建立藥物不良反應的訊號探測方法 以全國藥物不良反應資料庫為研究資料

The Study of Signal Detection from National Reporting System of Adverse Drug Reaction Database in Taiwan

蘇純慧(Chun-Hui Su)

指導教授 ： 高純琇

國立臺灣大學/醫學院/臨床藥學研究所/碩士(2006年)

https://doi.org/10.6342/NTU.2006.02463

摘要

研究目的
藥物不良反應之主動通報系統的建立，是上市後藥品安全監視的主要方法，目的是在短時間內發現上市前未知的藥物不良反應或已知藥物不良反應的發生率的改變。此篇論文，應用文獻中已建構的兩種簡易資料探勘法在臺灣的藥物不良反應資料庫，進行訊號偵測。並且比較兩種方法的可行性。
研究方法
收集全國藥物不良反應通報系統中九十二至九十四年度的通報案例，所有案例經由評估和譯碼。可疑藥品以Anatomical Therapeutic Chemical (ATC) code做為編碼，不良反應以Medical Dictionary for Medical regulatory Activities (MedDRA) 的Preferred term (PT) code做為編碼。將由各通報案例產生出的ATC-PT配對，以相對通報比率法 (proportional reporting ratios, PRRs) 與相對通報勝算比 (reporting odds ratio, RORs)進行訊號偵測。
研究結果
92-94年三年度納入分析的通報案例共8439件。通報案件的病患，男女比例為0.97：1，平均年齡為52.1±21.7歲，以70至80歲最多(17.2%)。通報案件的懷疑藥品之藥理分類，以神經系統作用藥物最常見(30.3%)，抗感染藥物次之(27.2%)，心血管腎臟用藥佔第三(11.9%)。根據MedDRA的System Organ Class (SOCs)，最常被通報的不良反應是皮膚與皮下組織方面的疾病(32.2%)，神經系統方面的異常次之(10.5%)，胃腸道的疾病為第三(10.0%)。不良反應的型態，以B型態為主(67%)。在藥物與不良反應相關性的評估方面，「確定」、「極有可能」與「可能」者，共佔91.3%。
三個年度的通報案例總計配對出14,072筆，共7,264種不同的ATC -PT配對，其中有974種配對(13.4%)的出現次數大於等於3。此974種配對經過訊號偵測後，由PRR法偵測出的訊號有648個配對(8.9%)，由ROR法偵測出的訊號有722個配對(9.9%)。其中有647個配對同時可以被此兩種方法偵測到，251組配對經PRR法與ROR法運算皆未被視為訊號。在進行第二階相對於同類藥品的訊號偵測，則同時可被PRR法及ROR法偵測到的訊號有244個配對(3.4%)。
討論
在647個同時可以被兩種方法偵測到的配對中，有17個配對(2.6%)是在文獻中，沒有明確記載其藥物不良反應相關性者，需要進一步研究確定藥物不良反應相關性。另外關於statin類與fibrate類降血脂藥物造成橫紋肌溶解症、顯影劑造成過敏性休克等藥物不良反應也應列入後續追蹤的項目之一。
結論
目前電腦化的訊號偵測資料探勘法，並無單一的閾值設定標準。使用PRR法及ROR法做為臺灣藥物不良反應資料庫的訊號偵測法，可能已經足夠。訊號偵測法可以當作為傳統藥物偵測法很好的輔助工具，資料探勘法產生的訊號，仍需要進一步的文獻回顧與研究，以確定其藥物不良反應的相關性與臨床意義，以建立用藥安全資訊。

關鍵字

藥物不良反應 ； 藥物不良反應資料庫訊號偵測 ； 資料探勘

並列摘要

Purpose
Spontaneous reporting systems (SRS) for adverse drug reactions (ADRs) remain a cornerstone of the Pharmacovigilance to early detect new ADRs and changes of the frequency of ADRs that are already known. In this study, we aimed to apply two simple data-mining algorithms developed in literature to the safety signal detection for the ADR database built in Taiwan. The feasibility of the two different methods was compared as well.
Methods
Reported cases collecting from 2003 to 2005 in the database of National reporting System of Adverse Drug Reactions in Taiwan were reviewed and coded with ATC code for suspected drugs and with MedDRA code for the reported adverse drug reactions. Preferred term (PT) of the MedDRA coding was used to represent the ADR. The priciples of proportional reporting ratio (PRR) and reporting odds ratio (ROR) were applied to ATC-PT pairs generated from the reports for signal detection.
Results
A total of 8439 reports over the 3-year period were included. The male/female ratio of patients was 0.97:1 and the average age of patients was 52.1±21.7 years old. The largest patients group was in the age range of 70-80 yrs (17.2%). The most frequently reported suspected drugs were the drugs used in nervous system (30.3%), followed by antiinfectives (27.2%), and cardiovascular- renal drugs (22.9%). According to MedDRA’s system organ classification, the most often reported ADRs were skin and subscutaneous tissue disorders (32.2%), followed by nervous system disorders (10.5%) and gastrointestinal disorders (10.0%). Type B ADRs are the main type of ADRs (67%). The causality of suspected drugs and ADRs was assessed as certain, probable, and possible for 91.3% of the cases.
A total of 14,072 ATC-PT pairs were generated from the reported cases, which belong to 7,364 different ATC-PT pairs. Among them, 974 (13.4%) ATC-PT pairs have the frequencies at least 3. After calculation, 648 signals (8.9%) were generated by PRR method and 722 signals (9.9%) were from ROR method. It is interesting to see that 647 different ATC-PT pairs (8.9%) were detected by both methods. Only 251 different pairs were not considered as signal by either method. If we took the 723 ATC-PT pairs to do the calculations with respect to the same drug class (based on ATC classification), a total of 244 signals (3.4%) were detected by both PRR and ROR methods.
Discussion
After searching, the relationship between suspected drug and ADR for 17 ATC-PT pairs (2.6% out of the 647 pairs) was not found in literatures. They may be a good candidate for further investigation. Moreover, antihyperlipidemia drug such as statins and fibrates related rhabdomyolysis and contrast medias related anaphylaxic shock are recommended for further carefully monitoring.
Conclusion
There is no single standard for signal detection method so far. The PRR method and ROR method may be good enough for safety signal detection for the ADR database of Taiwan. However, statistic approach signal detection method should be considered as a potential supplement to, and not substitute for, traditional pharmacovigilance strategies. Further clinical/pharmacological knowledge- based evaluation and research are required to confirm the casuality between the drug and the adverse drug reaction.

並列關鍵字

Adverse drug reaction ； ADR database ； signal detection ； data mining

參考文獻

1. Slone D, Shapiro S, Miettinen OS, Finkle WD, Stolley PD. Drug evaluation after marketing. Ann Intern Med 1979;90(2):257-61.

2. International Drug Monitoring: The Role of National Centres. Geneva: World Health Organization; 1972.

3. Taussig HB. A study of the German outbreak of phocomelia. The thalidomide syndrome. JAMA 1962;180:1106-14.

4. Klepper MJ. The periodic safety update report as a pharmacovigilance tool. Drug Saf 2004;27(8):569-78.

9. Egberts AC, Meyboom RH, van Puijenbroek EP. Use of measures of disproportionality in pharmacovigilance: three Dutch examples. Drug Saf 2002;25(6):453-8.