Publications

Detailed Information

Safety Signal Detection of Antidepressants : 항우울제 부작용 실마리정보 탐색: 건강보험청구자료와 자발적부작용보고자료 비교연구
A Comparison of Korean National Health Insurance Claims Database and Spontaneous Reporting System Database

Cited 0 time in Web of Science Cited 0 time in Scopus
Authors

JIANG XINYING

Advisor
홍송희
Issue Date
2021
Publisher
서울대학교 대학원
Keywords
signal detectionpharmacovigilanceHIRAKIDS KAERSantidepressants실마리정보 탐색약물감시데이터 마이닝건강보험청구자료자발적부작용보고자료항우울제
Description
학위논문(석사) -- 서울대학교대학원 : 약학대학 약학과, 2021.8. 홍송희.
Abstract
자발적부작용보고 시스템 (spontaneous reporting systems, SRS)에 비해 국민건강보험공단 청구자료 (HIRA)는 모든 피보험자의 진단 및 처방 정보가 포함되어 있지만, 약물 안전 감시에 적용되는 경우가 많이 없었다. 본 연구의 목적은 건강보험청구자료 (HIRA)와 자발적부작용보고자료 (KAERS)를 이용하여 항우울제 사용으로 인한 실마리정보를 탐색하고, 두 시스템에서 발견한 실마리정보의 특징을 비교하는 것이다.
본 연구에 사용된 자료는 2017년 HIRA와 KAERS 데이터이다. HIRA데이터의 경우는 실마리정보를 탐색하기 전에 모든 약물과 약물이상반응 (ADR)을 후향적 페어링하고 약물-이상반응 조합 (drug-ADR pair)을 추출했다. 두 시스템의 약물-이상반응 조합에 대하여 여러 가지 실마리정보 지표를 계산한 뒤 실마리정보의 중류 (class), 부작용일반성지표 (common ADR coverage, CAC)또는 허가상항내재지표 (labeling information coverage, LIC)측면에서 두 시스템을 비교 분석해 보았다. 부작용일반성지표는 흔한 ADR의 비율로 측정하고 허가상항내재지표는 mAP (Mean Average Precision)로 측정했다. 또한, 제약회사의 제품설명서 및 허가사항에 포함되지 않은 실마리정보는 protopathic bias평가또는 RR (Relative risk)평가로 확인되었다. Protopathic bias평가할때 LEOPARD (Observational Profiles of Adverse Events Related to Drug)를 이용했다. 그리고 성별, 나이, 시간대에 따라 실마리정보 분포의 변화를 관측하였다.
KAERS 데이터베이스에서 총 5,992건의 약물-이상반응 조합을 이용하여 총 51개 실마리정보를 발견했다. 제약회사의 제품설명서 및 허가사항에 포함되지 않은 실마리정보는 없었다. HIRA 데이터베이스에서 총 108,570건의 약물-이상반응 조합 만들어 총 62개 실마리정보를 발견했다. 이 중 5개는 제약회사의 제품설명서 및 허가사항에 포함되지 않은 실마리정보였다. KAERS에서 항우울제가 더 많은 장기 시스템의 장애와 관련이 있음을 보였다. KAERS에서 더 높은 mAP (EB05의 mAP : 1.00 [K] VS. 0.983 [H])를 보였지만 일반적인 부작용과 관련된 실마리정보는 HIRA보다 더 많이 발견되었다 (68.63 % [K] VS. 29.03 % [H]). 실마리정보를 확인할 때 LEOPARD를 통해 duloxetine과 myelopathy의 조합은 protopathic bias때문에 생기는 것으로 확인되었다 (P-value=0.01026). 각 연령 및 성별 그룹에서 실마리정보 탐색시 HIRA는 항상 KAERS보다 낮은 CAC와 LIC를 보였다. HIRA에서 시간대가 단축함에 따라 CAC는 감소하고 (29.03%, 27.87%, 27.27%) LIC는 증가했다 (0.983, 1.0, 1.0).
본 연구를 통해 실마리정보 탐색할 때 건강보험청구자료와 자발적부작용보고자료에서 발견한 실마리정보는 서로 다른 프로파일을 보였다. 건강보험청구자료를 통해 허가사항에 포함되지 않은 실마리정보를 발견되었으며 추가 연구를 통해 확인해야 한다. 앞으로 약물감시 업무에서 실마리정보 탐색 시 HIRA같은 건강보험청구자료를 더 많이 적용해야 할 것이다.
Background: The spontaneous reporting system (SRS), such as the Korea Adverse Event Reporting System (KAERS), has a limitation in detecting all safety signals because the reports do not come from all drug uses. On the other hand, the claims data of the Korean National Health Insurance Review & Assessment (HIRA) has information on drug-induced conditions for all insurance-covered patients along with their prior prescription records, which helps investigate the temporal association between drugs and adverse drug reactions (ADRs). Therefore, complementing the HIRA to KAERS for drug safety signal detection would generate a more substantial list of safety signals than KAERS alone. This study has the following objectives: 1) compare the profiles of the signals (signal classes, common ADR coverage (CAC), and labeling information coverage (LIC)) detected in HIRA and KAERS databases; 2) verify the validity of the signals detected but not covered by the labeling information with protopathic bias evaluation and relative risk (RR) assessment; 3) determine whether the signal profile depends on demographics (age and gender) and different time windows (4, 8, 12 weeks) used to define the prior drug exposure.
Methods: ADR signal detection on the KAERS and HIRA databases (1st January 2017 to 31st December 2017) was conducted with Bayesian and non-Bayesian methods. The signal classes were constructed based on System Organ Class (SOC) as well as types of antidepressants. CAC was computed as the proportion of common ADRs among all the signals detected. LIC was represented by the mean average precision (mAP). Protopathic bias was controlled using Longitudinal Evaluation of Observational Profiles of Adverse Events Related to Drugs (LEOPARD). RR for each drug-ADR combination was based on prescription drug events and follow-up of ADR conditions.
Results: The numbers of signals detected in the KAERS and HIRA databases were 51 and 62, respectively. Most of the signals detected in KAERS consisted of autonomic nervous system disorders of SOC (N=17, 33.3%) and TCA (N=21, 41.2%) antidepressants while those in HIRA consisted of central & peripheral nervous system disorders of SOC (N=31, 50%) and SSRI (N=22, 35.5%) antidepressants. HIRA had 5 of the signals detected that are not found in the drug labeling information while KAERS had 0. The signals detected from KAERS had a higher CAC (68.63% [K] VS. 29.03% [H]) as well as a higher LIC (mAP for EB05: 1.00 [K] VS. 0.983 [H]) than those from HIRA. The unlabeled signal of myelopathy on duloxetine was a protopathic bias (P-value = 0.01026). Three of the four unlabeled signals did not show a statistically significant association between drug events and ADRs (lower bound of RR < 1). As for demographic subgroups, HIRA always showed lower CACs and LICs than KAERS. As for different time windows of drug-ADR pairing in HIRA, CACs decreased (29.03%, 27.87%, 27.27%) with a narrower window while LICs increasing (0.983, 1.0, 1.0).
Conclusion: The safety signals detected for antidepressants in HIRA (healthcare claims database) and KAERS (SRS) databases exhibited different signal profiles. The signals detected but not covered by drug labeling information, which were only detected in HIRA, need to be verified with further research. Safety signal detection in both healthcare claims and SRS databases would provide additional regulatory insight for pharmacovigilance.
Language
eng
URI
https://hdl.handle.net/10371/178944

https://dcollection.snu.ac.kr/common/orgView/000000166486
Files in This Item:
Appears in Collections:

Altmetrics

Item View & Download Count

  • mendeley

Items in S-Space are protected by copyright, with all rights reserved, unless otherwise indicated.

Share