阳江高本底地区群组研究1979～1995年随访数据合并与统计分析方法

Data pooling and statistical analyses of cohort studies (1979 1995) in Yangjiang,China.

目的讨论阳江高本底地区群组研究不同阶段的资料合并与癌症危险的统计分析。方法调查分阶段进行，１９７０～１９７８年为回顾性调查，１９７９～１９８６年为动态人群前瞻性观察，１９８７～１９９５年为定群研究，后两个阶段的数据已经计算机化。采用记录连接方法进行１９７９～１９８６年和１９８７～１９９５年随访数据的合并。危险分析基于详细的人年列表，分组变量包括性别、到达年龄（ａｔｔｅｎｄｅｄａｇｅ）、随访年份和剂量分组。利用标准的Ｐｏｉｓｓｏｎ回归方法估计超额相对危险和与对照组相比高本底地区每个剂量组的相对危险。采用似然比χ２近似方法计算显著性检验的双侧Ｐ值和９５％置信区间。结果由于某些成员的出生日期在两个随访阶段数据库中有所不同及其他原因，约有１０％的群组成员未能完成记录连接。我们讨论了合并随访数据的三种方法。结论癌症危险分析基于１９７９～１９９５年记录连接合并资料，包括１２５０７９人，累计观察１６９８３５０人年，１０４１５人死亡，其中１００３人为癌症死亡。癌症危险估计的结果将有另文专门讨论。

Objective\ Cohort studies on cancer mortality among inhabitants in the high background radiation area (HBRA) of Yangjiang,China, was started in 1972.The purpose of the investigation was to explore cancer risks associated with chronic exposure to low dose ionizing radiation.In this paper,we describe data pooling and methods of cancer risk estimation used in Yangjiang HBRA study. Methods HBRA epidemionogical surveys were carried out in consecutive periods,i.e.,retrospective survey (1970 1978),dynamic population follow up (1979 1986),and fixed cohort study (1987 1995).The surveys included tracing and folow up of cohort members,ascertainment of causes of death,and measurements of indoor/outdoor gamma exposure and individual cumulative doses.Record linkage was used to combine the follow up data of the dynamic population (1979 1986) and the fixed cohort (1987 1995).Risk analysis was based on detailed tabulation of person years stratified by sex,age at risk,follow up interval,and dose group.The basic data for each cell in these tables included the numbers of cancer deaths,other deaths interested,and person years.These data were supplemented by cell specific person year weighted means of cumulative doses and age at risk.The person year tabulations were prepared using DATAB in Epicure.Standard Poisson regression method was used to estimate excess relative risk per sievert,and relative risks for each dose groups in HBRA compared with control group.Significance test (two sided P value) and 95% confidence intervals were calculated by χ 2 approximation to likelihood ratio method.Risks were estimated using AMFIT in Epicure. Results\ Due to differences in date of birth of some subjects and other reasons,about ten percent of subjects failed to link.Conclusion\ Cancer risk analysis was based on the pooled data for the period 1979 1995,which included 125079 subjects,accumulated 1698350 person years,and observed 1003 cancer cases among 10415 deaths.Cancer risks are discussed in another paper.