Review authors may present absolute effects in several ways, examples of which include risk differences, numbers needed to treat/harm, or attributable risk. Of these methods, evidence suggests that risk differences are most straightforward for evidence users to interpret (20). Risk differences represent the difference in risk between two groups (usually the difference between a group with or without an exposure or groups with higher vs lower exposure). Review authors may calculate the absolute risk difference using the formula below with an assumed risk (AR) (21). The assumed risk represents the assumed baseline risk of the outcome in a group with a defined level of exposure. This assumed risk may be retrieved from the eligible studies in a review or from a representative observational study. When review authors anticipate variation or uncertainty about the assumed risk, review authors may present risk differences corresponding to a range of plausible assumed baseline risks. For example, review authors may choose to present risk differences for a group with low risk, intermediate risk, and high risk of the outcome of interest. \(Risk difference=AR-AR\times RR\) Authors may also calculate risk differences using ORs or HRs using the formulae below (21, 22). \(Risk difference=AR-\frac{AR\times OR}{1-AR+(OR\times AR) }\) \(Risk difference=AR- 1-\text{e}\text{x}\text{p}(\text{ln}\left(1-AR\right)\times HR\) To calculate confidence intervals around the risk difference, review authors may apply the formula for risk difference to the lower and upper bounds of the confidence intervals of the relative effect (21). This method, however, does not account for the variance of the assumed risk. Review authors may use more complex methods like Propagating Imprecision (PropImp), MOVER, or MOVER-R to account for both the variance of the relative effect and the variance of the assumed risk (23–25). To improve interpretability, we recommend review authors to translate risk difference in percentage to the risk difference per 100 or 1,000 people (20). \(Risk difference per \text{1,000} people=RD \times \text{1,000}\) The clinical importance of a risk difference may depend on the underlying risk of events in the population. For example, a risk difference of 0.02 (or 2%) may represent a trivial change if the risk increases from 58–60% or a more important change if the risk increases 0–2%. Hence, we encourage review authors to present the assumed baseline risk, the corresponding risk at different levels of exposure, and the risk difference. The number needed to treat (NNT) is a common alternative way of presenting absolute risk—most commonly for studies of interventions. The NNT is defined as the expected number of people who need to receive the experimental intervention rather than the comparator intervention for one additional person to either incur or avoid an event (depending on the direction of the result) in a given time frame (26). The NNT, however, has some important limitations (27, 28). NNTs, for example, are often presented without confidence intervals, their confidence intervals are difficult to interpret when results are not statistically significant, and they may be misinterpreted if not reported without an accompanying baseline risk and duration. For these reasons, we encourage review authors to present risk differences. Reviews of nutritional and environmental exposures often inform public health decisions and policies for which the effects of exposures over large populations are important. To improve interpretability of reviews for public health, authors may present absolute effects for a population of interest. Such an approach is, however, more complicated because it needs to account for variations in risk and exposure across the population of interest and may necessitate more complex statistical modelling. For example, Gabet and colleagues (2021) present a systematic review and meta-analysis addressing the association between nitric oxide and breast cancer cases (29). They used results of the meta-analysis to estimate the proportion of breast cancer cases in France that could be attributed to nitric oxide using data on atmospheric concentration of nitric oxide in France, population density by age and gender. The review estimates that a reduction in concentration of nitric oxide to the lowest potential level would prevent 1,667 (95% CI 374 to 2,914) new cases of breast cancer in France per year. Authors of reviews that address surrogate outcomes (i.e., an outcome that is only important due to its correlation with another outcome), such as LDL cholesterol or blood pressure, instead of outcomes that are of direct importance to patients and the public, such as cardiovascular events or cardiovascular mortality, will need to translate the change in the surrogate outcome to a patient important outcome to optimize interpretability (30). We refer review authors to other sources that describe methods to translate effects of surrogate outcomes to outcomes that are of direct importance (30). Finally, we caution evidence users to avoid calculating absolute effects by summing the number of participants analyzed and number of events across studies. The most credible effect estimates come from meta-analyses of estimates adjusted for potential confounding factors and absolute effects calculated based on raw event rates will not account for confounders. In fact, adjusted and unadjusted estimates may indicate opposite directions of effect. |