Data Mining and Signal Detection in Pharmacovigilance

A signal is described by the World Health Organization as any information that is reported on a possible or potential causal relationship between a drug and the adverse event it spawns. This relationship can be of virtually any nature, so long as it concerns the drug and the subject, and it could be either new or one with a precedent.
Data mining can be described as the method of obtaining data from target groups to help the clinical study come to important assessments and conclusions. Many a time, it is not clear whether a drug’s expected benefits outweighs the potential risks it brings about or vice versa, till the drug goes for marketing authorization. In order to assess this to the extent possible, clinical pharmacologists weigh the benefit and risk evaluation of medicines using tools such as data mining. Data mining is done both at the individual level of a subject and at the macro level of the population at large. These two methods are usually inseparable from each other, in that almost no study is done exclusively for one group.
Given its ability to help pharmacologists discern the various patterns that emerge from a clinical study; data mining is acquiring a position of importance of late and is being used in almost all stages of drug development. This could range from the earliest stage, namely drug discovery and could go up to post-marketing surveillance.
The WHO’s Uppsala Monitoring Centre

In order to make the results of very clinical study done in every part of the world accessible to everyone – a formidable task without doubt – the WHO has formulated the Uppsala Monitoring Center. This is a universal database of all the results obtained from clinical research the world over. Although voluntary and missing data from a many studies; the UMC is a comprehensive attempt at establishing a data mining and signal detection system that is accessible to everyone concerned. The UMC can thus be considered the universal data mining and signal detection database.

With over 2.5 million case reports of various clinical studies done all around the world, the UMC has evolved over time as a data mining and signal detection database. It initially started by requiring principals of clinical studies to generate new drug and Adverse Drug Reactions (ADR) combinations every three months. With the growth in the number of studies and the variety of issues they threw up; this was no longer considered feasible. The UMC then started out to create its own method, by which the principles of making an objective initial assessment of all new drug and ADR combinations started getting implemented as they emerged. To this were added the requirements of bringing about a transparent selection of drug – ADR combinations for review, as well as suggest a quantitative aid to data mining and signal detection.

Today, the UMC uses several methodologies to carry out its task of being at the forefront of data mining and signal detection. It uses the Bayesian Confidence Propagation Neural Network, which uses Bayesian statistics within the architecture of a neural network for data mining and signal detection.

click to continue reading