- Jeffrey K Aronson
Centre for Evidence Based Medicine, Nuffield Department of Primary Care Health Sciences, University of Oxford, Oxford, UK
- Twitter @JKAronson
Etymologically, a signal is a mark that in some sense shows the way. By extension it demands to be followed. Indeed, unless it is followed it is of little value, and failure to follow a signal may lead to difficulty, or even disaster. One thinks, for example, of failure to follow an instruction given by a set of traffic signals. A published dosage regimen is a signal demonstrating how a drug should be used, and a pharmacovigilance signal warns about a potential adverse drug effect or reaction, of which one should take notice and follow up with more detailed studies.
The word signal may have derived from one of two similar IndoEuropean roots, SEK, to cut or SEKW, to follow. Both give us many English derivatives.
SEK, via the Latin verb secare, to cut, gives us secant, section and venesection, sector, and segment; adding prefixes we get dissect, insect, intersect, resect, and transect.
SEKW, via the Latin verb sequor, to follow, gives us sequel, sequential, consequent, subsequent, and obsequious; segue and seguidilla; sue, ensue, and pursue; suit, suite, and suitor; execute, persecute, and prosecute.
The word “sect” is ambiguous. In a now obsolete sense it meant a cutting from a plant, from SEK. However, in its most modern sense it means an organised group of people, often, although not always, united by a religious belief. It may come from SEKW; the members of a sect follow their leader and form a society. Or it may come from SEK, as a shortened version of the Latin phrase via secta, implying a group that cuts its own path through life and cuts itself off from others.
The Latin noun signum means a mark, perhaps because it is cut into a surface, such as stone, as a signet ring cuts into wax, or perhaps because it is a sign of something that has to be followed. Other meanings that imply following include a flag (something that a soldier follows) and a gesture. From signum we get seal and sigil and of course sign and its derivatives, assign, consign, design and designate, resign, and insignia. Signal comes to us from the Late Latin quasi-adjectival noun derivatives of signum, signallis, a signpost, and signalle, a signboard.
When elsewhere discussing the definitions of pharmacovigilance1 and surveillance,2 I mentioned the term “signal” several times, referring to its specialised sense in pharmacovigilance. And, like an electrical signal for example, a pharmacovigilance signal conveys specific information about its source and, like a traffic signal, spurs one to action or even demands it. The information it conveys is about a suspected adverse drug reaction and the action it demands is verification or refutation.
Adverse effects of drugs and reactions to them are not well studied during the early phases of drug development. Early studies are typically too small to detect any but the most common adverse outcomes, and randomised trials are typically designed to detect benefits rather than harms. The phrase “well tolerated” is all too often used when describing drugs in early development and even sometimes when adverse reactions are observed in clinical trials.
In later stages of development, adverse events tend to be poorly documented in clinical trials. Even if they have been collected, which is often not the case, they may not be reported among the published results, and even if they are reported they may be reported incompletely. Consequently, even though systematic reviews of adverse events in randomised clinical trials are possible, caveats may have to be issued. For example, in a systematic review of 183 studies of the use of macrolide antibiotics involving 252 886 participants, the authors reported that “few trials clearly listed adverse events as outcomes, reported on the methods used for eliciting adverse events, or even detailed the numbers of people who experienced adverse events in both the intervention and placebo group.”3
It is not therefore surprising that a lot of information about adverse events in people taking medications comes from anecdotal reports.
In a few instances, one or no more than a handful of case reports is sufficient to incriminate a drug as having caused an adverse reaction. For example, thrombophlebitis near the site of insertion of a catheter into a small vein soon after infusing a drug is highly likely to have been due to the drug; amiodarone provides an example of this.4 In 2006 Manfred Hauben and I described four types of such reactions and called them “between-the-eyes” reactions, because when they occur the causative association with the drug is obvious—it causes interocular trauma (hits you between the eyes).5 Here are the four categories, with examples and comparisons based on crime analogies:
● extracellular or intracellular tissue deposition of the drug or a metabolite; in this case the lesion has to be accessible for examination, for example by biopsy; renal stones composed predominantly or exclusively of a drug would qualify as a between-the-eyes adverse reaction to that drug6; this is analogous to catching the culprit at the scene of the crime;
● a specific anatomical location or pattern of injury; examples include damage due to extravasation of a drug at an intravenous administration site7 and ulceration due to topical aspirin8; this is analogous to witnessing the culprit committing the crime;
● physiological dysfunction or direct tissue damage that can be proved by physicochemical testing; photopatch testing for photoallergic reactions provides an example, if accompanied by proper control procedures to avoid the problems of false positive and false negative results9; this is analogous to recreating the crime scene in the presence of the suspect;
● infection as a result of administration of a potentially infective agent or because of demonstrable contamination; granulomatous hepatitis after BCG instillation into the bladder for transitional cell carcinoma, where the causative organism is identifiably the same, is an example10; this is analogous to identifying the culprit from fingerprints left at the scene of the crime.
In most cases single anecdotes will not suffice to establish causality. However, large numbers of anecdotal reports can provide a signal that something is going on. Large databases, such as the World Health Organization’s Vigibase, which gathers anecdotal reports from pharmacovigilance centres around the world, can detect such signals.11 The data sources that can contribute to detection of signals of suspected causality include individualities (i.e. single anecdotes or case series), observational studies (such as case-control studies), and interventional studies (randomised trials, singly or in systematic reviews and meta-analyses). Information from non-clinical sources, such as in vitro and animal studies, can also contribute.
The essential features of a pharmacovigilance signal are that:
● it is based on reports of an association between an intervention or interventions and an event or set of related events (e.g. a syndrome), including any type of evidence (clinical or experimental);
● each signal represents an association that is new and important and has not been previously investigated and refuted;
● a signal encourages further investigation, whether to verify or refute it from better evidence or to seek ways of preventing or minimising the adverse effect;
● intervention-event associations that are not related to causality or risk with a specified degree of likelihood and scientific plausibility should not be considered to be signals.
When enough reports of an association between an intervention and an event accumulate, statistical tests, such as proportional reporting ratios,12 gamma Poisson shrinkers,13 and Bayesian neural networks,14 provide evidence that the association is statistically significant. That is when the association can be declared to be a signal. This leads to the concept of a signal of disproportionate reporting, which refers to the numerical outputs of disproportionality analysis, when the frequency with which an adverse event is associated with a particular intervention significantly exceeds the frequency with which it is associated with all other interventions combined.15
Signals do not prove causality, but they do provide evidence of a suspected causal association sufficient for hypothesis formulation and further studies.
Defining a signal and its subtypes
Based on these features, Manfred Hauben and I proposed the following definition of a signal of suspected causality16:
signal of suspected causalityn. 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, [which would command regulatory, societal, or clinical attention, and] is judged to be of sufficient likelihood to justify verificatory [and, when necessary, remedial] actions
This definition was republished by the Council for International Organizations of Medical Sciences (CIOMS) in 2010, with the word “that” replacing the first bracketed section and omission of the second bracketed section.17 It was later adopted by a CIOMS Working Group, with a further amendment, omitting the section after “beneficial.”18
We also proposed the following definitions for subtypes of signal:
verified signaln. a signal of suspected causality that has been verified either by its nature or source (e.g. a definitive anecdote or a convincing association that has arisen directly from an RCT) or by formal verification studies
refuted signaln. a signal of suspected causality that has been subjected to attempted verification and has been refuted
indeterminate signaln. a signal of suspected causality that has been subjected to attempted verification and has been neither verified nor refuted
The first of these was also republished in CIOMS VIII in 2010,17 which we take to be a signal of approval.