After retaining the index hospitalization that met a strict definition of aneurysmal subarachnoid hemorrhage, 370 patients remained eligible. The longitudinal trajectory cohort included 318 patients with an albumin measurement within 48 hours of admission and measurements on at least three different days during days 0–14. Mean age was 60.5 years; 183 patients underwent microsurgical clipping, 126 received endovascular treatment, and nine received both.

Latent class mixed models with patient-specific random intercepts were used to identify longitudinal albumin patterns. Two-, three-, and four-class solutions with alternative time functions were compared using statistical fit, Bayesian information criterion, entropy, posterior membership probabilities, class size, and clinical interpretability. The selected three-class model used a natural cubic spline with four degrees of freedom and classified 90 patients as high-preserved, 148 as high-declining, and 80 as low-persistent.

The high-declining group began with albumin values close to those of the high-preserved group before falling rapidly to approximately 31–32 g/L, whereas the low-persistent group began with lower concentrations and remained low. The primary outcome was an unfavorable neurological status at discharge, defined as a modified Rankin Scale score of at least 3. Overall, 114 of 318 patients, or 35.8%, had this outcome.

It occurred in 13 of 90 patients in the high-preserved group (14.4%), 67 of 148 in the high-declining group (45.3%), and 34 of 80 in the low-persistent group (42.5%). In the principal complete-case analysis of 258 patients, adjustment for age, sex, treatment modality, hypertension, and a baseline severity indicator produced adjusted odds ratios of 3.65 (95% CI, 1.57–8.47; p=0.003) for the high-declining trajectory and 2.48 (95% CI, 1.02–6.02; p=0.044) for the low-persistent trajectory, each compared with the high-preserved trajectory; the global p value was 0.007. Those estimates did not demonstrate that either non-preserved trajectory was associated with a worse outcome than the other.

A direct adjusted comparison of high-declining with low-persistent albumin yielded an odds ratio of 1.47 (95% CI, 0.76–2.83; p=0.248). Albumin measured within 48 hours was not significantly associated with discharge outcome after adjustment, with an odds ratio of 0.97 per 1 g/L increase (95% CI, 0.91–1.03; p=0.286). Nadir albumin during days 0–14 was associated with the outcome, with an adjusted odds ratio of 0.76 per 1 g/L increase (95% CI, 0.69–0.84; p<0.001), although this post-baseline measurement may reflect evolving illness.

Landmark analyses supported a temporal association without eliminating reverse causation. After excluding patients who had been discharged or died by each landmark and requiring at least two measurements in the relevant window, each 1 g/L higher early nadir albumin was associated with lower odds of subsequent unfavorable discharge status. The adjusted odds ratio was 0.86 (95% CI, 0.79–0.93) in the day-3 analysis of 216 patients with 84 events and 0.84 (95% CI, 0.77–0.91) in the day-7 analysis of 250 patients with 91 events; both p values were below 0.001.

Restriction to patients with an available Hunt–Hess grade and multiple imputation for missing baseline severity produced trajectory associations in the same general direction. Sensitivity analyses also showed uncertainty in the trajectory findings. There was no evidence of a trajectory-by-treatment interaction between clipping and endovascular treatment (p=0.453), although treatment selection was not randomized.

Adjustment for recorded pneumonia, cerebral infarction, hydrocephalus, and intracranial infection attenuated the low-persistent estimate to 2.09 (95% CI, 0.81–5.41; p=0.129), while the high-declining estimate remained 3.35 (95% CI, 1.37–8.15; p=0.008). Adding trajectory class to a clinical model provided only a modest increase in apparent discrimination. The C-statistic rose from 0.754 to 0.776.

After 1,000 bootstrap resamples with class assignments held fixed, the optimism-corrected values were 0.735 and 0.753, respectively. The three-class model had only moderate separation, with entropy of 0.520, and 33% of patients had an assigned posterior probability below 0.70. A two-class spline model had the lowest Bayesian information criterion, only 1.72 points below the selected three-class solution, and the bootstrap procedure did not re-derive the trajectory classes.

The observational, single-center design limits causal inference and generalizability. Trajectory membership and nadir albumin incorporated measurements through day 14, when complications, treatment, nutritional changes, or deterioration could already have affected albumin levels. Routine rather than fixed laboratory testing meant fewer observations later in follow-up, while potential confounders including exogenous albumin administration, cumulative fluid balance, nutritional delivery, organ dysfunction, and delayed cerebral ischemia were not consistently available.

Some discharge modified Rankin Scale scores were retrospectively adjudicated, and standardized 90-day scores were not systematically available. The authors therefore characterized serial albumin trajectories as markers of the evolving systemic and clinical course rather than independently validated predictors, with prospective multicenter validation and standardized long-term outcome assessment needed.

Subject of Research: Association of longitudinal serum albumin trajectories with discharge neurological status after aneurysmal subarachnoid hemorrhage

Article Title: Longitudinal serum albumin trajectories and neurological outcomes after aneurysmal subarachnoid hemorrhage

Article References: Tang, H., Tang, Y., Wu, S., Zhang, Q., & Gong, Y. (2026). Longitudinal serum albumin trajectories and neurological outcomes after aneurysmal subarachnoid hemorrhage. Frontiers in Nutrition, 13, Article 1979070. https://doi.org/10.3389/fnut.2026.1979070

Image Credits: AI Generated

DOI: 10.3389/fnut.2026.1979070

Keywords: aneurysmal subarachnoid hemorrhage, serum albumin, albumin trajectories, neurological outcome, modified Rankin Scale, latent class mixed model, retrospective cohort, clinical nutrition, Longitudinal, serum, albumin, trajectories