Department of Epidemiology and Population Health, Albert Einstein College of Medicine, Bronx, NY 10461, USA.
Semin Nucl Med. 2010 Sep;40(5):357-63. doi: 10.1053/j.semnuclmed.2010.04.001.
The randomized clinical trial is widely viewed to be the gold standard for evaluation of treatments, diagnostic procedures, or disease screening. The proper design and analysis of a clinical trial requires careful consideration of the study objectives (eg, whether to demonstrate treatment superiority or noninferiority) and the nature of the primary end point. Different statistical methods apply when the end point variable is discrete (counts), continuous (measurements), or time to event (survival analysis). Other complicating factors include patient noncompliance, loss to follow-up, missing data, and multiple comparisons when more than 2 treatments are evaluated in the same study. This article provides an overview of the basic statistical approaches for analyzing clinical trials with binary, continuous or time-to-event outcomes as well as methods for handling protocol deviations due to noncompliance and early drop-out. Issues surrounding the determination of sample size and power of clinical trials are also discussed.
随机临床试验被广泛认为是评估治疗方法、诊断程序或疾病筛查的金标准。临床试验的正确设计和分析需要仔细考虑研究目标（例如，是否要证明治疗的优越性或非劣效性）和主要终点的性质。当终点变量为离散（计数）、连续（测量）或事件时间（生存分析）时，适用不同的统计方法。其他复杂因素包括患者不遵守规定、随访丢失、缺失数据以及在同一研究中评估超过 2 种治疗方法时的多次比较。本文概述了用于分析具有二分类、连续或时间到事件结局的临床试验的基本统计方法，以及处理因不遵守规定和早期退出而导致的方案偏差的方法。还讨论了临床试验样本量和功效确定的相关问题。
