(A-203) Evaluating Result Consistency of OMOP and Sentinel Common Data Models: A Case Study of Hospitalized Atherosclerotic Cardiovascular Disease Patients in Taiwan

Background: OMOP and Sentinel CDMs are two widely used frameworks in multi-national study, with OMOP standardizing both data structure and vocabulary, while Sentinel focuses only on structure. However, the standardization would cause information loss or distortion. It is important to determine an appropriate approach for generating the consistent results from Taiwan’s claim data.

Objectives: To evaluate the results consistency by adopting OMOP and Sentinel CDMs in a study of hospitalized atherosclerotic cardiovascular disease (ASCVD) patients.

Methods: We analyzed the Taiwan National Health Insurance Research Database (2016 - 2019) to examine baseline characteristics of hospitalized patients due to ASCVD events (ACS, MI, angina, stroke, TIA, PAD, PCI, or CABG). Patients were categorized into very-high-risk (VHRG) and high-risk groups (HRG) to compare the risks of ASCVD recurrence (composite and individual outcomes: ACS, MI, ischemic stroke, PAD, PCI, CABG) and mortality. Consistency was evaluated by comparing raw data results with those obtained after conversion into OMOP and Sentinel CDMs. The characteristics, including index ASCVD events, as well as history of comorbidities, and medications, were assessed using absolute standardized mean differences (ASMD), with consistency defined as ASMD < 0.1. The direction and difference of hazard ratios (HRs) for the ASCVD recurrence were evaluated.

Results: From the raw data, 327,742 eligible patients were included, compared to 331,121 in OMOP and 327,023 in Sentinel. The proportion of VHRG was 47.1, 48.1, and 44.7% in raw data, OMOP, and Sentinel, respectively. Index ASCVD events and history of medication were consistent between raw data and CDMs (ASMD < 0.1). The mean Charlson Comorbidity Index (CCI) was higher in OMOP (CCI in RAW: 3.76, OMOP: 4.09, Sentinel: 3.66; ASMD-Raw-OMOP: 0.11). The directions of HR for composite outcomes were the same and had differences less than 3% [HR in Raw: 1.50 (1.47-1.52), OMOP: 1.53 (1.50-1.56), Sentinel: 1.50 (1.48-1.52)]. Individual outcomes show the consistent results, except for PAD in OMOP which the HR was 1.3 higher than raw result [HR in Raw: 4.51 (4.05-5.03), OMOP: 5.81 (5.29-6.38), Sentinel: 4.20 (3.87-4.57)]. Vocabulary standardization influences both data and how to define the variables; therefore, the variables lacking a precise definition, such as PAD, are more likely to impact the consistency of inferential results.

Conclusions: OMOP and Sentinel CDM showed a high level of consistency with raw data in most aspects for hospitalized ASCVD patients, with minor variability stemming from vocabulary standardization. Both CDMs are suitable for generating real-world evidence in Taiwan if the key variables are not ambiguous.