You are here: Home DFG - Research Unit Projects Non-randomized


Evidence from non-randomized and partially randomized studies

Duration: 2004-2010

Principal investigator: Dr. Claudia Schmoor (IMBI)

Researchers: Dr. Claudia Schmoor (IMBI), Dr. Christine Gall, Dr. Erika Graf (IMBI), Dr. Susanne Stampf (IMBI)



It is generally accepted that randomized controlled trials (RCTs) are the gold standard for the comparison of treatments in clinical trials. Randomized treatment assignment shall provide reliable unbiased estimates of causal treatment effects. Nevertheless, there are reasons to perform non-randomized clinical studies in situations where randomization is not possible, feasible, or not enforceable. But observational intervention studies might show association structures instead of causal relationships due to confounding. The crude association of treatment with outcome then fails to admit a causal conclusion.
To cope with this problem in settings with time--constant associations, proposals of two kinds have been introduced: Those that use the relationship between prognostic factors and the outcome variable, such as multiple regression, and those that are based on the relationship between prognostic factors and treatment assignment, such as propensity score or instrumental variable methods. The statistical properties of the propensity score applied to different situations and used in certain forms e.g. for stratification and matching are still not fully explored and further research is needed. In this project, we explore the behaviour of propensity score methods and compare it to conventional regression approaches. Additional emphasis lies on the assessment of their practical applicability.
Another focus of the project is the investigation of causal effects in data sets with a more complex association structure, especially with interdependancies changing over time. Thus, the influence of time-dependent confounders must be correctly modeled which is split in two steps. A structural model is defined to characterize the causal relation between treatment regime and outcome and the confounding mechanism is formulated separately. The common ground of the two focal points is causal modeling by means of counterfactuals.



  • Schmoor C, Gall C, Stampf S, Graf E. Correction of confounding bias in non-randomized studies by appropriate weighting. Biom J 2011; 53(2):369–387.
  • Stampf S, Graf E, Schmoor C, Schumacher M. Estimators and confidence intervals for the marginal odds ratio using logistic regression and propensity score stratification. Statist. Med. 2010; 29(7-8):760–769.
  •  Graf E, Schumacher M. Letter to the editor: Comments on the performance of different propensity score methods for estimating marginal odds ratios. Statist. Med. 2008; 27(19):3915–3917.
  • Senn S, Graf E, Caputo A. Stratification for the propensity score compared with linear regression techniques to assess the effect of treatment or exposure. Statist. Med. 2007; 26(30):5529–5544.
  • Schmoor C, Caputo A, Schumacher M. Evidence from nonrandomized studies: A case study on the estimation of causal effects. Am J Epidemiol 2008; 167(9):1120–1129.
  • Müller C, Caputo A, Schumacher M, Raab G, Schütte M, Hilfrich J, Kaufmann M, von Minckwitz G. Clinical response by palpation during primary systemic therapy with 4 dose-dense cycles doxorubicin and docetaxel in patients with operable breast cancer: further results from a randomised controlled trial. European Journal for Cancer 2007; 43:1654–1661.
  • Hochholzer W, Trenk D, Bestehorn H, Fischer B, Valina C, Ferenc M, Gick M, Caputo A, Büttner H, Neumann F. Impact of the degree of peri-interventional platelet inhibition after loading with clopidogrel on early clinical outcome of elective coronary stent placement. Journal of the American College of Cardiology 2006; 48:1742–1750.