Prognostic value of 48-h reassessment of a laboratory-derived frailty index in ICU patients with sepsis: model development in MIMIC-IV and external transport evaluation

Background Static laboratory-derived frailty index (FI-lab) measures have been associated with adverse outcomes in sepsis, but the value of reassessing acute multisystem laboratory abnormality burden after initial resuscitation and transporting a prespecified model to a new clinical setting remains uncertain. Methods We conducted a retrospective 48-h landmark study using MIMIC-IV for model development and an independent single-center ICU sepsis cohort from January 2020 to December 2025 for supportive single-center transport evaluation. FI-lab-23 was calculated from 23 routine laboratory variables using prespecified thresholds, with at least 17 components required, and was interpreted as an acute laboratory deficit-accumulation index rather than chronic frailty. The primary extended model added FI-lab-23 measured at 24–48 h to clinical variables; the strict secondary model additionally included lactate clearance. Fixed-horizon logistic regression predicted 28-day mortality among patients alive and remaining in the ICU at 48 h. MIMIC-IV coefficients were frozen and applied without refitting to the local cohort. Performance was assessed by discrimination, Brier score, calibration, and exploratory decision-curve analysis. Baseline/change parameterizations were treated as secondary specification analyses; sensitivity analyses addressed testing intensity, component completeness, and a fixed 10-component common-core FI. Results The primary MIMIC-IV development cohort included 6,153 patients with 1,594 deaths. Each 0.05 increase in 24–48 h FI-lab-23 was associated with higher 28-day mortality (OR: 1.178, 95% CI: 1.149–1.207). Optimism-corrected AUC was 0.710. Of 722 local landmark-eligible patients, 328 met the primary complete-case requirements (85 deaths). In this selected cohort, the frozen model achieved an AUC of 0.774, Brier score of 0.160, calibration intercept of 0.109, calibration slope of 1.273, and observed-to-expected ratio of 1.074. Baseline FI-lab did not improve fit after the 48-h level was known, and baseline-by-change interactions were not significant. The direction of association was preserved across sensitivity analyses, although effect magnitude varied with component-completeness criteria. Conclusions FI-lab-23 reassessed at 24–48 h provided updated prognostic information for 28-day mortality in ICU patients with sepsis. The frozen MIMIC-IV model showed encouraging but imprecise performance in a selected single-center complete-case cohort. These findings constitute supportive transport evidence rather than definitive external validation, and larger multicenter studies are required before clinical implementation.