Data Mining Clinical Trial Data: Dredging or Innovative Exploration?

• Activity Number: 149
• Type: Luncheons
• Date/Time: Monday, August 8, 2005 : 12:30 PM to 1:50 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Mani Lakshminarayanan*+ and Jose Alvir
• Companies: Pfizer, Inc. and Pfizer, Inc.
• Keywords: Knolwedge Discovery ; Statistical Learning ; Large database ; Text Mining ; Classification ; Graphical Tools

Abstract

Data mining, another term for exploratory data analysis, is a methodology that makes use of computationally feasible statistical techniques in finding interesting structures and relationships, unknown a priori to the researcher, within a large database. Clinical trial databases are typically large, with data from many trials involving several thousands of patients. For the primary goal, namely getting a drug approved before marketing, pharmaceutical sponsors are mandated to prespecify the hypothesis and analyses for all confirmatory trials. But there are always additional interests in generating hypotheses in support of postmarketing needs and designing new trials, not usually done through standard reporting of clinical trials. For example, in phase II/III studies, one might be interested in exploring or confirming the therapeutic efficacy of a new drug. For this reason, it is important for statisticians to explore tools and methodologies needed to explore the vast database, which are not covered in a clinical study report. This group also will discuss whether a process such as this should be considered as dredging or knowledge discovery.