Association of early anion gap with 28-day mortality: a landmark cohort study in MIMIC-IV with multicenter validation...

Early laboratory measures in ICU studies are often defined over a post-admission window while follow-up starts at ICU admission, which can introduce time-window and immortal time bias. We evaluated the association between early anion gap (AG) and 28-day mortality using prespecified landmark analyses with external validation. We analyzed adult first ICU stays in MIMIC-IV (derivation) and eICU-CRD (external validation). Early AG measures were the first AG within 0-12 h (AG_first_12h; 12-h landmark) and the maximum AG within 0-24 h (AGmax_24h; 24-h landmark). In patients with albumin measured within 24 h, albumin-corrected AG was calculated using the Figge correction (cAG = AG + 2.5 × (4.0 - albumin [g/dL])) and summarized as cAGmax_24h. The primary outcome in MIMIC-IV was 28-day all-cause mortality measured from each landmark; in eICU-CRD, 28-day in-hospital mortality was used as a proxy endpoint, with patients discharged alive before day 28 treated as survivors. Associations were estimated using multivariable logistic regression; Cox models were additionally performed in MIMIC-IV, and restricted cubic spline analyses were conducted in both databases. Because several covariates were summarized over 0-24 h, fully adjusted 12-h models were interpreted as extended-adjustment sensitivity analyses. In MIMIC-IV (12-h landmark cohort, N = 57,220), 28-day mortality increased from 6.3% in the lowest to 24.2% in the highest AG_first_12h quartile. In the extended-adjustment model, the highest versus lowest quartile was associated with higher mortality (OR 2.32, 95% CI 2.11-2.54), and each 1 mmol/L increase in AG_first_12h corresponded to OR 1.06 (95% CI 1.05-1.06). In the temporally aligned 24-h landmark cohort, associations persisted for AGmax_24h (Q4 vs. Q1 OR 1.98, 95% CI 1.81-2.17) and for cAGmax_24h in the albumin subset (Q4 vs. Q1 OR 1.91, 95% CI 1.64-2.22). In eICU-CRD, effect estimates were directionally consistent for AGmax_24h (Q4 vs. Q1 OR 1.64, 95% CI 1.43-1.89) and cAGmax_24h (Q4 vs. Q1 OR 1.79, 95% CI 1.51-2.11).Higher early AG measures, including albumin-corrected AG, were consistently associated with increased 28-day mortality across two large ICU databases using a landmark framework that improves temporal alignment. These findings support early AG as a readily available marker of metabolic and physiologic derangement, but they do not establish incremental prognostic value beyond existing severity scores or demonstrate clinical utility for risk stratification.