The impact of dropouts on the analysis of dose‐finding studies with recurrent event data

Abstract

SUMMARYThis work is motivated by dose-finding studies, where the number of events per subject within a specified study period form the primary outcome. The aim of these studies is to determine the efficacy of a new drug compared to an active control or placebo. In particular, we are interested in identifying the dose-response relationship and the target dose for which the new drug can be shown to be simultaneously safe and as effective as the control.Given an outcome which is pain-related, we expect a considerable number of patients to drop out before the end of the study period. The impact of missingness on the analysis and diverse models for the missingness process must be carefully considered.The recurrent events are modeled as over-dispersed Poisson process data, with dose as a regressor. Additional covariates such as age may be included. Constant and time-varying rate functions are examined. Based on these models the impact of missingness on the precision of the target dose estimation is evaluated. Diverse models for the missingness process are considered, including dependence on covariates and number of events. The performances of five different analysis methods (a complete case analysis; two analyses using different single imputation techniques; a direct likelihood analysis; and an analysis using pattern-mixture models) are assessed via simulation studies. It is shown that the target dose estimation is robust if the same missingness process holds for the target dose group and the active control group. Furthermore, we demonstrate that this robustness is lost as soon as the missingness mechanisms for the active control and the target dose differ. Among the explored missing data handling methods it is shown that the direct-likelihood approach performs best, even when a missing not at random mechanism holds.

Sensitivity analyses for partially observed recurrent event data

“…In terms of coverage, all approaches and parameters except the nuisance parameter ϕ show proper coverage properties. The parameter ϕ shows under‐coverage for most scenarios considered, a limitation that has been noted previously . Overall, the asymptotic and bootstrap imputation approaches have very satisfactory properties when being compared with the benchmark approach under MAR: the direct likelihood approach.…”

Model‐based dose finding under model uncertainty using general parametric models

“…S obtained from such a model (at a particular timepoint of interest) then inherits the missing data assumptions of the underlying first stage model (see also [38] for a discussion of missing data in the context of dose-finding trials). Model-based dose finding methods, such as the extended MCP-Mod, provide better understanding of the dose-response relationship, generally translating into more accurate dose selection for confirmatory trials.…”

“…S obtained from such a model (at a particular timepoint of interest) then inherits the missing data assumptions of the underlying first stage model (see also [38] for a discussion of missing data in the context of dose-finding trials).…”

Dose Response Signal Detection under Model Uncertainty

“…For example, Neal (2006) and Wakana et al (2007) extended the original approach to Bayesian methods and Klingenberg (2009) applied the MCP-Mod approach to proofof-concept studies with binary responses. Benda (2010) proposed a time-dependent dose finding approach for repeated binary data, while Akacha and Benda (2010) studied the impact of dropouts on the analysis with recurrent event data. Several authors investigated extensions of the original MCP-Mod approach to response-adaptive designs; see Miller (2010); Bornkamp et al (2011); Tanaka and Sampson (2012).…”