequencies of each CCI component were similar between groups, except peripheral vascular disease, dementia, and hemi- or paraplegia.

Of the 423 patients who underwent KT, 185 (43.7%) received <1 year of pre-transplant dialysis before KT, including 66 (15.6%) who underwent pre-emptive KT. The median pre-transplant dialysis duration was 29.6 (interquartile range, 9.7–57.4) months. There were 9 (2.1%) in-hospital deaths after KT: 6 were due to recurrent AMI and 3 were from an unknown cause. There were 13 (3.1%) in-hospital MACE, including 6 cardiovascular deaths and 7 cases of coronary artery disease treated with PCI. The cumulative incidences of graft failure (restart of dialysis or re-transplantation) were 2.4%, 5.2%, and 8.9% at 1, 5, and 10 years after KT, respectively (Supplementary Figure S3).

During the mean follow-up period of 48.3 ± 38.6 months (dialysis group, 45.7 ± 37.7 months; KT group, 61.3 ± 40.7 months), 542 patients in the dialysis group and 41 patients in the KT group died, representing incidence rates of 112.2 and 19.0 per 1,000 person-years, respectively. Except for unknown cause, the most common cause of mortality was CVD, followed by cancer and infection in both groups (Supplementary Figure S4). All-cause mortality was significantly lower in the KT group than in the dialysis group (p < 0.001) based on Kaplan–Meier curve analysis (Figure 2). The 1, 5, and 10 years cumulative incidences of all-cause mortality were 12.6%, 39.1%, and 60.1% in the dialysis group and 3.1%, 7.2%, and 14.5% in the KT group (Table 2). The adjusted hazard ratio (HR) of KT for all-cause mortality was 0.17, with a 95% confidence interval (CI) of 0.12–0.24 (p < 0.001).

The incidence of MACE was also significantly lower in the KT group than in the dialysis group (p < 0.001; Figure 2). The 1, 5, and 10 years cumulative incidences of MACE were 15.6%, 37.6%, and 52.1% in the dialysis group and 6.6%, 13.8%, and 29.1% in the KT group. The adjusted HR of KT for MACE was 0.38, with a 95% CI of 0.23–0.51 (p < 0.001; Table 2).

The incidences of all MACE components were significantly lower in the KT group than in the matched dialysis controls (Figure 2 and Table 2). KT provided the most protection against cardiovascular death, as indicated by the lowest subdistribution HR (HR, 0.23 [95% CI, 0.12–0.42]; p < 0.001). For cardiovascular mortality, the 1, 5, and 10 years cumulative incidences were 3.5%, 11.3%, and 20.5% in the dialysis group and 0.7%, 1.2%, and 9.5% in the KT group. KT was also protective against recurrent AMI (HR, 0.59 [95% CI, 0.38–0.93]; p = 0.023) and stroke (HR, 0.33 [95% CI, 0.23–0.46]; p < 0.001), compared with maintaining on dialysis. Additionally, the incidence of coronary revascularization (PCI or CABG), regardless of the specific diagnosis, was significantly lower in the KT group than in the dialysis group (HR, 0.38 [95% CI, 0.27–0.52]; p < 0.001).

The protective effects of KT for all-cause mortality and MACE were seen in all subgroups, especially in higher-risk patients, such as those >65 years of age, patients with an interval from AMI to cohort entry date of <6 months, and those with CHF (Figure 3). However, when compared within the stratified time intervals during the early period after cohort entry, the KT group had a higher risk of recurrent AMI in the first 3 months post-KT, compared with the dialysis group (HR, 3.30 [95% CI, 1.46–7.47]; p = 0.004) (Supplementary Tables S3, S4).