Survival or time-to-event analysis: common pitfalls of retrospective approach

SUMMARY Retrospective analysis of the archival clinical data collected outside the controlled trials is the main source of false conclusions due to data manipulation. This flawed method produces common errors especially in case if life duration is used as an endpoint in survival analysis. It is pointed out that the mixing of searching and / or descriptive-promoting goals with the evidence aimed conclusions is incorrect and should be avoided. The main feature of evidence based study is the fact that the design of the study, inclusion / - exclusion criteria, recruitment time and analysis plan are developed and exist before the experimental data will appear. Special modeling program was written to generate the data and to simulate wrong procedures of data manipulation and analysis. In the paper the examples of late post factum selection of patients are demonstrated and the resulting bias of estimates is discussed. The main conclusion is that retrospective time to event analysis of the passively collected data is not free of mistakes. Such type of the study and analysis should be used with limitation in discovering and descriptive studies.