Risk Ratio vs Odds Ratio: Optimal presentation in GRADE evidence profiles

Session Type
Communicating evidence including misinformation and research transparency
Foroutan F1, Naude C2, McCaul M2, Daniel A3, Guyatt G1
1McMaster University, Canada
2Stellenbosch University, South Africa
3World Health Organization, Switzerland

Background: Evidence profiles (EPs), used for evidence-to-decision processes by guideline developers, present relative effects, baseline risk, and risk difference (RD). Due to ease of interpretation, reviewers most often report risk ratios (RR) and calculate RD with 95% CI through application of the RR to baseline risk.
Objectives: To highlight limitations of applying RR to baseline risk for RD calculation in EPs.
Methods: Through methods support for a World Health Organization guideline, we produced EPs summarizing the effectiveness of specially formulated foods compared to counselling alone in children with moderate wasting on anthropometric recovery (desirable outcome). We wanted to present the RD stratified by mid-upper arm circumference (MUAC). Without receiving SFFs, patients with MUAC in the lower end of the moderate wasting range (115 to 119mm; worse prognosis) had a 50% chance of anthropometric recovery, whereas those in the higher end (120 to <125mm; better prognosis) had a 72% chance. We initially applied the treatment effect to the baseline probability of each prognostic group, encountered problems, and identified a solution.
Results: As absolute effects of treatments are always larger in those with a higher baseline risk of poor outcomes, we anticipated finding greater benefit in patients with worse prognosis compared to patients with better prognosis. However, applying the RR to each baseline chance produced counterintuitive results. Applying the treatment RR of 1.29 produced a RD of 145 more anthropometric recoveries per 1000 for those with worse prognosis and 207 more per 1000 for those with better prognosis, suggesting that children with better prognosis will achieve greater benefit compared to those with worse prognosis. We subsequently changed the output from RR to OR, which resulted in 180 more anthropometric recoveries per 1000 in those with worse prognosis and 125 more per 1000 in those with better prognosis. This RD agreed with our expectation that patients with worse prognosis can expect greater benefit from the intervention.
Conclusions: When the probability of a desirable outcome is high, OR applied to the baseline probability will yield a more accurate estimate of absolute benefit than the RR.
Patient, public and/or healthcare consumer involvement: No direct involvement