The assessment of adverse events is one of the key challenges in pharmacovigilance. Bayer now realized an innovative approach to increase the efficiency in this process: The software Cognium® determines the key properties “listedness“ and “company causality“ for reported product-event combinations. In a proof-of-concept it could be shown that the self-learning system yielded an initial assessment quality of about 90 % compared with human experts. Now, the solution was implemented by Cognium® creator Lenolution and system integration partner Syncwork.
The processing of adverse events is one of the main challenges for drug producers in pharmacovigilance. Increasing case numbers, ever-changing regulatory reporting requirements, as well as new case sources, such as social media, consume a growing amount of resources in pharmacovigilance departments.
A new approach to make case processing more efficient was now taken at Bayer AG by employing the innovative software Cognium®. The system calculates an essential information in the analysis of adverse event reports. Its core features compute assessments for two important aspects of each product-event combination:
The causality assessments are based on Bradford Hill criteria (strength, consistency, specificity, temporality, biological gradient, plausibility, coherence, experiment, analogy). As result, Cognium® provides its assessment including probability values, such as “with 80 % certainty the adverse event is caused by the product“. Since the system adjusts the algorithm parameter settings based on experience, it improves the results with growing historic data (self-learning component).
Listedness: Is this combination already listed, i.e., known to the producer?
Causality: Can a relation between the product and the event be derived from the available reported data?
Cognium® was developed by the Berlin-based Lenolution GmbH and implemented at Bayer with system integration partner Syncwork AG. After a test period of several months Magnus Lerch, CEO of Lenolution, stated: “We could even improve on the quality that was already reached in the proof-of- concept. The assessments are consistent with those of human experts in about 90 %. We expect that this percentage will even get better with the time when the software is fully exploiting its learning features.“
Bayer project manager Klaus Ristau: “The additional calculation of this information is of particular interest to us, since it will focus our quality measures where the need is most important. Now, in a first phase we will test the solution in a productive setting by integrating Cognium® in our quality management process. However, we see a tremendous potential for improving the efficiency in our case assessment process.“