Influences on inferences. Effect of errors in data on statistical evaluation

Background. Inadvertent random and systemic errors introduced into data sets and manipulation of data are well‐defined sources of discrepancies in statistical evaluation of clinical trials. In this study, the authors show the influence of errors on the widely used statistical result, P values. Methods. Using data from a retrospective study of patients with Hodgkin disease treated at the University of Minnesota between 1970 and 1984 and observed to 1988, we introduced various errors into the data to study the impact on results. Results. Inadvertent random and systemic errors affect statistical results. Data entry and transcription errors, vague definitions of endpoints and prognostic factors, and the omission and selection of patients are examples of frequent errors that affect statistical evaluation. Conclusion. The results and inferences of many studies are sensitive to systemic errors and data manipulation. Great care must be given to the clear definitions of terms, exclusion and inclusion criteria, group assignments, treatment protocols, and the subgroups on which statistical analysis is performed. Clinicians and statisticians must work together to improve the performance and interpretation of clinical trials.