Interaction between oxygen saturation and renal function on 30-day mortality in emergency department patients

Emerging evidence suggests that bidirectional lung–kidney crosstalk may influence outcomes, but this has not been systematically evaluated in unselected emergency department populations. We therefore examined the association between peripheral oxygen saturation (SpO₂), estimated glomerular filtration rate (eGFR), and 30‑day mortality, and tested whether the prognostic association of oxygenation with mortality differs across levels of kidney function (and vice versa), using an SpO₂×eGFR interaction term to model effect modification We analyzed 12,651 adults with complete data on SpO₂, creatinine-derived eGFR, lactate, C-reactive protein (CRP), RETTS triage, and prespecified covariates from the Skåne Emergency Medicine (Skåne 17/18) cohort (2017–2018). We fitted multivariable logistic regression models including SpO₂ and eGFR (Model 1) and then added an SpO₂×eGFR interaction term (Model 2). Nested models were compared using a likelihood-ratio test, and discrimination was compared using AUROC (DeLong test) based on model-predicted probabilities. In a predefined subgroup with arterial blood gases (n = 3,068), we evaluated eGFR in relation to PaO₂/FiO₂ (P/F). In the full cohort, SpO₂ and eGFR were significantly correlated. Adding the SpO₂×eGFR interaction term improved model fit versus the main-effects model (LRT ΔDeviance = 15.77, p = 7.17 × 10⁻⁵), but discrimination was essentially unchanged (AUROC 0.744 vs. 0.745; ΔAUROC 0.0009; 95% CI − 0.00227 to 0.00046; DeLong p = 0.193). In the interaction model, higher SpO₂ and eGFR were associated with lower 30-day mortality (OR 0.81, 95% CI 0.77–0.85; OR 0.85, 95% CI 0.80–0.91), and the interaction term indicated stronger protection when both were higher (OR 0.90, 95% CI 0.86–0.95). Exploratory subgroup analyses suggested the interaction effect was most pronounced among patients presenting with chest pain. SpO₂ and eGFR showed evidence of interaction in relation to 30-day mortality. Although adding the interaction term improved model fit, it did not meaningfully improve discrimination compared to main model (without interaction). This suggests that, in unselected ED populations, measuring and interpreting SpO₂ and eGFR remains clinically useful, whereas explicitly modeling their interaction is unlikely to add substantial predictive benefit.