利用巨量資料分析長期藥物曝露與癌症風險之間的關聯性

Big Data Approach To Explore The Associations Between Long Term Drug Exposures And Cancer Risk

烏斯馬(Usman Iqbal)

指導教授 ： 李友專

臺北醫學大學/醫學科技學院/醫學資訊研究所/博士(2016年)

摘要

醫療健康觀察資料是大家熟知巨量資料同時日益被廣泛運用。在藥物流性病學上，針對有興趣的醫療議題以及為了迅速獲得結果資料，我們需要發展了一套高效能及具有成本效益的系統來即時探索不同變數。我們依據現在藥物流行病學的統計方法和第三等級實證醫學（EBM）研究設計，設計了一套通用分析方法。

我們使用的資料庫為台灣國民健康保險署（健保署）所提供的1998 到 2011年的百萬人抽樣歸人檔資料。在研究初期先建立資料庫、並發展前端與後端的系統，最後驗證這些結果。
此系統為線上系統，其包含以下五個步驟：定義世代、定義結果、定義介入因子、共同變數選擇及結果輸出。在藥物流行病學的巨量資料分析上，研究人員往往有不同的假設，而在系統的步驟引導之下，不同假設的研究將不會遺漏應該選擇的變數。（參閱http://10.1.2.125/）

透過OMOSC系統的線上分析工具可以有規模性探索及檢視長期用藥與癌症關聯，因此將有利於針對長期用藥與癌症的風險進行大規模的線上研究，同時也可直接地幫助缺乏資料探勘技術的醫護專業人員進行研究。

此建構的線上系統將利用巨量資料並自動建立病例組與對照組，來了解慢性長期用藥暴露與癌症的風險。結果會以勝算比(Odds ratio, OR)顯示，同時可藉由控制干擾變數來獲得調整後的勝算比(Adjust Odds ratio, AOR) 及95%的信賴區間(Confidence Intervals ,CI)。我們使用SAS統計軟體分析相同的資料庫來驗證OMOSC系統的結果。它可以幫助進行大量線上研究以節省時間和具備成本效益。

對OMOSC系統來說分析藥物與癌症風險的巨大挑戰包含該系統需要容易執行、可實現平易近人的，可實現的，並可能對已使用藥物有所影響。而臨床試驗無法執行的因素包含文化、花費、道德、政治以及社會議題。所以本研究的大規模研究模型將藉由巨量資料分析，在健康照護產業裡扮演很重要的角色，此能提供絕佳的機會去解決科技、資訊和使組織在藥物評估上的議題，去影響更多更廣泛的領域。

關鍵字

健康資訊模型 ； 長期藥物 ； 癌症風險 ； 致癌性 ； 藥物流行病學 ； 藥物毒性 ； 藥物安全 ； 資訊設計 ； 藥物再定位 ； 對照組的觀察性研究 ； 實證研究

並列摘要

The health care observational data which is also now known as bigdata is leveraging day by day. To utilize it in an effective way for generating health outcomes of interest specifically for pharmacoepidemiology, we need to develop a system which could be efficient and cost effective to do online explorations with different parameters.

We designed a generic methodology based upon existing statistical and Evidence Based Medicine (EBM) level 3 research study designs method for pharmacoepidemiology studies. We used Taiwan Bureau National Health Insurance (BNHI) claim database one million cohort sample population from 1998 – 2011 year. The basic phases involved in this study were to create the database, develop back-end and front-end and then validate those results. This system is a web-based online which contains five steps: Define cohort, define outcome, and define intervention criteria and covariate assessment selections and final outcomes. It leads to the researcher to follow steps with different assumptions so, the research would not lost what and where they should select the parameters with different assumptions to explore big data for pharmacoepidemiology. (Visit at: http://10.1.2.125/)

This online analytical tool has capability to massively explore and visualize big data for long term use drugs and cancers through OMOSC system which will help to do mass online studies for long term use drugs and cancer risk. It would help to direct the health care professionals with lack of datamining skills to lead the study. The constructed online system would generate automatically case and controls by utilizing large databases for long term drug exposures and cancer risk. The results are shown in odds ratio (OR) and if selected some confounding factors then could also get adjusted odds ratio (AOR) for risk estimation with 95% Confidence Intervals (CI). We used SAS statistical software on the same dataset to validate the OMOSC system results. It could help to do massive online studies which will saves time and cost effective.

The OMOSC system will be capable for drugs and cancer risk evaluation on a societal scale which is a big challenge that is approachable, achievable, and has implications towards those developing or using medications. Since the clinical trials are impossible to conduct due to cultural, cost, ethical, political or social obstacles. Therefore, this kind of research model would play an important role in health care industry by providing an excellent opportunities for solving the technological, informatics, and organizational issues towards other broad domains of drugs evaluation by utilizing large-scale databases.

並列關鍵字

Health Informatics Model ； Long-Term Use Drugs ； Cancer Risk ； Carcinogenicity ； Pharamaco-Epidemiology ； Drugs Toxicity ； Drug Safety ； Informatics Design ； Drug Repositioning ； Observational Studies With Controls ； Evidence Based Research Studies

參考文獻

ABBOTT, R. D., & CARROLL, R. J. (1984). INTERPRETING MULTIPLE LOGISTIC REGRESSION COEFFICIENTS IN PROSPECTIVE OBSERVATIONAL STUDIES. American journal of epidemiology, 119(5), 830-836.

Austin, P. C. (2011). An Introduction to Propensity Score Methods for Reducing the Effects of Confounding in Observational Studies. Multivariate Behavioral Research, 46(3), 399-424. doi: 10.1080/00273171.2011.568786

Avorn, J. (2006). Dangerous deception—hiding the evidence of adverse drug effects. New England Journal of Medicine, 355(21), 2169-2171.

Benson, K., & Hartz, A. J. (2000). A comparison of observational studies and randomized, controlled trials. New England Journal of Medicine, 342(25), 1878-1886.

Black, N. (1996). Why we need observational studies to evaluate the effectiveness of health care. BMJ : British Medical Journal, 312(7040), 1215-1218.