Rossi G, Zanobetti A, Marchi M
Istituto di Fisiologia Clinica, C.N.R., Pisa.
Epidemiol Prev. 1995 Mar;19(62):90-8.
This work gives an overview of design and analysis of temporal studies using aggregated data in air pollution epidemiology. In the last years time series are often used to study the short-term association between ambient air pollution levels and aggregated health data. Health endpoints are usually daily mortality and/or daily hospital admission data from routine health registers. Air quality data are commonly obtained from one (or a few) fixed site monitoring stations. To detect the temporal association between the time-pattern in air pollution and the time-pattern in health data particular attention needs to be given to the autocorrelation structure, to the seasonality and long term trend in the data, and to the weather variables. Poisson regression with autocorrelated residuals is the suitable statistical method to analyze time studies. Furthermore, the pollutant variable can be analyzed at different lag-times to account for short latency periods in the manifestation of diseases. Studies with temporal aggregated data show the same disadvantages of the ecologic studies, although, in this case, confounding is less of a problem. Temporal studies usually are based on a large database, so that sufficient power can be achieved to detect even weak associations. Finally, the exposure information on subjects is often better characterized by short-term fluctuations in ambient air quality than is the case in geographic aggregations.
这项工作概述了在空气污染流行病学中使用汇总数据进行时间序列研究的设计与分析。在过去几年中，时间序列常用于研究环境空气污染水平与汇总健康数据之间的短期关联。健康终点通常是来自常规健康登记处的每日死亡率和/或每日住院数据。空气质量数据通常从一个（或几个）固定站点监测站获取。为了检测空气污染的时间模式与健康数据的时间模式之间的时间关联，需要特别关注自相关结构、数据中的季节性和长期趋势以及天气变量。具有自相关残差的泊松回归是分析时间序列研究的合适统计方法。此外，可以在不同的滞后时间分析污染物变量，以考虑疾病表现中的短潜伏期。使用时间汇总数据的研究显示出与生态学研究相同的缺点，不过在这种情况下，混杂问题不太严重。时间序列研究通常基于大型数据库，因此即使是检测微弱关联也能获得足够的效力。最后，与地理汇总情况相比，受试者的暴露信息通常通过环境空气质量的短期波动能得到更好的描述。
