Robertson C, Boyle P
Division of Epidemiology and Biostatistics, European Institute of Oncology, Milano, Italy.
Stat Med. 1998 Jun 30;17(12):1305-23. doi: 10.1002/(sici)1097-0258(19980630)17:12<1305::aid-sim853>3.0.co;2-w.
Age-Period-cohort models are widely used by epidemiologists to analyse trends in disease incidence and mortality. The interpretation of such models is fraught with difficulty in view of the exact linear dependency between the three variables. It is the purpose of this paper to review, compare and contrast some of the more common approaches to this problem based on Poisson regression and a linear model for the log rates. Also the results of using the different approaches on a single series of data on breast cancer incidence among females in Scotland from 1960-1989 are presented for comparison. Recommendations as to the merits and drawbacks of the approaches are also given in the conclusions. Models which are based upon the estimable contrasts such as local curvatures and deviations from linearity are most suitable.
年龄-时期-队列模型被流行病学家广泛用于分析疾病发病率和死亡率的趋势。鉴于这三个变量之间存在确切的线性相关性，对此类模型的解释充满困难。本文旨在回顾、比较和对比基于泊松回归和对数率线性模型的一些更常见的解决该问题的方法。此外，还给出了对1960 - 1989年苏格兰女性乳腺癌发病率的单一数据集使用不同方法的结果以供比较。结论中还给出了关于这些方法优缺点的建议。基于可估计对比（如局部曲率和线性偏差）的模型是最合适的。
