This study was a longitudinal study conducted at the stroke centers of two tertiary hospitals from April 2023 to June 2024 in Xuzhou, China. The inclusion criteria for patients were as follows: (1) diagnosed with AIS by a neurologist according to a focal neurologic deficit and a corresponding infarct on computed tomography (CT) or magnetic resonance imaging (MRI), (2) age 18 years or older, and (3) <2 weeks since stroke onset. The exclusion criteria were as follows: (1) diagnosed with mental disorder, cognitive or emotional impairment prior to stroke onset, (2) having a transient ischemic attack (TIA), and (3) life expectancy < 6 months.

Two trained neurologists recruited eligible patients for this study. Two trained research assistants interviewed patients and their primary family caregivers in person to collect sociodemographic and stroke-related information, combined with medical records.

Baseline assessment (T1) was performed during the acute hospitalization of patients (within 7–10 days following acute stroke) and follow-up assessments were conducted by telephone interviews at 3 months (T2) and 6 months (T3) after stroke onset (22, 23). Supplementary Table S1 presents the measures collected at each time point.

The pertinent covariates collected at baseline (T1) were composed of sociodemographic factors (age, sex, marital status, educational level, residence, occupational status, household income/yuan per year), severity of stroke [National Institutes of Health Stroke Scale (NIHSS)] (35, 36), classification of stroke [Oxfordshire Community Stroke Project (OCSP) classification, Trial of ORG 10172 in Acute Stroke Treatment (TOAST) classification] (37, 38), therapeutic options (intravenous thrombolysis (39, 40), endovascular thrombectomy (41, 42), conservative therapy (39, 43)), medical comorbidities (hypertension, diabetes, dyslipidemia, atrial fibrillation), risk factors (history of stroke, smoking status, drinking status, body mass index [BMI]), and stroke rehabilitation. A detailed coding of covariates is listed in Supplementary Table S2. Endovascular treatment included intra-arterial thrombectomy, arterial thrombolysis, and emergency angioplasty within 6 to 24 h of stroke onset (44). BMI was calculated by weight (kg) divided by square of height (m²), and ≥24 kg/m² was defined as overweight or obesity referring to the Chinese criteria for adults (45). Stroke rehabilitation was provided by occupational therapy, physical therapy, and speech and language therapy during hospitalization (46).

To ensure data quality, all neurologists and research assistants involved in data collection received standardized and homogeneous training, which covered interpretation of the study protocol, standardized operation procedures for scales, interview skills, and data recording requirements. Only those who passed the post-training assessment were permitted to participate in the study, so as to ensure the consistency of assessments. A database was established using structured electronic data capture forms with EpiData 3.1 software, where logic checks and range restrictions were set to prevent invalid data entry. All data underwent double data entry by two independent research assistants, followed by cross-verification to ensure the accuracy of data entry.

The GBTM assumes that the population has inter-individual heterogeneity and aims to identify homogeneous classes of individuals following similar developmental trajectories over time (17). We applied GBTM to identify potential trajectories of mRS and BI from stroke onset to recovery period (within 6 months) in AIS patients (47). The trajectory models tested from 2 trajectory classes to 6 trajectory classes (i.e., linear, quadratic, and cubic) were fitted until the best model was determined. The optimal number of mRS or BI trajectory groups was determined based on the following criteria: (1) Bayesian information criterion (BIC), with a lower absolute value indicating a better fit; (2) average posterior probability (AvePP) of assigning an individual to a trajectory group ≥ 0.70; and (3) proportion of the participants belonging to each trajectory group > 5% (21, 48). Trajectories of mRS or BI were analyzed with the lcmm package (49).

Baseline characteristics of AIS patients were described as numbers (n) and proportion (%) for categorical variables, while continuous variables were described as mean ± standard deviation (SD) or median and interquartile range (IQR). Analysis of variance, Kruskal-Wallis H test, Chi-square test, or Fisher’s exact test were conducted to compare the baseline characteristics of included AIS patients across distinct trajectory groups. A series of multinomial logistic regression models were performed to explore the association between early depressive symptoms and trajectories of functional recovery (mRS, BI) over time. Firstly, we ran the univariate analysis to test the association between early depressive symptoms, covariates and functional recovery trajectories, respectively (Supplementary Tables S3, S4). Then, only those significant ( p<0.05) univariate factors were allowed to enter into the final multivariate logistic regression models, and multicollinearity was assessed among all variables using the variance inflation factor. To assess the effect of early depressive symptoms on trajectories of functional recovery, we incorporated covariates adjustment into the models in turn. For the mRS trajectories, (1) Model I was the base model adjusting for baseline depressive symptoms; (2) Model II: model I plus stroke-related factors (NIHSS, stroke rehabilitation, OCSP classification, hypertension, diabetes, atrial fibrillation, and smoking status); (3) Model III: model II plus sociodemographic factors (age, sex, and marital status). For the BI trajectories, (1) Model I was the base model adjusting for baseline depressive symptoms; (2) Model II: model I plus stroke-related factors (NIHSS, stroke rehabilitation, OCSP classification, hypertension, and diabetes); (3) Model III: model II plus sociodemographic factor (sex). Odds ratios (ORs) and 95% confidence intervals (CIs) for the effect of early depressive symptoms on mRS or BI trajectories were reported.

To further assess the robustness of the set of logistic regression models, the E-values were computed to evaluate the minimum strength of association, on the OR scale, that an unmeasured confounder would need to have with both early depressive symptoms and distinct trajectories of functional recovery to fully nullify the observed association, conditional on the measured covariates (50). Considering the sample size distribution among BI trajectory subgroups and the stability of the model, another sensitivity analysis was performed by excluding the small “moderate high–rapid decline” subgroup (accounting for 8.3% of the sample). The BI trajectory model was refitted to re-examine the association between early depressive symptoms and functional recovery trajectories. All statistical analyses were performed using R software version 4.3.2 and IBM SPSS version 26.0. The statistical significance was set at two-sided p < 0.05 throughout.

Figure 1 shows the flow diagram of AIS patients assessed at each time point. In this study, we included patients who had mRS and BI measurements at three time points (T1, T2, and T3) and excluded those with missing data due to loss to follow-up or death. A total of 219 patients were enrolled in the study. Participants who withdrew from the study (n = 11) showed no significant differences in depressive symptoms ( p = 0.217) at baseline, age ( p = 0.468), gender ( p = 0.154), marital status ( p = 0.067), and comorbidities ( p > 0.05), but had higher NIHSS scores ( p = 0.007). It could be explained that those patients were more likely to discontinue their participation from the study due to more severe brain damage. Table 1 presents the baseline characteristics of the study patients, which was consistent with those of stroke patients reported by the China National Stroke Registry (CNSR) (51, 52).

Three distinct mRS trajectory groups were identified as the best model according to the recommended statistical criteria (Figure 2A; Supplementary Table S5). The AIS patients were divided into 3 groups based on their mRS trajectories: Group 1 (mild trajectory, 41.6%) contained 91 patients with mRS scores below 2 during the first 6 months after stroke onset, indicating excellent functional recovery; Group 2 (moderate trajectory, 31.5%) contained 69 patients with mRS scores began at close to 4 during acute hospitalization and gradually decreased to around 3 during 3-month follow-up and then maintained this level until 6-month follow-up, indicating poor functional recovery; Group 3 (severe trajectory, 26.9%) contained 59 patients with mRS scores began at around 4 but then continued to increase rapidly until the end of the follow-up, indicating worsening functional recovery. Longitudinal changes in functional status after stroke onset across different mRS trajectory groups are shown in Supplementary Figure S1.

Table 2 presents the sociodemographic and clinical characteristics of patients in three mRS trajectory groups. Patients in the severe trajectory were more likely to be older and unmarried, had a greater proportion of posterior circulation infarct (POCI), and higher admission NIHSS score, whereas patients in the mild trajectory were more likely to be women, had a greater proportion of lacunar circulation infarcts (LACI) and small artery occlusion (SAO), and a lower burden of hypertension, diabetes, atrial fibrillation and stroke rehabilitation ( p < 0.05).

Five distinct BI trajectory groups were similarly identified by our model based on the criteria and clinical interpretability (Figure 2B; Supplementary Table S5): Group1(low-rapid rise trajectory, 9.1%) included 20 patients whose BI scores were the lowest in the acute stage then rose rapidly over time, representing a gradual improvement in ADL; Group 2 (moderate low-stable trajectory, 10.0%) included 22 patients with BI scores started below 40 then stabilized between 50 and 55; Group 3 (moderate-progressive rise trajectory, 28.8%) included 63 individuals and followed a similar pattern to the group 2, with BI scores fluctuating between 50 and 90, and these two groups exhibited mild to moderate dependence in ADL; Group 4 (moderate high-rapid decline trajectory, 8.3%) included 18 patients referring to those who initially had BI levels similar to the group 3 but continued to decline acutely over the following time, representing a progressive deterioration in ADL; Group 5 (high-stable trajectory, 43.8%) included 96 patients with persistently high BI levels during the first 6 months after onset, representing the functional independence. The functional changes following acute stroke between different BI trajectory groups are shown in Supplementary Figure S2.

The sociodemographic and clinical profiles of patients by BI trajectories are shown in Table 3. Patients in the high-stable trajectory tended to have lower NIHSS scores on admission, a higher proportion of SAO, and a lower rate of stroke rehabilitation, whereas those in the low-rapid rise trajectory were more likely to carry a greater burden of hypertension and diabetes ( p < 0.05).

Tables 4, 5 present the association between early depressive symptoms and trajectories of functional recovery (mRS, BI) during the 6-month follow-up period after onset of stroke in multivariable logistic regression models with adjustment for relevant covariates, respectively.

For mRS trajectories (Table 4), the unadjusted model (Model I) revealed that compared with the mild trajectory group, early depressive symptoms increased odds of patients belonging to the moderate (OR 12.54, 95% CI 5.88–26.77) and severe (OR 53.25, 95% CI 14.96–189.55) trajectory group. After adjusting for sociodemographic (age, sex, and marital status) and stroke-related factors (NIHSS at admission, stroke rehabilitation, OCSP classification, hypertension, diabetes, atrial fibrillation, and smoking status), the fully adjusted model (Model III) confirmed the results. Additionally, stroke severity at baseline independently predicted poor recovery trajectories, either in unadjusted or adjusted analyses ( p < 0.001).

For BI trajectories (Table 5), the unadjusted model (Model I) demonstrated that early depressive symptoms were associated with an increased likelihood of patients being in the low-rapid rise (OR 38.0, 95% CI 4.87–296.71), moderate low-stable (OR 12.67, 95% CI 3.49–45.99), moderate-progressive rise (OR 2.85, 95% CI 1.48–5.49), and moderate high-rapid decline (OR 34.0, 95% CI 4.33–267.02) trajectory groups, compared to the high-stable trajectory group. After adjusting for all covariates (Model III), the association between early depressive symptoms and the first three trajectory groups was no longer significant, but remained significant for the moderate high-rapid decline trajectory group (OR 12.93, 95% CI 1.49–112.42). The stroke severity at baseline was also significantly related to the BI trajectory groups throughout models ( p < 0.001 ).

We calculated the E-values to assess the robustness of observed associations against potential unmeasured confounder factors, as shown in Supplementary Table S6. The E-values for the associations between early depressive symptoms and mRS trajectories were 5.18 (moderate vs. mild) and 9.35 (severe vs. mild). Similarly, for the relationship between early depressive symptoms and BI trajectories, the E-value for the OR was 6.65 in the comparison between moderate high-rapid decline and high-stable trajectory group. This indicates that the observed associations could potentially be explained by an unmeasured confounder factor associated with both the exposure and the outcome, with a relative risk greater than the reported E-values, conditional on the measured covariates. The refitted model identified three distinct BI recovery trajectories (Supplementary Figure S3): a “moderate-low functional recovery” (Class 1, 15.4%, BI slowly improved from ≈20 to ≈50), a “high-stability” (Class 2, 59.2%, BI ≈80 maintained from T1, set as reference), and a “gradual improvement”(Class 3, 25.4%, BI increased from ≈40 at T1 to ≈80 at T3). In the adjusted multinomial logistic regression (Supplementary Table S7), early depressive symptoms remained significantly associated with the less favorable “moderate-low functional recovery” trajectory (OR = 3.62, 95% CI: 1.36–9.67, p = 0.010) but not with the “gradual improvement” trajectory (OR = 4.70, 95% CI: 0.90–24.69, p = 0.067). This confirms that the core finding—linking early depressive symptoms specifically to unfavorable functional recovery trajectories—is robust and not substantially altered by the exclusion of the smallest subgroup.