2.1 Data sources
The National Health Insurance (NHI) system of the Republic of Korea
covers approximately 98% of the overall Korean population [9]. The
Health Insurance Review and Assessment Service (HIRA) is a government
organization that reviews and assesses all NHI claims. The HIRA database
provides demographic, diagnostic, treatment, and prescription
information for all inpatients and outpatients as anonymized data
[9]. Our study was approved by the Institutional Review Board of
Seoul National University (No. E2002/001-008) and the requirement for
written informed consent was waived owing to the anonymized data.
We used two separate databases from Korea’s claims data provided by the
HIRA for derivation and validation cohorts. The derivation cohort
included the entire population of the elderly (≥65 years) diagnosed with
arthritis between July 2016 and June 2017 in Korea (n=2,570,122) using
customized HIRA research data (M20200324406). The diagnosis of arthritis
was identified using ICD-10 codes applicable for rheumatoid arthritis
and osteoarthritis (Table S1). This cohort was followed up until
December 31, 2018 (time span, up to 30 months).
For the external validation
cohort, we used the HIRA sample data from 2019 (HIRA-APS in 2019), which
included 10% of the total Korean elderly population (approximately
700,000 patients).
2.2 Study population
A nested case-control study was
conducted on a development cohort. The inclusion criteria were patients
who used NSAIDs for more than 30 consecutive days allowing for a 15–day
gap from July 2016. NSAID treatment included non-selective (ns) NSAIDs
and selective COX-2 inhibitors (ATC code for NSAIDs in Table S1).
Patients were followed up for up to 10 days after the end of NSAID
prescription. The patient selection process is described in Fig 1.
Patients diagnosed with esophageal varices, GI cancer, or Mallory–Weiss
syndrome within 6 months of NSAID initiation were excluded.
NSAID users in the external
validation cohort included those who used NSAIDs after July 2019 for
more than 30 consecutive days, allowing a 15–day gap, and were followed
up for 10 days after the end of NSAID prescription. The exclusion
criteria were identical to those used in the development cohort.