暴露與疾病關聯之五區域假說檢定

Abstract

描述暴露與疾病之間的關係，是流行病學研究的中心課題。為了這個目的，流行病學家常會使用勝算比指標，呈現勝算比之點估計、p值以及信賴區間。本研究將勝算比參數空間劃分為五區域，分別對應於強保護因子、弱保護因子、無相關、弱危險因子及強危險因子。隨著五區域的劃分，能夠簡潔清楚地描述暴露與疾病的關係，包含暴露與疾病相關與否，以及方向與強弱（假如有相關的話）。我們特別為此五區域的劃分量身訂作出一套統計方法，包含假說檢定、信賴區間以及樣本數計算。我們會論證五區域方法，與傳統方法相比能夠更有統計效率地描述暴露與疾病之間的關係。我們使用本研究方法重新分析三筆流行病學案例。我們提供簡單易用的R函數程式給研究者使用。五區域方法值得推薦為流行病學資料分析的常規方法。

Characterizing exposure-disease associations is a central issue in epidemiology. For this purpose, epidemiologists often adopt the index of odds ratio, presenting its point estimate, p-value and confidence interval. In this study, the parameter space of the odds ratio is to be partitioned into five regions, corresponding to ‘strong protective factor’, ‘weak protective factor’, ‘no association’, ‘weak risk factor’, and ‘strong risk factor’, respectively. With the five regions demarcated, a putative exposure-disease association can be described succinctly and clearly, including its presence or absence, and its direction and strength (if any association). The authors develop a suite of statistical methodologies tailored specifically to such five-region demarcation, including methods for hypothesis testing, confidence interval estimation and sample size calculation. The authors show that the five-region methods can be more statistically efficient than the traditional methods in describing exposure-disease associations. Three published results are re-analyzed to demonstrate the methodologies. Easy-to-use R code is provided. The five-region methods are recommended for routine use in epidemiologic data analysis.