This study used data from the Korean Stroke Cohort for Function and Rehabilitation (KOSCO). The KOSCO study is a 10-year, long-term follow-up study of stroke patients using a prospective multicenter design to investigate residual disability, activity limitation, and quality of life in patients who experienced a first-ever stroke. Patients with ischemic and hemorrhagic stroke were included, while patients with transient ischemic attack were excluded. The detailed rationale and protocol of KOSCO have been previously described [8].

The KOSCO data from three participating hospitals, collected from August 2012 to May 2015, were reviewed. A total of 1,287 patients with a left hemispheric lesion were selected. Of the 1,287 patients, 624 (48.5%) with preserved language function (Korean version of the Frenchay Aphasia Screening Test [K-FAST] score >25) and 274 (21.3%) with a coinciding right hemispheric stroke lesion were excluded. Another 125 (10.1%) patients were excluded owing to missing follow-up assessments, and 29 (2.3%) patients were excluded owing to missing magnetic resonance imaging (MRI) data. Altogether, a total of 235 first-time stroke patients were included in the final analysis (Fig. 1).

Data on demographic and clinical characteristics including age, sex, obesity, hypertension, hyperlipidemia, diabetes mellitus, atrial fibrillation, family history of stroke, education level, the National Institutes of Health Stroke Scale (NIHSS) score at admission, smoking status, and alcohol use were collected from the KOSCO study records. Included comorbidities were recorded according to the following definitions: hypertension (systolic blood pressure >160 mmHg, diastolic blood pressure >90 mmHg, or history of hypertension or medical treatment), diabetes mellitus (elevated blood glucose level >126 mg/ day or history of diabetes or medical treatment), atrial fibrillation (documented by standard electrocardiogram [ECG], long-term ECG, or history of atrial fibrillation or medical treatment), hyperlipidemia (low-density lipoprotein >160 mg/dL, elevated total cholesterol level >240 mg/dL, or history of hyperlipidemia or medical treatment), and obesity (body mass index [BMI] ≥25). Education level was classified based on the highest diploma that the patient acquired as follows: no formal education, primary education (primary school diploma, 1–6 years of education), secondary education (middle or high school diploma, 7–12 years of education), and higher education (university or graduate school diploma, more than 13 years of education). The NIHSS scores were grouped into three categories: mild (NIHSS score 0–5), moderate (NIHSS score 6–13), and severe (NIHSS score ≥14) [9].

Stroke lesion volume and involvement of languagerelated cortical lesions were assessed. To determine language-related cortex involvement, Broca’s area and Wernicke’s area (posterior third of the STG) were selected as regions of interest (ROIs) (Fig. 2). The MRI scans performed at the time of admission were reviewed. Each lesion was manually drawn on diffusion-weighted imaging with lesion mapping software (MRIcro software version 1.4; Chris Rorden’s Neuropsychology Lab, Columbia, SC, USA). The lesion volume of each patient was extracted using a binarized lesion mask. The lesion area was assigned a value of 1 in the binarized lesion mask with lesion mapping software. The lesion volume was calculated by multiplying the number of voxels assigned a value of 1 in the binarized lesion mask by the voxel volume and converted into cm³ . Before the results were analyzed, brain MRI scans from 30 randomly selected patients were used to verify lesion volume reproducibility. ROI involvement and lesion volume were assessed by two blind readers, and inter-rater reliability was evaluated prior to statistical analysis. Interclass correlation tests (ICC) for lesion volume, assessed by two blind readers, showed excellent reliability (p>0.9). Interclass correlation tests for the involvement of Broca’s area and Wernicke’s area (posterior third of the STG) were 0.882 and 0.845, respectively.

Language function was evaluated using the K-FAST, which includes assessment of comprehension, verbal expression, reading, and writing. Patients were assessed with K-FAST at 7 days, 3 months, 6 months, and 1 year after stroke onset. Furthermore, to evaluate changes in language function according to the severity of the initial aphasia, participants were categorized into three groups according to the initial K-FAST score: mild (K-FAST score 20–25), moderate (K-FAST score 11–19), and severe (K-FAST score 0–10) [10]. In this study, we defined the primary outcome of aphasia as delta K-FAST score: the difference between the initial K-FAST and the 1-year K-FAST scores.

For statistical analysis, we used descriptive statistics for demographic and clinical characteristics. Nominal and ordinal data obtained from a baseline review of medical records and initial stroke features were assessed using frequency analysis. Scale factors were analyzed using average analysis. Differences in demographic and clinical data among severity groups were analyzed using Pearson χ² or Fisher exact test for categorical variables and the Kruskal-Wallis rank test for numerical variables that did not meet normality assumptions. Inter-rater reliability in measuring lesion volume and determining ROI involvement was calculated with ICC. Repeated measures ANOVA was conducted to assess the difference in language function at 7 days, 3 months, 6 months, and 1 year in the total study population, and in subgroups. To correct for baseline differences in variables among the groups, age, initial K-FAST score, total stroke lesion volume, obesity level, education level, initial NIHSS score, and languagerelated cortical region involvement were included as covariates when conducting ANOVA. Bonferroni correction was used in the post-hoc analysis of ANOVA. McNemar test was conducted to evaluate differences in patient distribution in the severity groups after 1 year. To determine related factors for language function recovery in terms of delta K-FAST, linear regression was initially conducted, and multiple regression analysis was performed for significant values. Additionally, variance inflation factor (VIF) values were investigated for the multicollinearity of independent variables, and the cutoff value was set to less than 5. The p-values less than 0.05 were considered statistically significant. All analyses were conducted using SPSS version 24.0 (IBM, Armonk, NY, USA).

Among the groups, there were significant differences in the age at onset, stroke lesion volume, initial K-FAST score, level of obesity, education level, initial NIHSS severity, and language-related cortical region involvement. Full patient demographic and clinical characteristics are shown in Table 1.

The mean values of K-FAST scores assessed at 7 days, 3 months, 6 months, and 1 year after onset were 15.5±7.6, 19.8±8.1, 21.1±7.8, and 22.0±7.6, respectively, and differed significantly among time points (p<0.05) (Fig. 3A). However, sub-group analysis revealed that only the severe group had significant differences in mean K-FAST scores among time points (p<0.05) (Fig. 3B). Mean K-FAST scores among groups were significantly different at all assessment time points, using an independent samples t-test with post-hoc Bonferroni correction (p<0.01). Improvement of language function was also observed by the shift of patient distribution in the K-FAST severity groups after 1 year. At the initial assessment, 39.0% of cases were categorized as mild; 34.3%, as moderate; and 26.7%, as severe. After 1 year, only 8.5% of cases were categorized as severe and 40.3% were found to have normal language function (K-FAST score >25) (Fig. 3C). All groups had significant changes in distribution at 1 year (p<0.05).

Linear regression indicated the age at onset, obesity level, education level, initial NIHSS score, initial K-FAST score, total lesion volume, and language-related cortical region involvement as factors affecting delta K-FAST. Among the above-mentioned factors, when assessed with multivariate analysis, age at onset, education level, initial NIHSS score, initial K-FAST score, and total lesion volume area were identified as factors influencing the recovery of language function (Table 2). Among the variables included for multivariate analysis, there were no variables with multicollinearity (VIF<5). Age at onset, initial NIHSS score, and initial K-FAST score were shown to negatively affect language function recovery, while education level and stroke lesion volume positively influenced language function improvement during the 1 year after stroke. For the final multivariate regression model, adjusted R² value was 0.395, with a p-value <0.001 for Fstatistics.