<p>The datas of participants in MIMIC.</p>

Objective To investigate the association between logarithmically transformed lactate-to-hemoglobin ratio (Ln_LHR) and 28-day mortality in sepsis patients, addressing the critical need for reliable prognostic biomarkers in this high-mortality condition. Methods This retrospective cohort study analyzed 20725 adult sepsis patients from the MIMIC-IV database (2008–2019). The primary exposure was Ln_LHR calculated from measurements within 24 hours of ICU admission, while the primary outcome was 28-day all-cause mortality. Covariates included demographics, physiological parameters, severity scores, and interventions. We employed multivariable logistic regression and restricted cubic splines to identify potential non-linear relationships.Finally,mediation analysis was used to assess the factors affecting sepsis mortality in Ln_LHR. Results Elevated Ln_LHR was independently associated with increased 28-day mortality after comprehensive adjustment (OR:1.43, 95% CI:1.35–1.52). We identified a significant threshold effect at Ln_LHR: −0.625, above which mortality risk increased dramatically (OR:3.812, 95% CI:3.131–4.642). Subgroup analyses revealed the predictive efficacy of Ln_LHR exhibited significant variation across various factors, including age, the utilisation of norepinephrine, and the severity score. Mediation analysis revealed that minimum temperature, as a potential mechanism linking Ln_LHR to 28-day mortality, accounted for 9.2% of the total association. Conclusion Ln_LHR may represent a promising, readily available prognostic biomarker for sepsis mortality risk stratification, with a clinically meaningful threshold effect. This composite marker integrates critical pathophysiological information and may enhance risk assessment and guide clinical decision-making in sepsis management.