Letter to the editor: Odds ratio versus risk ratios: Rejuvenating an important debate and an opportunity to learn

Dear Editor, We read with great interest the letter by Dr. Montero [1], which provides thoughtful comments on our recent systematic review and meta-analysis comparing bedside versus conventional operating room surgery in critically ill neonates [2]. We appreciate his careful reading and engagement with our work. At the same time, we wish to clarify several methodological points and address certain criticisms that, in our view, merit further discussion. It is correct that risk ratios (RRs) appear more intuitive and are often recommended when outcomes are common [3]. However, apparent simplicity does not guarantee validity. As Cook has persuasively argued [4], RRs may be misleading when heterogeneity of baseline risk exists, since the assumption of a homogeneous RR across individuals rarely holds. In contrast, odds ratios (ORs) are less dependent on such assumptions and often provide a more robust and generalizable measure, especially when synthesizing data across heterogeneous populations. The frequent claim that ORs “overestimate” associations applies only when they are erroneously interpreted as approximations of RRs. When understood correctly, both measures estimate different but equally legitimate population parameters. Indeed, ORs remain the standard effect measure in meta-analyses, particularly where included studies vary widely in baseline risk. In such contexts, ORs are not only mathematically coherent but also methodologically preferable, since RRs may actually understate intervention differences when outcomes are common [4]. Moreover, the symmetry of the OR represents a distinct advantage. As noted by both Cook [4] and Senn [5], ORs are reciprocal with respect to the definition of the outcome and the exposure, avoiding the arbitrariness inherent in RRs, which may differ depending on whether one defines “success” or “failure.” This property was particularly relevant in our review, where the designation of “intervention” and “comparator” was reversed at the analysis stage for clinical clarity; such reclassification alters RRs but leaves ORs invariant. We also acknowledge the broader debate summarized by Cummings [6], who emphasizes the collapsibility of RRs and their intuitive appeal. These features are valuable in certain contexts, but they do not negate the advantages of ORs in meta-analytic applications where baseline risks differ substantially and outcomes are not rare. Indeed, the non-collapsibility of ORs, often portrayed as a drawback, reflects their distinct mathematical property and does not undermine their validity as an effect measure. What is essential is that ORs be reported transparently and interpreted as such, rather than conflated with RRs. We further recognize the limitations inherent in our field. Our PROSPERO record reflects the evolving nature of neonatal surgical research, where prespecified outcomes (e.g., recurrence, Clavien—Dindo classification) could not be uniformly captured across studies. This deviation was transparently reported. Likewise, we discussed at length the potential for confounding by indication, given that bedside surgery is often reserved for the most fragile neonates. Excluding such patients would have misrepresented clinical reality. Regarding operative time, we respectfully disagree with the assertion that our GRADE rating was indefensible. While reasonable differences in judgment may exist, the consistency and objectivity of the included findings justified upgrading certainty in this instance. Nevertheless, we acknowledge that a more conservative rating would also be defensible. Finally, we concur that individual patient data (IPD) meta-analyses and adjusted observational studies are needed to more definitively disentangle confounding factors. In the interim, our analysis, conducted with standard random-effects modeling and continuity correction for sparse data, offers the most comprehensive synthesis currently available and highlights areas where future research must be directed. We thank Dr. Montero for prompting this important methodological discussion. Far from a sterile statistical dispute, the debate between ORs and RRs is an opportunity for the field to advance toward both clarity and rigor. To our knowledge, this meta-analysis offers the most comprehensive synthesis of evidence to date and represents a valuable foundation for future studies, ultimately providing useful guidance for clinicians facing critical decisions in fragile neonatal populations.