The impact of bias on test decisions in survival analysis
| Autor: | Rückbeil, Marcia Viviane |
|---|---|
| Přispěvatelé: | Hilgers, Ralf-Dieter, Willmes-von Hinckeldey, Klaus |
| Jazyk: | angličtina |
| Rok vydání: | 2019 |
| Předmět: | |
| Zdroj: | Aachen 1 Online-Ressource : Diagramme (2019). doi:10.18154/RWTH-2020-00969 = Dissertation, Rheinisch-Westfälische Technische Hochschule Aachen, 2019, Kumulative Dissertation |
| DOI: | 10.18154/rwth-2020-00969 |
| Popis: | Dissertation, Rheinisch-Westfälische Technische Hochschule Aachen, 2019; Aachen 1 Online-Ressource : Diagramme (2020). = Dissertation, Rheinisch-Westfälische Technische Hochschule Aachen, 2019 Randomized controlled trials are considered the gold standard for establishing the efficacy of a new intervention with a low risk for bias. Chronological bias and selection bias can, however, lead to biased results even in randomized controlled trials. The choice of a suitable randomization procedure can reduce the expected influence of bias on the test decision and is therefore an essential part of the study design. The objective of this thesis is to develop a method for the selection of a suitable randomization procedure in the presence of bias in survival studies. In this thesis, a selection bias model for exponentially distributed survival data and a model for the combined representation of selection bias and chronological bias in survival studies are developed. For the comparison of different randomization procedures, the expected type I error probability is considered as the evaluation criterion. For this purpose, an analytic formula of the biased distribution of the F-statistic is derived based on the first bias model. For the second generalized bias model, an approximation formula for the biased distribution of the log-rank statistic is obtained. Based on the biased distributions of the test statistics, the probability of rejecting the null hypothesis can be calculated in dependence of the realized randomization sequence. If the null hypothesis is true, this probability corresponds exactly to the type I error probability. The presence of selection bias leads to an increase in the expected type I error probability for all randomization procedures. It can be seen that randomization procedures with stronger balancing behavior lead to a more serious increase. In the presence of both chronological bias and selection bias, no general statement can be made as to which randomization procedures adequately controls the expected type I error probability. In this case, a study specific evaluation must be performed which takes into account the anticipated risk for selection bias and chronological bias. Published by Aachen |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|