“Causal or casual”? The challenge of pharmacovigilance
Imagine that, after starting a new therapy, a patient laments a symptom she usually does not experience. Since she feels confident that the new drug and the symptom are causally related, she reports the episode to her national drug agency. How should the drug agency interpret such report? Did the drug actually have a causal role in the onset of the undesired effect, as the patient suspects, or was the symptom arising from other, concomitant changes in the patient’s life? Or again, was it due to a development of the patient’s initial condition, which the drug was supposed to treat?
The practice of pharmacovigilance often requires to be able to assess causality from one single case, or from a series of few similar cases. This is an unsolved challenge not only practically and methodologically, but also conceptually. We are used to think about causal relationships as something that tend to repeat themselves, and to generate regularity. Instinctively, in order to check how plausible it is that a certain effect was caused by a certain drug, we look for how many times the same drug previously provoked the same effect in other patients. But what when we do not have such information? What if that single case, or few cases, represent the first and only evidence available? This is usually the challenge that pharmacovigilance needs to face.
Causality assessment approaches: some differences and some similarities
In the decades, several tools were developed in order to assist with the task of assessing how probable causality is – just from the analysis of one single case. Some of these methods are based on an algorithm, and they go through a series of questions while they assign a certain score depending on the answers. The plausibility of causation is given by a total, final score. One commonly used type of such algorithms is the Naranjo method, for instance. Other approaches to causality assessment, such as the WHO-UMC approach, are skeptical of the use of scores and are based on the open argumentation of experts. Again, other methods are more statistical, and especially based on Bayesian probability.
Although these approaches are different, they have some common traits. For instance, they all require investigation of other possible causes of the adverse effect, other than the suspected drug, and they value evidence of dose-response of the effect (which is when the intensity of the symptom changes depending on the variation of the drug’s dose). In CauseHealth Risk and Safety, we have revised such common traits and we are going to re-organise it a dispositionalist causality assessment. We aim to a dispositionalist tool which could guide the evaluation of evidence of causation in the single case, or the evaluation of a case series.
A singularist theory of causation for the causal assessment in single (or few) cases
We have several motivations for starting this effort. First, we think that we have a conceptual advantage for the purpose of causality assessment in the single case, since we start from a singularist theory of causation: casual dispositionalism. Since causality assessment in pharmacovigilance happens case-by-case, an understanding based on frequency of effect, regularity or difference making at population level will have limited use in this case. From the dispositional perspective, causation happens because of intrinsic properties (dispositions) at play in every single instance. From a dispositionalist understanding of causation, it is at least in theory possible to understand something about the causal relationships at play by a thorough analysis of one single instance of causation. For this reason, we suggest that a dispositional perspective might be a better starting point for argumentation that there is (or that there is not) evidence for a signal of harm from a drug. But if we accept to adopt a dispositionalist view of causation, how would the ideal causality assessment tool in pharmacovigilance look like? This is what we are working on answering in CauseHealth Risk and Safety.
Patient’s pre-disposition: the cause or only a trigger?
There are other practical motivations for our purpose of a dispositionalist tool for causality assessment. When a signal of harm from a certain drug is proposed, the validity of the evidence is sometimes objected by authorities or by other experts. The reason for these objections is often a known pre-disposition of the patient to the symptom (for instance because of background conditions) or a known disposition of the patient’s context to the symptom (because of confounding of other drugs with the same or similar ADR). However, a patients’ predisposition should not automatically exclude the possibility that a drug played a causal role for triggering the effect. If there is enough evidence for a drug’s intrinsic property to dispose toward an effect, for instance, this should be in some cases sufficient to recommend action. Action could be intended as further investigation, or communication of the risk or caution and monitoring.
A transparent causality assessment should describe, argue and motivate
A third concern we want to address is that the detection of a signal depends on the causality assessment of the single cases, or of a case series. However, this causality assessment should be as transparent as possible, especially concerning assumptions and interpretations of the causality assessor. Implicit subjective interpretations of ‘possible’, ‘plausible’ and other concept, for instance, might be inherited un-scrutinised. This can be partly avoided if the causality assessor does not express an overall evaluation about the plausibility of causal relationship. Instead, the assessor could simply describe the available evidence for the dispositions at place (dispositions of the drug, of the patient, and of the interactions drug-patient). When the evidence is from the literature, this should be quoted. The assessor should also motivate her reasoning about the evidence and its strength/ significance, with the help of the provided tool.
This last suggestion echoes the advances made in the last decade by feminist epistemologies. These insights were powerful in transforming the conviction that a unanimity of scientific evaluation is what we should all wish, when we need to make evidence-based decisions. On the contrary, feminist epistemologists urge that a variety of judgements are going to generate a more informed decision, as long as all judgements are open about their premises and their basic assumptions.
This is what we aim in our forthcoming proposition. Starting from the basic assumption of casual dispositionalism, we wish to provide pharmacovigilance experts with a tool to argue their causality assessment in the most open and informative way for decision makers.
CauseHealth Risk and Safety 