Statistical Techniques to Detect Fraud and other Data Irregularities in Clinical Questionnaire Data

Abstract

The detection of fraud and other systematic data irregularities in clinical trials is an important issue. While awareness of the problem is growing and willingness to combat it is clear, there still appears to be a lack of detection procedures suitable for routine implementation by trial coordinators. The shortage is particularly acute for discrete data, since the majority of methods which are available have been developed for continuous responses. In this paper, we examine the suitability of existing methods for discrete outcomes and propose a new technique for questionnaire data in both an informal graphical mode and as a randomization test. This method exploits the underlying correlation structure of a questionnaire and the difficulty in fabricating such details. A data set concerning a trial of a novel drug for treatment of schizophrenia, in which the Brief Psychiatric Rating Scale was used to assess patient mental health, is used for illustration.