This thesis addresses potential threats to the validity of observational epidemiological studies. Examples of these potential sources of bias are confounding, missing data, selection bias, and... Show moreThis thesis addresses potential threats to the validity of observational epidemiological studies. Examples of these potential sources of bias are confounding, missing data, selection bias, and measurement error. Although various methods have been developed to mitigate these biases, it is often unclear which methods can be used in which empirical settings. It is also common that issues discussed in methodological studies are overlooked in clinical research. Thus, we investigated problems ofmissing data, selection bias, and measurement error occurring in several specific observational settings and discuss how to optimally handle them. Show less
Over the last decades, increasingly so in the last years, epidemiological methods have been refined, making it challenging to keep abreast of all methodological developments. The choice of the data... Show moreOver the last decades, increasingly so in the last years, epidemiological methods have been refined, making it challenging to keep abreast of all methodological developments. The choice of the data analytical method directly influences the interpretation and clinical meaning of results of an analysis, yet it is undesirable that technical considerations define the subject of the investigation. Having a deeper understanding of the impact that data analytical decisions can have on the interpretation of numerical results of a study would help to apply analytical tools that are both suitable and appropriate to answer clinical questions. The aim of this thesis was to investigate the impact of choices regarding the design and statistical analysis of a study on the meaning of its numerical results in two sets of case studies in research into causal effects (Part I) and prediction research (Part II). The thesis concludes with a discussion on the role and importance of clinical research questions and estimands. Clearly defining a clinically relevant estimand ensures that data analytical decisions yield meaningful results. Making targeted research questions central to quantitative clinical research can reduce fallacious confidence in (complex) methods and can add to intelligibility of findings. Show less
