二项抽样下相对风险比的区间估计比较.pdfVIP

  摘 要 相对风险比是流行病学研究中的重要指标之一，它是指流行病学研究中暴 露于某种危险因子的群体和未暴露于该危险因子的群体的发病率之比，体现了 暴露与疾病的关联强度。精确地得到相对风险比指标的区间估计，对病因推断 具有重要意义。但是，由于相对风险比指标的估计量是两个概率值的估计量的 比，一般很难得到它的精确分布，因此在已有文献中都会采用渐近方法估计相 对风险比的置信区间。本文讨论了二项抽样下相对风险比区间估计的方法，包 括文献中已有的 Delta 方法，对数转换方法，Fieller 定理的改进方法，基于似然 比检验的方法，基于得分检验的方法，以及本文通过鞍点逼近方法构造的置信 区间，并运用蒙特卡罗模拟对这六种方法进行较为全面的比较。模拟结果的评 价准则主要是运用实际覆盖率和名义水平的接近程度以及平均置信区间长度两 个方面的比较，从而确定各个方法的适用情况和优劣程度。最后，本文运用两 个实际的例子进行更为直观的说明。 关键词：流行病风险指标；相对风险比；区间估计；鞍点逼近   1   Abstract The relative risk (RR) between the exposed and the non-exposed groups is certainly one of the most important indices to quantify the strength of the association between a given disease and a suspected risk factor in epidemiological studies. To obtain an accurate interval estimation of RR is very important to give the deduction of the cause of disease. But as a ratio of two probabilities, it’s hard to get its exact distribution. So previous literature concentrated on the asymptotic methods of interval estimation. This thesis discusses the interval estimation methods under binomial sampling. The methods are including delta method, logarithmic transformation, improved Fieller’s theorem , likelihood test transformation, score test transformation, and the interval estimation of RR based on saddlepoint approximations. The six methods are compared by extensive Monte Carlo simulation. Criteria appropriate to the evaluation of the six methods include: closeness of the achieved coverage probability to its nominal value; expected interval wi