大數據研究之機會與限制

The Opportunities and Limitations of Big Data in Clinical Research

近幾年「大數據」已成為臨床和公共衛生研究的熱門詞彙，不同領域的專家給予「大數據」的定義也有所不同，泛指「資料量的龐大」、「多樣的資料來源」和「非結構化」作為主要特色。本文聚焦在大數據最重要的核心：「連結多樣資料來源」而獲得有效推論並應用在臨床醫療與公共衛生。利用大數據作推論時要使用的觀察性資料研究方法是最重要的，從此來看，流行病學方法是大數據研究的基礎，龐大的資料量基本上可以減少推論的隨機誤差，卻不能將系統誤差加以控制，需透過合適的研究與分析方法才能達到正確的推論，否則「大數據」可能得到精確卻錯誤的結論，本文將以文獻中的例子進行說明，並對於使用健康資料的隱私和保密議題進行討論。

＂Big data＂ has become a popular term in clinical and public health research in recent years. As there is no authoritative definition about ＂big data,＂ different authors and institutions have defined its scope and content differently, with ＂large amount of data,＂ ＂variety of data sources,＂ and ＂unstructured data＂ as commonly cited attributes. In this article the authors focused on the most important ingredient of big data concept – utilizing data from disparate data sources and taking advantage of advance in information technology to draw valid inferences that are useful to medical and public health practice. While big data may hold promising potential to generate and test hypothesis with massive amount of data, the fundamental principles of making inference with observational data still apply. From this perspective the discipline of epidemiology is the foundation of current ＂big data＂ research. Research with large amount of data may substantially reduce random error, but systematic errors cannot be addressed by the size of the datasets and need to be reduced through sound research methods and robust analysis. Without appropriate study design and analysis, large datasets may most likely yield precise but wrong answers. Some examples in published literature are utilized in the article to illustrate the above principles. Lastly, privacy and confidentiality considerations for research utilizing health data are discussed.