Causality assessment is important for the benefit/risk evaluation of a drug, especially when post-marketing safety signals emerge. Its purpose simply is to answer the question “causal or coincidental?” In some cases, the assessment is rapid because there is a self-evident causal relationship (i.e. a case of anaphylactic shock reaction during or just after the injection of a vaccine or intake of other drugs); in the vast majority of cases, answering this question is much more difficult and requires step-by-step approaches with specific methods.

For instance, on 10 March 2021 the EMA probably used the observed vs expected approach when it stated that “the number of thromboembolic events reported in vaccinated people seems not to be higher than seen in the general population”.

The benefit/risk balance is a very basic concept, but its statistical computation is in fact fairly complex for many reasons. For instance, the therapeutic effect (benefit) is a pharmacological response expressed by almost all treated persons; in the case of COVID vaccines, we observed 65 to 97% positive responses, i.e. efficacy, in the clinical trials. Untoward effects are quite different as they occur only in certain persons when certain risk factors, often unknown, are present; moreover, we have to bear in mind that a significant part of untoward effects of medicines present on the market are not yet identified. These adverse effects could have a low to extremely low probability of occurrence so computations and modelling must be based on adequate probability distributions.

In the 18th century, Jean le Rond d’Alembert, using the example of the variolation against smallpox extensively discussed the contribution and limits of statistics to determine the benefit/risk balance. Among other pioneering concepts that seem today forgotten, d’Alembert said that:

1. The benefit/risk balance generally differs for the person and the population;
2. Vaccination imposes a risk of adverse drug reaction (ADR) for all while, for a given person, the gain is hypothetical (random), because this person could have or could not have  the chance to be in contact with the virus, i.e. to be infected, and to develop or not severe form or death PBad = PInf x PHosp or Death during the relevant time window, e.g. life span;
3. The risk is generally limited to some days or weeks after vaccination while the benefit is incremental, i.e. increases over time (assuming « life time » protection);
4. It is generally misleading to present a benefit/risk balance as an average for the whole population. This could lead to inappropriate public health decisions, because both risk and benefit generally vary greatly with age, often in opposite directions; consequently, the benefit/risk balance depends on the age when the vaccination is performed, so life expectancy plays a major role in the calculation.

Regarding the signal about cerebral thrombosis, data available tend to support a causal role of Vaxzevria (AstraZeneca vaccine) in the occurrence of CVSTs. Unfortunately, many ingredients are missing for the computation, for instance the duration of vaccine protection or the duration of the COVID-19 pandemic, which could disappear now or be still present in 10 years. Further uncertainties relate with the effectiveness of the AZ vaccine against (present and future) variants and the availability of « safer » alternatives (vaccines or not) in the next future.

If we assume that this vaccine is (and will remain) effective against variants (present or future) without considering other associated (known or unknown) ADRs, computation shows that the benefit/risk balance of Vaxzevria  remains a priori very good, but according to statistics, a clear concern exists for young adults and, probably particularly women. In older adults (>64), the benefit/risk balance remains excellent.

An important point to retain is that a good communication is more important than statistics: some fatal cases, e.g. in healthy young women, can kill the product, even the campaign, more surely than a brilliant computation.

To conclude, statistics and modelling are a powerful tool for preparing and adjusting public health decisions and the results can be easily strengthened by sensitivity analyses. In the present case, a specific and independent pharmacoepidemiology study should be carried out as soon as possible in order to better appraise the possible link between vaccination and the occurrence of CVSTs and the magnitude of the relative risk in order to be more precise for regulatory decisions and for the choice of target populations.

• Bégaud B, Jones JK. Assessing Causality from case-reports. Chapter 14th In Brian Strom’s Textbok of Pharmacoepidemiology. 3d ed.Wiley 2021.
• Semenzato L, Botton J, Drouin J et al. Chronic diseases, health conditions and risk of COVID-19-related hospitalization and in-hospital mortality during the first wave of the epidemic in France: a cohort study of 66 million people. The Lancet Regional Health - Europe, Volume 8, 2021
• EMA COVID-19 vaccines