Background: Conducting pharmacoepidemiological research globally comes with considerations as each country has a distinct healthcare system. In the US, claims data are widely used in such research where in Japan those are hospital-based. There is minimal evidence on the comparison of the US and Japan databases for real-world evidence generation.
Objectives: To assess the feasibility of replicating a cardiovascular disease US-based claims analysis in Japanese databases.
Methods: We assessed 2 Japanese hospital databases (Medical Data Vision - MDV - and RWD maintained by the Health, Clinic and Education Information Evaluation Institute with support from Real World Data Co. Ltd. – RWD -) aiming at replicating a US study using Optum Clinformatics Database (CDM). MDV is a hospital administrative database covering DPC hospitals. RWD is an electronic medical record (EMR) dataset covering DPC and non-DPC hospitals. Optum CDM is an administrative claims database containing pharmacy, medical claims along with lab and mortality data. Descriptive statistics were used to assess the availability and quality of data elements needed to replicate the US study.
Results: The Japanese data structure did not mirror the US one necessitating careful adaptation of variable definitions for enrollment, hospitalizations, comorbidities, treatments, clinical values and mortality. The identification of inpatient visits in the RWD required a proxy capturing EMR hospitalization events with ≥2-day duration. In both databases, tracking patients across hospitals was not possible leading to under-reporting of hospitalizations and other emergency-care visits. Careful assumptions on the observability of patients had to be made. Both databases included prescriptions from hospitals only; thus, captured inpatient and outpatient treatment exposures only in the same facility. ICD-10 and procedure codes were used to capture comorbidities. ATC-WHO codes from Optum CDM were adapted to define drug exposure in MDV (ATC-EphMRA). English-translated generic drug names were limited in MDV data. Lab values in both databases required attentive reformatting. In MDV, adaptation for mortality definition was required using DPC variables such as “death within 24 hours of hospitalization” and “destination after discharge”.
Conclusions: When replicating US claims analyses using Japanese databases, specific considerations should be planned ahead, notably the language challenge, differences between the 2 healthcare and payment systems, and therefore, captured data. Other aspects include non-Japanese-specific considerations such as formatting adjustments, dealing with missingness, and creating proxies to workaround the data contents.
