All too often we in pharmacovigilance consider the generation of data to be our primary purpose. However, data is of little use if we cannot use it to generate meaningful information. In pharmacovigilance it’s possible to have very large data sets, especially for modern pharmaceutical products: if we consider Atorvastatin, the most widely used statin globally, in 2019 over 112 million prescriptions were issued in the United States alone with the next four most popular statins having 94 million prescriptions. This equates to something in the order of 40 million people receiving this class of product in the USA. It is clear we do not have the capacity to monitor individual patients to see if there are any events that might constitute an adverse reaction to therapy, so we must look for alternative ways to monitor the real-world safety of drugs.

One way we do this is to look at the data we collect to see if there is any obvious trend suggesting an adverse effect. This trend need not be subtle to escape detection: if we look at thalidomide for example, despite the catastrophic teratogenic effects being obvious, the link was not made with the drug for some time after marketing, notwithstanding the reported incidence of abnormalities in the infant’s limbs being around 20%. This particular adverse reaction was ultimately to lead to the Medicines Act in the UK, and hastened the advent of modern pharmacovigilance.

Clearly, we need a more sensitive tool to ensure that even rare adverse events are detected quickly to limit any harm to the treated population and to enable remedial action to be taken. Step forward signal detection! The CIOMS working group 8, defines a signal in drug surveillance as “information that arises from one or multiple sources (including observations and experiments), which suggests a new potentially causal association, or a new aspect of a known association, between an intervention and an event or set of related events, either adverse or beneficial, that is judged to be of sufficient likelihood to justify verificatory action”.

Signal detection is by no means limited to the world of pharmacovigilance, but with the massive increase in Ecommerce, it is a major tool for many sectors. If we consider financial transactions, there are over 1 billion credit card transactions globally each day. The banks have extremely efficient AI algorithms to detect potentially suspicious use and will act to suspend an account. Similarly, in online betting where potential scams are increasingly common, signal detection software scans for abnormal betting patterns across a variety of sports. Given that the sports betting market is worth some $66 billion annually it is clear that the computing systems needed to monitor such use are enormously complex and expensive.

In this regard, drug safety operates with much smaller amounts of data. Because in most countries at present, we don’t have systems that will track a patient from prescription to an event automatically, we have to rely on what is reported to us by the prescriber, other health care professionals and, of course, the patients themselves. We know that the majority of unwanted effects of treatment are under reported, so it is critical that our data storage systems are capable of accurately holding those events we have recorded, in a way that allows us to generate signal detection statistics.