Forensic Pharmacometrics: Part 1 - Data Assembly
Thaddeus H. Grasela* (1), Jill Fiedler-Kelly (1), Darcy J. Hitchcock (1), Elizabeth A. Ludwig (1), Julie A. Passarell (1)
(1) Cognigen Corporation, Buffalo, New York, USA
Introduction: As modeling and simulation results become integral to a program’s outcome, the consequences of lapses in data assembly and analytic result quality can jeopardize the role of pharmacometrics in contributing to the transition to model-based drug development. There are few standards available to define measures of acceptability and suggest strategies for assessing the “fit for purpose” of analysis datasets or model building efforts. These quality assurance activities might include, for example, a review of programming logic and coding, as well as the assumptions used to re-create dosing histories.
Objectives:
Describe a case study of a forensic assessment of analysis-ready datasets performed as part of a due diligence effort.
Describe methods used in the forensic assessment that identified problems and errors in the previously constructed datasets and propose proactive quality assurance activities.
Methods: A series of quality assurance checks comparing the analysis-ready datasets to the source data files were developed by both data programmers and scientists addressing their individual areas of expertise. Three teams, operating in parallel and consisting of a scientist and a data programmer, were constituted to focus on different aspects of the PK and PK/PD datasets and modeling. A review of the previously prepared technical reports was used to identify the assumptions and strategies that went into the original data assembly and model-building efforts.
Results: The forensic analysis of the datasets revealed a mismatch in demographic data with corresponding dosing and PK data in a large percentage of patients and systematic errors in the creation of dosing histories, including improper use of NONMEM®-derived data items (ADDL) and incorrect dose amounts in subsets of patients across studies. The descriptions of patient disposition and data deletions were insufficient in supplying reasons or rationale for the programming logic errors discovered.
Conclusions: A gap currently exists in defining the criteria for judging the quality of data assembly efforts along with the comprehensiveness of data programming, technical report, and other supportive work product documentation. Strategies for this assessment can be used as a basis for independent validation of pharmacometric work products prior to use in critical decision-making activities, as well as in the development of standards for quality assurance activities.