Effects of a Misattributed Cause of Death on Cancer Mortality.

Misattribution of Cause of Failure;Competing Risks;Mortality Model;Isotonic Estimation;Kullback-Leibler Divergence;Statistics and Numeric Data;Science;Biostatistics

Ha, JinkyungTaylor, Jeremy M. ;

University of Michigan

关键词: Misattribution of Cause of Failure;    Competing Risks;    Mortality Model;    Isotonic Estimation;    Kullback-Leibler Divergence;    Statistics and Numeric Data;    Science;    Biostatistics;   

Others  :  https://deepblue.lib.umich.edu/bitstream/handle/2027.42/89748/jinha_1.pdf?sequence=1&isAllowed=y

瑞士|英语

来源: The Illinois Digital Environment for Access to Learning and Scholarship

【 摘 要 】

The cause for the observed trends in prostate cancer mortality is unclear. Several authors have indicated that incorrectly classified causes of death have played a role in recent mortality trends. This dissertation is devoted to analyzing competing risks survival data with a misclassified cause of death and evaluating the hypothesis that misattribution of cause of death provides partial explanation for mortality trends.In the first project, we derive nonparametric maximum likelihood estimators (NPMLE) for cause-specific cumulative hazards. It is shown that constrained NPMLE obtained through EM algorithm is not consistent in continuous time setting. On the other hand, naive NPMLE is consistent although it cannot be guaranteed to be non-decreasing. We also investigate other isotonic approaches such as the supremum (SUP) method and the Pooled-Adjacent-Violators (PAV) algorithm.In the second project, we consider semiparametric proportional hazards regression models with covariates. Due to a misattributed cause of death, the standard profile semiparametric maximum likelihood approach cannot be directly used to eliminate the baseline hazards in estimating regression coefficients. We propose Kullback-Leibler estimating equation (KLE) which does not require the parametric assumptions for baseline hazards.Finally, in the third project, we apply these estimation approaches to a mortality model and assess the effect of attribution bias on the recent trend in mortality rates with data obtained from the Surveillance, Epidemiology and End Results (SEER) Program. We find thatthe shape of mortality is only altered under a calendar time-varying misattribution mechanism.