JMIR Publications

ABSTRACT

Background:

Adverse events (AEs) are frequently associated with the use of medical devices, yet effective post-market surveillance remains limited because device usage is rarely captured in a structured form and voluntary reporting is incomplete. Common data model (CDM)-based distributed research networks allow clinical evidence to be shared across institutions without transferring patient-level data, but existing CDMs contain little of the device-specific information required for medical device vigilance.

Objective:

We aimed to develop a Korean Common Data Model for Medical Device Vigilance (K-CDM MDV) that supports structured, real-world data (RWD)-driven active surveillance of device-related AEs, and to evaluate its feasibility using electronic medical record (EMR) data from a single tertiary hospital.

Methods:

The K-CDM MDV was designed from voluntary Medical Device Reports (MDRs) of several countries and from device- and AE-related elements of the Observational Medical Outcomes Partnership (OMOP), Sentinel, and PCORnet CDMs, and extended with standardized device usage data and Unique Device Identifiers (UDIs) to interoperate with existing CDMs. As a pilot application, EMR records of patients who used a hemodialysis machine or a coronary stent at Konyang University Hospital (KYUH) between 2012 and 2019 were pseudonymized, transformed, and loaded into the K-CDM MDV. Two scenario-based hypotheses were examined: the time-dependent risk of hypotension for two hemodialysis machines, and the severity of AEs across coronary stent types. Descriptive analyses, t tests, relative risks, odds ratios, survival analysis, and analysis of variance were performed in R (version 4.1.0); statistical significance was set at p < 0.05.

Results:

The K-CDM MDV consists of five tables (DEVICE_EXPOSURE, DEVICE_AE, DEVICE_AE_INVOLVED, DEFECTIVE_MD, and META_DEVICE) and connects to external standard CDMs through the DEVICE_EXPOSURE table. In total, 8,064 hemodialysis machine use records and 3,156 coronary stent implantation records were transformed into the model. AEs were reported in 161 of 4,832 uses (3.3%) for device A and in 87 of 3,241 uses (2.7%) for device B; the survival probability of device A decreased to approximately 0.97 after two hours of hemodialysis, whereas that of device B remained near 1.00 until four hours. For coronary stents, information on thrombosis could not be obtained from the existing coded data, and crude AE incidence by stent type ranged from 0.4% to 14.3%. The comparison of AE severity across stent types was not statistically significant (p = 0.4405; F = 1.003), with a small effect size (eta-squared = 0.037).

Conclusions:

The K-CDM MDV offers a standardized and interoperable structure that captures both detailed device usage and the causes of AEs. Although it was developed and evaluated in a single-institution setting, its design supports integration across multiple institutions and provides a scalable foundation for national-level active medical device surveillance and for informed policy development.