Semiparametric Methods for the Analysis of Progression-Related Endpoints

摘要: Use of progression-free survival in the evaluation clinical interventions is hampered by a variety issues, including censoring patterns not addressed usual methods for analysis. Progression can be right-censored before or interval-censored between inspection times. Current practice calls imputing events to their time detection. Such an approach prone bias, underestimates standard errors andmakes inefficient use data at hand. Moreover composite outcome prevents inference about actual treatment effect on risk progression. This thesis develops semiparametric and sievemaximum likelihood estimators more formally analyze progression-related endpoints. For special case where death rarely precedes progression, Cox-Aalen model proposed regression analysis time-to-progression under intermittent inspection. The general setting considering both progression examined with Markov Cox-type illness-death various schemes. All resulting globally converge truth slower than parametric rate, but finitedimensional components are asymptotically efficient. Numerical studies suggest that new perform better imputation-based alternatives moderate large samples having higher rates censoring.