W 05: Unusual Data Pattern Analysis in a Large Pharmaceutical Company
Julie Solbjerg Appel
Clinical Data Surveillance Specialist
Novo Nordisk A/S Denmark
We aim to summarize the lessons learned from the practical implementation of unusual data pattern detection in clinical trials, the process of analysis and the communication of observations to different stakeholders.
ORAL PRESENTATION SCHEDULED: Session 2A at 9:50 - 10:00 AM
SAS JMP Clinical was used to search for unusual data patterns in clinical trial data in Novo Nordisk. Standards for further data drilling and visual outputs were created and the anchoring in the organisation focused on building collaboration with different professional disciplines.
Co-Author: Mette Krogh Beyer
A unit with two full-time analysts with a background in trial conduct was established. During 2015 the analysts analysed 18 trials with the aim of identifying unusual data patterns that could jeopardize the integrity of trial data. A standard set of analyses and visual statistical outputs were applied across trials independent of trial teams and trial design. Subsequently, the analyst supported the trial teams in interpreting the outputs to avoid interpretation errors due to noise or false-positive findings. The conclusions from the unusual data pattern analyses led to decisions and actions handled through existing procedures. To facilitate interpretation and decision making an algorithm was created accounting for the timing of unusual data pattern analyses e.g. during or after trial conduct. The timing of analysis potentially affected decisions and who to involve in decision making. The algorithm included an overview of possible actions to guide the trial teams. Initially, the trial teams were themselves accountable for creating appropriate documentation of any observations.
As experience was gathered, the analyses became better adapted to the specific trial designs and their potential risks. Thus, the unusual data pattern analysis now assumes a more appropriate combination of statistical outputs and graphics tailored to trial risk and therapeutic area. Furthermore, a process for how to document the analysis performed, observations found and actions taken is being implemented across trial teams to facilitate a uniform documentation practice.
Due to data complexity, multivariate dimensionality and noise the standard JMP Clinical output are not immediately understood by the diverse professional disciplines involved in trial conduct. The outputs from the analysis and further data drilling exercises therefore needs interpretation and simplification before being presented to trial teams. A framework of standard analyses and subsequent standard output is necessary. Additionally, the analyst should serve in a guiding role in the discussion of the significance and relevance of the statistical findings to avoid biases especially in terms of false-positives or non-important observations.
In conclusion, a simple and transparent process is essential. This should build on an interdisciplinary model where the standard statistical output and in-depth data drilling is supplemented with an evaluation process involving diverse disciplines within trial conduct. With this setup, the analyst assumes leadership of the process and sets standards to establish a coherent collaborative approach and to translate findings into decisions and actions manageable by the trial teams.
Additional Authors: Mette Krogh Beyer, Thue Johansen, Sinna Lisa Vange