Table 2 Bradford Hill’s causality considerations as applied to the association between humanitarian assistance and excess mortality
Consideration | Application to the HA-mortality association | Relevance (this author’s suggestion) |
|---|---|---|
Strength (effect size) | A large observed effect size indicates that it is relatively more likely that HA is indeed causing a decrease in mortality | − (The effect of HA will depend on the appropriateness and performance of HA, i.e. the intensity of exposure: a small effect size does not particularly suggest lack of causation, as it may simply reflect insufficient HA.) |
Consistency (reproducibility) | A comparable effect of HA is observed in multiple crises of comparable typology that benefit from similar humanitarian responses | + (In practice, variability in crisis and response characteristics, as well as evaluation data and methods, may preclude reproducibility of effects observed in one crisis across other crisis scenarios.) |
Specificity | Both HA and mortality are measured and quantified in an unequivocal way, and not in generic terms | ++ (Mainly relevant for the exposure: as discussed in the text, the metric by which HA is quantified should reflect specific parameters, e.g. appropriateness, coverage etc. that determine its theoretical effect on mortality.) |
Temporality | The decrease in mortality occurs after the increase in HA | ++ |
Biological gradient | A (negative) ‘dose–response’ relationship is observed between HA and mortality: the more appropriate and performant HA, the lower mortality | ++ |
Plausibility | The observed effect is consistent with the body of knowledge on what is known about the mechanism whereby specific humanitarian services (e.g. measles vaccination) affect mortality, and more generally about causal factors for mortality. The observed effect is also consistent with what is known about the specific context of the crisis in question, how HA was implemented and what other factors may have affected mortality | ++ |
Coherence | Published evidence, e.g. from trials, shows that the humanitarian services offered as part of the response are effective against mortality | + (More applicable to specific and new health problems of unknown cause. Somewhat overlapping with plausibility.) |
Experiment | If HA is withdrawn the observed effect disappears. Alternatively, HA is offered only to a randomly selected group, and an effect is observed among those receiving HA but not among those not receiving it | − (Unethical, unless a new, specific intervention is proposed, the effectiveness of which is insufficiently established.) |
Analogy | A similar causal association has been observed in other (non-crisis) settings where vulnerable people have received a given package of services. This could, for example, be stable but low-income countries where a basic package of services is introduced at scale | + (It might be challenging to identify analogous associations outside of humanitarian responses. This consideration would probably not add much beyond the list above.) |