Development of the theory and program for the survival time model with covariates subject to measurement error.

点击翻译按钮获取中文摘要

The paper describing the theoretical aspects of the method for the proportional hazards measurement error model, or generally referred to as the Cox-regression model, has been published from the Environmental Health Perspectives and a FORTRAN program implementing the method from the Computer Methods and Programs in Biology and Medicine. We have also constructed a method for finding asymptotically unbiased estimates adjusting for the misclassification in discrete convariate models and reported the results at the Joint Conference for the International Society for the Envirobnmental Epidemiology and the Exposure Analysis (REsearch Triangle Park, North Carolina, 1994), which is under preparation for submitting for publication in certain European journal.Analysis of the effect of the radiation doses on the A-bomb survivors followed-up by the Radiation Effect Research Foundation is as important application of the results of this study project. Not very reliable method has been proposed by so … More me statisticians of RERF for the analysis adjusting for the uncertainty of the dose-estimation, which failed to obtain even variance estimation of the estimates. Considering the importance of the results of the analysis for social life and political decision, we have developed a theory to apply the method obtained from this research project to the A-bomb survivors data to find mor reliable estimates for the radiation dose effects on human mortality. The method basically assumes the Poisson regression model, like RERF,and proposes a simple correction to the likelihood, which yields asymptotically unbiased estimates with valid variance estimates. Since we have obtained the summary data from RERF,we will analyze it with the method after close investigation into the data structure.We have also obtained a magnetic tape including the Framingham Heart Study Follow-up Data from NIH to apply the developed method to find out the effect of cholesterol and/or blood pressure on the cardiovascular risk adjusting for the measurement error in those laboratory test results.Thus we have successfully finished the study project, published the main results from well circulated journals and are going to apply the results to some of the important cohort data. Less