This multicenter cross-sectional study was conducted across 26 tertiary rehabilitation centers in China between January 2023 and December 2024 (Figure 1). The study protocol was approved by the Ethics Review Committee of Huashan Hospital, Fudan University (approval number: HIRB2022-510). The study was registered in the Chinese Clinical Trial Registry (registration number: ChiCTR2200063611). Written informed consent was obtained from all participants or their legal representatives. The study adhered to the Declaration of Helsinki and STROBE guidelines for observational studies (13).

A total of 1,741 adults (≥18 years) with clinically confirmed ischemic or hemorrhagic stroke were enrolled. Patients were excluded based on clinical conditions that could substantially confound the assessment of functional recovery or precipitate clinical deterioration during evaluation, including: (1) Acute Uncontrolled Medical Conditions: Severe or poorly controlled hypertension, advanced heart failure, severe active infection (e.g., pneumonia), diabetic ketoacidosis, uncontrolled or frequent seizures, or other acute medical problems judged by the treating clinician as likely to worsen during assessment. (2) Uncontrolled Systemic Disorders: Overt thyroid dysfunction (hyper- or hypothyroidism), severe hepatic or renal impairment (including dialysis dependence), active acute rheumatologic disease, or clinically significant electrolyte imbalances. (3) Data Quality: Records with missing key covariates or outcomes, implausible values, or duplicate entries were excluded during the final data cleaning process, yielding the final analytic cohort.

All clinical and demographic data were collected and managed using a secure, electronic data capture platform (REDCap; https://himedc.huashan.org.cn:5288/) hosted at the Huashan Hospital, Fudan University coordinating center, ensuring data privacy and quality control across all participating sites.

Potential confounding variables were collected to ensure comprehensive adjustment in the multiple linear regression models. These covariates included Demographic and Lifestyle Factors (Sex, Age, Education level, Body Mass Index [BMI], Smoking status, and Alcohol status), Comorbidities (History of Hypertension, Coronary Artery Disease [CAD], Diabetes Mellitus [DM], and Previous stroke history), and Stroke Characteristics. The detailed stroke characteristics encompassed Day_of_illness (time span in days from stroke onset to functional assessment), Stroke diagnosis (Ischemic vs. Hemorrhagic), and Lesion location (lesions involving the Basal Ganglia[BG], Brainstem[BS], and Cerebellum[CB]) as confirmed by neuroimaging reports.

Participants were stratified into quartiles based on ALB levels: Q1 (<37.7 g/L), Q2 (37.7–40.0 g/L), Q3 (40.0–42.8 g/L), and Q4 (≥42.8 g/L). Descriptive statistics are presented as mean ± standard deviation (SD) for normally distributed continuous variables, median (interquartile range) for skewed data, and frequency (%) for categorical variables. Group comparisons were performed using ANOVA or Kruskal-Wallis tests for continuous variables and χ² tests for categorical variables, as appropriate. Statistical significance was set at p < 0.05 (two-tailed).

Multivariable linear regression models assessed associations between ALB levels and functional outcomes (FMA, ADL, BBS, MoCA). Three models were constructed: Model 1 Unadjusted; Model 2: Adjusted for sex, age, education, smoking, alcohol status, BMI, hypertension diabetes mellitus (DM), coronary artery disease (CAD), and previous stroke; Model 3: Further adjusted for Model 2 covariates plus illness duration (days) and lesion locations (BG, BS, CB). ALB was analyzed continuously (per 1-g/L increase) and categorically (quartiles: <37.7 g/L [reference], 37.7–40.0 g/L, 40.0–42.8 g/L, ≥42.8 g/L). Results are presented as regression coefficients (β) with 95% confidence intervals (CI). Linear trend tests across quartiles were performed. Significance was defined as p < 0.05.

Subgroup analyses were conducted using stratified multivariable linear regression models to assess the heterogeneity of associations between ALB and functional outcomes (FMA, ADL, BBS, MoCA) across prespecified subgroups. Subgroups included age (<60 vs. ≥60 years), sex (male vs. female), BMI (<24 vs. ≥24 kg/m²), and lesion locations (BG, BS, CB). Interaction terms (ALB × subgroup) were tested, with significance defined as p < 0.05. Results are presented as β-coefficients per 1-g/L ALB increase with 95% confidence intervals (CI).

The baseline characteristics of the 1,741 post-stroke patients, categorized by quartiles of ALB (Q1: < 37.7 g/L to Q4: ≥42.8 g/L), are presented in Table 1. Significant differences across the albumin quartiles were observed for Sex (p < 0.001), with males constituting a larger proportion in the highest quartile (77%). Continuous variables, including Age (p < 0.001), BMI (p < 0.001), and ALB itself (p < 0.001), also differed significantly across the groups. Notably, the mean age decreased progressively from Q1 (65.7 ± 11.2 years) to Q4 (57.0 ± 13.2 years). Significant differences were also found in the prevalence of Hypertension (p = 0.04) and Diabetes Mellitus (DM) (p = 0.012), with the highest prevalence of DM (13.9%) recorded in Q4. Furthermore, Day_of_illness differed significantly (p < 0.001), showing the longest median duration in Q1 (36.0 days) and the shortest in Q4 (23.0 days). Education, smoking, alcohol use, and comorbidities (CAD, stroke history, ischemic stroke) did not differ significantly among groups. All four primary functional outcome scores—FMA, ADL, BBS, and MoCA—showed significant differences across the quartiles (p < 0.001 for all), with mean or median scores consistently increasing from the lowest albumin quartile (Q1) to the highest (Q4).

Table 2 summarizes the results of the multiple linear regression analysis examining the association of ALB with FMA, ADL, BBS, and MoCA scores. The analysis included three models, with Model 3 being maximally adjusted for sex, age, education, smoking, drinking, BMI, hypertension, DM, CAD, previous stroke, Day_of_illness, and lesion locations (BG, BS, and CB). ALB levels exhibited significant positive associations with all functional outcomes across incremental adjustment models (Table 2). In the fully adjusted Model 3, ALB, analyzed as a continuous variable (per 1 g/L increase), was significantly associated with higher scores in FMA (β = 1.35, 95% CI: 0.99 ~ 1.72, p < 0.001), ADL (β = 1.77, 95% CI: 1.44 ~ 2.1, p < 0.001), BBS (β = 1.02, 95% CI: 0.78 ~ 1.26, p < 0.001), and MoCA (β = 0.30, 95% CI: 0.21 ~ 0.40, p < 0.001). When ALB was analyzed categorically (quartiles), a significant dose-dependent relationship was observed for all outcomes. Patients in the highest ALB quartile (≥42.8 g/L) exhibited markedly higher FMA (β = 15.11, 95% CI: 11.09–19.12), ADL (β = 19.35, 95% CI: 15.76–22.93), BBS (β = 11.61, 95% CI: 9.00–14.22), and MoCA scores (β = 3.12, 95% CI: 2.07–4.17) compared to the lowest quartile (all p < 0.001). Trend tests confirmed progressively stronger functional improvements across ascending ALB quartiles (all p < 0.001). Associations remained robust after adjusting for demographics and clinical covariates.

The figure displays the coefficients (β) and 95% confidence intervals (95% CI) representing the change in functional score per 1 g/L increase in serum ALB, stratified by predefined baseline characteristics. Each subgroup analysis was conducted using a multiple linear regression model (similar to Model 3) where the association between ALB and the outcome was assessed within each stratum. The models were comprehensively adjusted for all covariates, excluding the stratification factor itself.

The stability of the independent association between ALB (analyzed as a continuous variable per 1 g/L increase) and the four functional outcomes (FMA, ADL, BBS, and MoCA) was evaluated across predefined clinical subgroups, with the results presented in Figure 2.

Subgroup analyses revealed significant effect modification by sex for FMA (p = 0.017), with a stronger albumin-FMA association in females (β = 1.81, 95% CI: 1.12–2.51) than males (β = 1.15, 0.72–1.58). Overall, the positive association between higher ALB levels and better functional outcomes remained consistent (β > 0) across most subgroups defined by Age (< 60 vs. age ≥60 years), BMI (< 24 vs. ≥24 kg/m²), stroke onset time and lesion location (Basal Ganglion, Brainstem, and Cerebellum presence). No Significant Interaction for Other Subgroups. The association of ALB with FMA, ADL, BBS, and MoCA scores did not differ significantly by age, BMI, stroke onset time or the presence of lesions in the basal ganglion, brainstem, or cerebellum.

The major strengths of this study include the large sample size (N = 1741) drawn from multiple centers, enhancing the external validity of our findings. Crucially, we employed highly specific, validated measures for each functional domain (FMA, BBS, MoCA, ADL), moving beyond single gross disability scores. Furthermore, the inclusion of a comprehensive set of covariates, including lesion location (BG, BS, CB) and Day_of_illness, allowed us to establish the independent association of ALB, minimizing residual confounding. The demonstration of a robust dose-dependent relationship across ALB quartiles strengthens the causal inference. Despite its strengths, this study has several limitations inherent to its cross-sectional design. First, the cross-sectional nature precludes the establishment of temporal or causal relationships; while we adjusted for time since stroke, reverse causality (i.e., severe disability leading to poor nutritional intake and low ALB) cannot be entirely excluded. Second, ALB levels reflect both nutritional status and inflammation; we did not adjust for specific inflammatory markers (e.g., CRP), which could introduce residual confounding. Third, the cohort was derived from rehabilitation centers, which may introduce selection bias towards patients eligible and stable enough for intensive rehabilitation.