邊際充分成份原因模型下辨識與估計拮抗機制

Identifying and Estimating the Antagonism with Marginal Sufficient Component Cause Model

周孟穎(Meng-Ying Chou)

指導教授 ： 林聖軒

國立交通大學/理學院/統計學研究所/博士(2020年)

摘要

充分成份原因是因果推論領域用來探討發生結果的可能原因的方法。在流行病學上為了減少疾病的發生，需要考慮造成疾病的可能原因。而充分成份原因適合用來描述這些原因與風險因子之間的作用機制。許多研究文獻由充分成份原因的架構下，提出方法來檢定兩個因子的交互作用。並且也有學者把充分成份原因和反事實結果結合，在經驗法則之下來估計各種機制的大小。然而現有的估計方法建立在因子對結果有沒有預防效果或是單調遞增的條件的假設，在此假設之下，估計的機制只可以一併估計協同作用與促效作用，或是由對二元變數重新編碼來估計拮抗作用。同樣的在檢定交互作用的方法，協同作用與拮抗作用無法同時被檢定。當拮抗作用與協同作用同時存在時，需要其它方法來一起估計與檢定。因此單調遞增假設下，無法由現有的方法中把拮抗機制從協同與促效機制辨識出來。
本研究經由去除單調遞增假設來延伸邊際充分成份原因模型，考慮因子對結果的影響有預防效果的可能，推論可以產生的邊際機制。本研究方法可以對邊際拮抗機制、邊際促效機制、邊際協同機制三種交互作用存在的情況下，辨識出拮抗機制，也同時估計與檢定三種機制。在考慮以上機制交互作用可能的七種組合情況，模擬分析的結果顯示誤差相當小，覆蓋率信賴區間的覆蓋率接近95%。此結果顯示如果資料的結構和mSCC模型的設定相同，本研究方法可以用來作為估計與檢定以上三種機制交互作用。於資料分析部分我們採用台灣世代研究的肝癌資料。考慮兩個因子的拮抗、促效與偕同機制存在的情況下使用mSCC模型來分析，結果顯示拮抗交互作用的檢定結果顯著，也就是說在HCV抑制HBV的機制然後導致肝癌發生可以經由mSCC模型來說明。

關鍵字

邊際充分成份原因模型 ； 充分成份原因模型 ； 反事實結果 ； 因子有預防效果 ； 單調遞增假設

並列摘要

Sufficient component cause (SCC) is the principle concept in causal inference research field. SCC is applied when one investigates the causes of a result.. In epidemiology, to reduce the prevalence of a disease, it is necessary to consider the possible causes. There had been many literatures which provided methods for testing the effect of two variables interaction. Some researchers integrate the SCC and counterfactual methods to measure and estimate the mechanistic interaction effect.
However these methods were based on two assumptions. One assumption in SCC scenario was risk factors had no preventive effect on the outcome. The other in counterfactual outcome scenario was the counterfactual outcome had monotonicity with respect to risk factors. Under the assumptions, there were only synergistic mechanism and agonistic mechanism. To estimate the antagonistic mechanism, one of the binary factor had to be recorded. The synergistic and agonistic mechanism could not be tested along with the antagonistic mechanism at the same time when the factor was recorded. It required another way to identify and test the agonistic and synergistic mechanism at the same time when the both mechanism was considered co-existed.
In this research, we extended the marginal sufficient component cause (mSCC) model by considering two risk factors had preventive effects on the response variable. We inferred the possible mechanism with preventive effect setting in mSCC model. With this setting, the marginal antagonism, agonism and antagonism could be identified with surrogate variables. They could be tested and estimated at the same time. We perform simulation to validate out method. The biases approximated to zero and the coverage rates were closed to 95% in each of the seven scenarios from the three mechanistic interactions combinations. And we applied the mSCC model to evaluate the antagonism in the Taiwanese cohort study of HCC data. The result suggested that there was significant antagonism among HCV and HBV that leaded to HCC when considering the three mechanisms.

並列關鍵字

Marginal sufficient component cause ； sufficient component cause ； counterfactual outcome ； o preventive effect assumption ； monotonicity assumption

參考文獻

參考文獻

1. VanderWeele, T.J., Invited Commentary: The Continuing Need for the Sufficient Cause Model Today. American Journal of Epidemiology, 2017. 185(11): p. 1041-1043.

2. Rothman, K.J., CAUSES. American Journal of Epidemiology, 1976. 104(6): p. 587-592.

3. VanderWeele, T.J. and J.M. Robins, The Identification of Synergism in the Sufficient-Component-Cause Framework. Epidemiology, 2007. 18(3): p. 329-339.

4. Lee, W.-C., Assessing Causal Mechanistic Interactions: A Peril Ratio Index of Synergy Based on Multiplicativity. PLOS ONE, 2013. 8(6): p. e67424.