"Novel Adaptive Trial Designs for Studies with a Composite Endpoing of Morbidity and Mortality"

Public Health/Biostatistics

Committee: 
Chung-Chou H. Chang, (advisor) Department of Biostatistics, Department of Medicine
Andriy I. Bandos, Department of Biostatistics
Lu Tang, Department of Biostatistics
Victor B. Talisa, Department of Critical Care Medicine

Abstract:
Adaptive designs that allow for prospectively planned modifications during data collectionhave been widely used in clinical trials in recent years. Comparing to non-adaptive design, adaptive design may detect a true treatment effect with a lowersample size under the samestatistical power. In this dissertation, Ipropose several novel adaptive trial designs for studies with a composite endpointof morbidity and mortality.In the first project, Ideveloped a novel Bayesian response adaptive randomization (RAR) design for a composite endpointof organ support free days(OSFD). Iapplied thismethod to design a multicenter, unblinded, phase II or III trial for treating sepsis patients admitted to the intensive care units with the endpoint ofOSFDby Day 28.Comparingwith other existing designs, non-adaptive or adaptive, the proposed methodallocatesmore patients tothe best performing treatmentarm(s)and shows higher power. In the secondproject, Iextendedtheadaptation method in the first project to address the heterogeneity of treatment effects stemming from patients' baseline characteristics. Four sepsis phenotypes were considered as categorical covariates in thisBayesian covariate-adjusted response adaptive(CARA)design. I also extended win ratio method to the adaptive designand incorporatedstratum-specificwin ratiosinto theadaptive randomization(WR-CARA). Simulations showedthatboth Bayesian CARAand WR-CARAmethods resulted in a higher proportion of patients assigned to the best performing treatmentarm(s)in each sepsis phenotype when compared with the RARmethod. Bayesian CARA had the highest statistical power among those methodscomparedbecause itbest capturesthe underlying OSFD distribution. The WR-CARAapproach is a good alternative when the underlying distribution is unknown.

Public health significance:
All adaptive methods we proposed allocate more patients to the superior arm comparing withthe existing methods. The two methods in the second projectincorporatepatients’baseline characteristics in the design so that heterogenous treatment effects are taken into consideration. This dissertationwill promote the development of adaptive designs and innovation for public health and medical research.