Efficient quality review for modeling input dataset
Vincent Buchheit, Pui Tang, Aurélie Gautier, Thomas Dumortier, Grégory Pinault, Jean-Louis Steimer
Modeling & Simulation, Novartis Pharma AG, Basel, Switzerland
Objectives: The objective of this poster is to share our experiences regarding modeling input dataset validation and to solicit feedback and collaboration from other Pharma industries.
Methods: Drug development is a succession of clinical trials, where statisticians and programmers play a major role in the data reporting and data analysis. In addition, pooling data across studies, within a compound, or across compounds is becoming nowadays a routine activity in most pharmaceutical companies, also in the Modeling and Simulation (M&S) Programming Group at Novartis. While we develop tools and methodologies to facilitate data pooling, we are still facing the following challenge: “How can we perform an efficient quality review of our pooled modeling input dataset, and therefore validate the data?”
Results: For this purpose, the double programming process is routinely applied in pharmaceutical industries. An independent programmer re-produces the same dataset based on data specifications. The validation is completed when both datasets match. It can takes up to several weeks before completion. This Quality Control (QC) method is adequate to ensure that the data-generating program does what it is supposed to do. However, it does not guarantee that the data is scientifically accurate. From a modeling and simulation point of view, it’s important to identify upfront most of the data issues that could impact the model development. The data issues are diverse: missing covariates, discrepancies between units and associated laboratory measurements, inconsistencies between dose history and pharmacokinetic samples, inconsistencies with the clinical studies report, different SAS(R) formats applied for the same variable (race, ethnicity) .... Some key graphics, such as the pharmacokinetic concentration data versus time since previous dose or plots from the scheduled timepoint against the calculated elapse time, or a summary table of all covariates by study and centers are efficient ways to identify potential issues. We will report on other graphical and statistical approaches, including the re-use of a qualified model on comparable data.
Conclusions: The use of Model Based Drug Development is more and more advocated in the pharmaceutical industry. The modeling approach contributes to answer some of the key questions, such as “What is the best dose?”, or “What is the best dose regimen?”. The matter of data quality is essential.