In the context of dichotomous outcomes, healthcare interventions are intended either to reduce the risk of occurrence of an adverse outcome or increase the chance of a good outcome. All of the effect measures described in Section 9.2.2 apply equally to both scenarios.
In many situations it is natural to talk about one of the outcome states as being an event. For example, when participants have particular symptoms at the start of the study the event of interest is usually recovery or cure. If participants are well or alternatively at risk of some adverse outcome at the beginning of the study, then the event is the onset of disease or occurrence of the adverse outcome. Because the focus is usually on the experimental intervention group, a study in which the experimental intervention reduces the occurrence of an adverse outcome will have an odds ratio and risk ratio less than 1, and a negative risk difference. A study in which the experimental intervention increases the occurrence of a good outcome will have an odds ratio and risk ratio greater than 1, and a positive risk difference (see Box 9.2.a).
However, it is possible to switch events and non-events and consider instead the proportion of patients not recovering or not experiencing the event. For meta-analyses using risk differences or odds ratios the impact of this switch is of no great consequence: the switch simply changes the sign of a risk difference, whilst for odds ratios the new odds ratio is the reciprocal (1/x) of the original odds ratio.
By contrast, switching the outcome can make a substantial difference for risk ratios, affecting the effect estimate, its significance, and the consistency of intervention effects across studies. This is because the precision of a risk ratio estimate differs markedly between situations where risks are low and situations where risks are high. In a meta-analysis the effect of this reversal cannot easily be predicted. The identification, before data analysis, of which risk ratio is more likely to be the most relevant summary statistic is therefore important and discussed further in Section 9.4.4.4.