Detecting data fabrication in clinical trials from cluster analysis perspective

Detecting data fabrication is of great importance in clinical trials. As the role of statisticians in detecting abnormal data patterns has grown, a large number of statistical procedures have been developed, most of which are based on descriptive statistics. Based upon the fact that substantial data fabrication cases have certain clustering structures, this paper discusses the potential for the use of statistical clustering method in fraud detection. Three clustering patterns, angular, neighborhood and repeated measurements clustering, are identified and explored. Correspondingly, simple and efficient test statistics are proposed and randomization tests are carried out. The proposed methods are applied to a 12-week multi-center study for illustration. Extensive simulations are conducted to validate the effectiveness of the procedures.