Grant: Discovery Grants Program - Individual

One goal of survival analysis is to model the distribution of the failure time measured between an initial event and a terminal event which includes numerous applications to medical science. A statistically challenging feature of survival analysis is that not all subjects' failure times are adequately observed. In an incident cohort study with follow-up, subjects, who are condition free, enter the study, are followed until onset of the condition and are then subsequently followed to their failure event, their withdrawal date from the study or to the end of the study period. The observed periods of time are known as right-censored failure times. In contrast, in a prevalent cohort study with follow-up, subjects are screened for a condition and only those individuals who are alive and have screened positively are selected into the cohort. Those prevalent subjects are then followed forward until they either withdraw from the study, fail during the follow-up period or survive to the end of the study. We refer to the observed durations of prevalent cohort subjects as left-truncated right-censored failure times as the prevalent cohort subjects were sampled according to a screening test and their failure event is known to occur after the end of the study follow-up period. When sampling from a database or collecting data from both incident and prevalent cohort studies with follow-up, the observed durations are right-censored or left-truncated right-censored, respectively. Although the two different studies yield different types of data, it is key that statistical procedures use all available data to provide the most accurate statistical inferences. A secondary feature for some types of prevalent cohort data is that the reported onset dates of the subjects recruited into the study may only act as proxies for the true unobserved onset dates. Although survival analysis methods have been developed for individual cohort data sets, few techniques have been explored for combined cohort data sets. Through the funding support of the NSERC Discovery grant, I will develop novel survival analysis methodologies that utilize all data collected from incident and prevalent cohort studies with follow-up simultaneously. I will investigate how uncertainty in the observed onset dates of prevalent cohort studies affect the standard survival methodologies in addition to developing new statistical models which incorporate the measurement error into the survival analyses. The statistical procedures developed through my research program will not only extend existing survival analysis procedures to the combined cohort setting but will benefit applied researchers in the fields of health/medicine. My research program will provide the statistical techniques for estimating survival using all available data from multiple cohort studies as well as the techniques to model the survival time of conditions with insidious onset processes where the true onset dates are unknown.