Baseline and 72-h dynamic laboratory deficit burden for mortality through day 90 after acute ischemic stroke in critical care: landmark analyses with temporal validation.
Gu G, Shi H, Wang G, Chen J • Frontiers in neurology • 2026
Baseline FI-Lab improved internally validated prediction of mortality through day 90 among patients alive and evaluable at 24 h, while at 72 h dynamic change added limited global predictive information and the four-level phenotype remained exploratory.
Key Findings
Results
Adding baseline FI-Lab to a clinical logistic model significantly improved 90-day mortality discrimination at the 24-hour landmark in temporal validation.
Validation AUC improved from 0.774 to 0.808 (paired difference 0.034, 95% CI 0.022–0.047; p < 0.001).
Brier score was reduced from 0.162 to 0.152 (difference −0.011, 95% CI −0.015 to −0.006; p < 0.001).
The complete-case 24-h cohort comprised 3,186 patients (development: n = 1,890, 559 events; temporal validation: n = 1,296, 339 events).
The analysis used patients alive with evaluable 0–24 h FI-Lab from MIMIC-IV v3.1, with the outcome defined as death after the 24-h landmark through day 90 after ICU admission.
Results
Adding baseline FI-Lab also improved discrimination at the 72-hour landmark, increasing validation AUC from 0.752 to 0.791.
The complete-case 72-h cohort comprised 2,918 patients (development: n = 1,754, 473 events; validation: n = 1,164, 287 events).
The 72-h cohort required patients to be alive with evaluable baseline and 24–72 h FI-Lab measurements.
Among 3,376 total admissions, 39 deaths occurred by 24 h and 148 by 72 h.
Results
At the 72-hour landmark, adding dynamic FI-Lab change direction or continuous change to baseline FI-Lab provided only marginal and statistically unsupported improvements in AUC.
Adding change direction yielded an AUC of 0.797, and adding continuous change yielded an AUC of 0.798, compared to 0.791 for baseline FI-Lab alone.
Neither improvement over baseline FI-Lab alone was statistically supported.
The four-level phenotype combining baseline level and change direction had an AUC of 0.767, which was lower than the baseline FI-Lab model.
Results
The four-level FI-Lab phenotype (combining baseline level and directional change) remained exploratory and did not outperform baseline FI-Lab alone for global prediction.
The four-level phenotype AUC of 0.767 was below the baseline FI-Lab model AUC of 0.791 at the 72-h landmark.
The authors characterized the phenotype as 'exploratory.'
Decision curves and calibration assessments were included alongside discrimination metrics.
Methods
The study used an internal temporal validation design with MIMIC-IV v3.1 data, separating patients into development and validation cohorts based on calendar time rather than random splitting.
Two separate landmark cohorts were constructed: a 24-h cohort and a 72-h cohort, each with distinct development and validation splits.
The landmark approach ensured that prediction times preceded the 90-day outcome follow-up window.
Sensitivity analyses were pre-defined for the revised analysis, and methods included discrimination, calibration, decision curves, and paired bootstrap differences.
The authors noted that external validation is required before clinical use.
Background
The laboratory-based frailty index (FI-Lab) was conceptualized as capturing both acute physiological derangement and underlying vulnerability during critical illness.
The authors noted that during critical illness, FI-Lab 'may reflect acute derangement and underlying vulnerability,' motivating analysis of both its baseline and dynamic roles.
Baseline and 24–72 h FI-Lab values were required for inclusion in the 72-h cohort.
What This Means
This research suggests that a score calculated from routine laboratory tests — called a laboratory-based frailty index (FI-Lab) — measured within the first 24 hours of ICU admission can meaningfully improve predictions of whether a stroke patient will survive to 90 days. Using a large critical care database (MIMIC-IV), the researchers tested this by splitting patients into earlier and later time groups (a method called temporal validation), and found that adding the FI-Lab score to standard clinical information raised the accuracy of 90-day mortality prediction from an AUC of 0.774 to 0.808 — a statistically significant improvement. The score also improved how well-calibrated the predictions were, meaning the predicted probabilities better matched actual outcomes.
The researchers also examined whether tracking how the FI-Lab score changed between hospital admission and 72 hours (going up, going down, or staying the same) added extra predictive value on top of the initial score. They found that while the baseline score at 72 hours was also informative, the additional benefit of knowing whether the score improved or worsened was small and not statistically convincing. A four-category grouping combining baseline level and change direction performed worse than using the baseline score alone and was considered preliminary or exploratory.
This research suggests that routinely collected laboratory data could help clinicians identify acute ischemic stroke patients in the ICU who are at higher risk of dying within 90 days, potentially enabling more targeted monitoring or care discussions. However, the authors emphasize that this study used only one internal dataset, and the findings need to be confirmed in different hospitals and patient populations before the approach could be considered for real-world clinical decision-making.
Gu G, Shi H, Wang G, Chen J. (2026). Baseline and 72-h dynamic laboratory deficit burden for mortality through day 90 after acute ischemic stroke in critical care: landmark analyses with temporal validation.. Frontiers in neurology. https://doi.org/10.3389/fneur.2026.1886148