- 學位論文
流行病學研究之因果機轉交互作用分析
Analysis of Mechanistic Interactions in Epidemiological Studies
摘要
如何評估生物交互作用，一直是流行病學家及臨床研究者相當關注的議題。傳統上，評估交互作用會檢定羅吉斯、布阿松、庫克司迴歸等累乘性模式的交互乘積項，由於發展完整及統計便利性，成為研究者探討生物交互作用的重要方法之一。近期，透過因果機轉交互作用來討論生物交互作用的方式，也獲得熱烈地討論。充分成份因果模式，將因果關係認為是因果機轉的集合體，若兩個因子參與同個因果機轉的作用，則認為存在此模式下的協同作用，即我們所稱的因果機轉交互作用。探討因果機轉交互作用的現有指標，包含「交互作用之相對超額風險」及「稜鏡」試驗，都使用在沒有失去追蹤及沒有競爭死因的世代研究。本論文中，以充分成份因果模式為主體，發展在不同流行病學研究設計下的因果機轉交互作用之檢定。在世代研究中，「互補對數」的連結函數恰可連接充分成份因果模式與世代資料，於是我們發展「互補對數迴歸」，可評估多階層因子或多因子間與疾病的關聯，並使得主效應項的迴歸係數指數化後將是校正後該暴露改變的「風險比」，檢定交互乘積項的迴歸係數則為檢定因果機轉交互作用。在病例對照研究中，若為罕見疾病，藉由勝算近似為對數風險，便能使用我們提出的「互補對數迴歸」進行多因子的探討；若為非罕見疾病，我們提出利用疾病盛行率的外部資訊來估計疾病風險，便可使用「稜鏡」試驗來進行因果機轉交互作用的檢定。我們呈現此方法在任何情形下，都能維持正確的型一誤差，而且有著優越的檢力。在長期追蹤研究中，我們考慮設限資料，提出「因果機轉交互作用」試驗，不需比例危險假設，甚至在危險曲線交叉時仍保持穩健的統計性質，比起「交互作用之相對超額風險」及「稜鏡」試驗有著更好的檢力，而且交互作用的訊號並不會隨著追蹤時間增加而消失。我們提出的方法都擁有良好的統計性質，若想探討暴露與疾病間的因果機轉交互作用，建議使用相對應的研究設計下的檢定方法。
並列摘要
The assessment of biological interactions is important for epidemiologists and clinicians alike. We traditionally test the cross-product terms in a multiplicative model, such as logistic, Poisson, or Cox regression, to assess the interactions. For its well-development and statistical convenience, the use of multiplicative model is prevalent in assessing biological interaction. Recently, using mechanistic interaction to assess biological interaction receives much attention. The sufficient component casue model conceptualizes causation as a collection of causal mechanisms. If two factors participate in the same causal mechanism, we will say there is syngergism between the two factors in the sufficient cause sense, that is, mechanistic interaction. Indices specific to mechanistic interactions have been proposed in a cohort study, including the ‘relative excess risk due to interaction’ (RERI) and the ‘peril ratio index of synergy based on multiplicativity’ (PRISM). In this thesis, we proposed the methods of assessing mechanistic interactions in different types of epidemiological studies. In cohort studies, we cast the PRISM test into a regression framework with the complementary log link. In a complementary log regression, the exponentiated coefficient of a main-effect term corresponeds to an adjusted ‘peril ratio’, and the coefficient of a cross-product term can be directly to test for mechanistic interaction. In case-control studies of rare diseases, log perils can be approximated by odds, and then a complementary log regression can be valid. In case-control studies of non-rare diseases, we proposed a method to incorporate information regarding disease prevalence to estimate disease perils, and then adopted a PRISM test to assess the mechanistic interaction. The proposed method can maintain type I error rates at nominal significant level, and have great powers in all scenarios. In long-term follow-up studies, we proposed the ‘mechanistic interaction test’ (MIT) for censored data. MIT can maintain reasonably accurate type I error rates, even when the hazard curves crossed others. MIT have greater powers than RERI and PRISM tests, and its synergy signals do not disappear as time progresses. To test for mechanistic interactions, we recommended using the proposed methods in this thesis in light of their desirable statistical properties.
參考文獻
延伸閱讀
- 陳政亮（2011）。流行病學與社會分析的因果關係。科技醫療與社會，(12)，233-242。https://doi.org/10.6464/TJSSTM.201104.0233
- Huang, J. W. (2010). A型H3N2流行性感冒病毒之基因演化與抗原性演化之關聯性研究 [doctoral dissertation, National Chiao Tung University]. Airiti Library. https://doi.org/10.6842/NCTU.2010.01063
- Lee, L. T. (2006). 流行病學與衛生政策分析. 慈濟醫學雜誌, 18(4), 253-257. https://doi.org/10.6440/TZUCMJ.200608.0253
- Lo, C. (2019). Methodology Development and Application of Untargeted Metabolomics [master's thesis, National Taiwan University]. Airiti Library. https://doi.org/10.6342/NTU201900431
- 林佳逸（2015）。Odds-based Causal Mechanistic Interaction Tests in Case-Control Studies〔碩士論文，國立臺灣大學〕。華藝線上圖書館。https://doi.org/10.6342/NTU.2015.10697