Cortisol-to-albumin ratio and mortality in vasopressor-dependent critical illness: a four-database cohort study

Total cortisol is frequently measured during vasopressor-dependent critical illness, but changes in cortisol-binding proteins complicate its interpretation. We examined the association of the cortisol-to-albumin ratio (CAR) with mortality and evaluated CAR alongside total cortisol and albumin considered separately. We used the Medical Information Mart for Intensive Care (MIMIC)-IV for derivation, the eICU Collaborative Research Database (eICU-CRD) and Northwestern ICU (NWICU) database for external replication, and MIMIC-III CareVue for historical sensitivity analysis. Eligible adults had suspected infection, early vasopressor support, and cortisol and albumin measured by 24 hours after shock onset. CAR was total cortisol (µg/dL) divided by albumin (g/dL), giving a numerical scale of µg/g. At a 24-hour landmark, continuous CAR was the primary exposure; CAR ≥ 10 and ≥15 µg/g were exploratory. Additional analyses addressed severity, missing data, survival, interaction, and outcome-specific meta-analysis. Albumin was available in 381 of 603 stays with time-ordered cortisol measurements. The 24-hour landmark MIMIC-IV cohort included 335 stays and 142 deaths. The adjusted ORs were 1.09 (95% CI, 0.99–1.19) per 10 µg/dL for total cortisol and 1.29 (95% CI, 1.03–1.62) per 10 µg/g for continuous CAR. Database-specific adjusted ORs for continuous CAR per 10 µg/g were 1.55 (1.36–1.76) in eICU-CRD, 1.31 (0.98–1.77) in NWICU, and 1.10 (0.96–1.25) in MIMIC-III CareVue. All point estimates were above 1, although 95% CIs included 1 in NWICU and CareVue. In MIMIC-IV, the continuous-CAR OR was 1.25 (1.00–1.55) after Sequential Organ Failure Assessment (SOFA) adjustment and 1.16 (0.93–1.44) after Acute Physiology Score III (APS III) adjustment. CAR and separate-component models had similar discrimination. Continuous CAR estimates were directionally consistent across datasets, but CAR did not outperform modeling cortisol and albumin separately. CAR does not directly measure free cortisol and should not be used to diagnose critical illness-related corticosteroid insufficiency (CIRCI) or guide corticosteroid treatment.