The study was conducted in Division I, Department of Neurology, Dongguan People's Hospital (Dongguan, China) between September 1, 2018, and July 31, 2019. The inclusion criteria for the study were (1) age >18 years, (2) first or recurrent early subacute ischemic stroke occurring within 7 days before admission (the ischemic stroke diagnosis was made in accordance the American Heart Association Stroke Council criteria) (18), and (3) that a complete brain MRI examination had been performed. The exclusion criteria were (1) transient ischemic attack, cerebral hemorrhage, epidural or subdural hematoma, or subarachnoid hemorrhage; (2) prestroke cognitive impairment, which was assessed using the informant questionnaire on cognitive decline in the elderly (IQCODE) in hospital (19); (3) coexisting conditions (beside stroke) that affect cognitive function (e.g., major depression or anxiety, chronic alcoholism, previously diagnosed psychiatric conditions, epilepsy, severe traumatic brain injury, Parkinson's disease, multiple sclerosis); (4) lack of complete clinical or MRI data; (5) death during hospitalization; (6) severe stroke indicated by a National Institutes of Health stroke scale (NIHSS) score ≥15 points at discharge; (7) patients with severe comorbidities (e.g., liver, kidney, heart, or respiratory failure or malignant tumors); (8) severe hearing disabilities, sight disabilities, or language disorders at discharge according to physical examination; and (9) refusal by patients or their relatives to provide informed consent.

Demographic and clinical variables (i.e., age, sex, years of education, onset time, vascular risk factors, history of stroke, and neurological deficit status assessed using the NIHSS) were collected. Preexisting cognitive status was evaluated using the IQCODE within 3 days after admission. The IQCODE questionnaire consists of 16 items that assess impairment in the patient over the previous 10 years. The IQCODE score ranges from 16 (marked improvement in all items) to 80 (marked worsening in all items). A total score >51 points is considered to indicate a preexisting cognitive impairment (19).

Brain MRI scanning, including T1-, T2-, diffusion-, and susceptibility-weighted imaging sequences were performed on each participant using a 3.0-T system (Sonata; Siemens Medical, Erlangen, Germany) within 7 days after admission. Diffusion-weighted image spin-echo echo-planar imaging (repetition time [TR]/echo time [TE]/excitation: 2,162/76/1; matrix: 128×128; field of view (FOV): 230 mm; slice thickness/gap: 6 mm/1 mm; echo-planar imaging factor: 47; acquisition time: 25.9 s) with three orthogonally applied gradients was used with a b value of 0 and 1,000. Axial spin echo T1 (TR/TE/excitation: 488/15/1; FOV: 230 mm, slice thickness/gap: 6 mm/1 mm; matrix: 256 × 256; time of acquisition: 1 min 24.8 s) and turbo spin echo T2 (TR/TE/excitation: 3,992/110/2; turbo factor: 15; FOV: 230 mm; slice thickness/gap: 6 mm/1 mm; matrix: 512 × 512, time of acquisition: 1 min 55.8 s) images were also captured.

A neurologist (JFQ) measured the MRI variables as follows:

(1) Assessment of lesions involving the CCP. Lesions involving the CCP were assessed using CHIPS (16), which was developed based on immunohistochemical tracings of the cholinergic pathways in humans, superimposed onto a structural MRI image (11). Measurement zones within four index slices of white matter regions were demarcated on axial T2-weighted images of the lateral ventricles and the third ventricle. Slices: (A) low external capsule layer (four separate zones: bilateral anterior and posterior zones), (B) high external capsule layer (six separate zones: bilateral anterior, posterior, and cingulate gyrus zones), (C) corona radiata (six separate zones: bilateral anterior, posterior, and cingulate gyrus zones), and (D) semioval center (four separate zones: bilateral anterior and posterior zones). The severity of white matter hyperintensities (WMHs) in each zone was scored as (1) normal (0 points), (2) <50% of the region involved (1 point), and (3) ≥50% of the region involved (2 points). Severity scores were weighted according to the distribution density of cholinergic fibers. The weighting coefficients were as follows: 4 for slice A, 3 for slice B, 2 for slice C, and 1 for slice D. The score for each region was obtained by adding all points from the measurements of small zones and multiplying them by the weighting coefficient. The maximum score per hemisphere was 50, and the total maximum score per scan when combining scores for all four regions was 100. The CHIPS total score showed a good correlation with lesion volume within the cholinergic pathways (Spearman's = 0.87, p < 0.0001). Higher CHIPS scores suggest more severe CCP impairment (16). Lesions involving the CCP included acute infarction, old infarction, or WMHs (Figure 1) (20).

(2) Acute infarcts. Acute infarcts were defined as regions of restricted water diffusion identified on diffusion-weighted imaging with b values of 1,000 together with hypointensity on the corresponding apparent diffusion coefficient map. The sites of acute infarcts were divided into cortical regions (frontal, temporal, parietal, and occipital lobes), subcortical regions (white matter, basal ganglia, and thalamus), and infratentorial regions (i.e., brainstem and cerebellum). The contours of an acute infarct on diffusion-weighted imaging were manually outlined. The total volume was subsequently calculated by multiplying the total area by the sum of the slice thickness and the gap.

(3) White matter lesions (WMLs). The severity of WMLs, including deep white matter hyperintensities (DWMH) and periventricular hyperintensities (PVH), were scored using fluid-attenuated inversion recovery images with the Fazekas scale (range 0–3) (21).

(4) Medial temporal lobe atrophy (MTLA). MTLA was assessed using the Scheltens' scale (22) in which coronal MRI sections are used to judge the severity of MTLA based on standard images (range from 0 to 4, no atrophy to severe atrophy).

(5) Cerebral microbleeds. These were defined as small (2–10 mm) hypointense lesions with a clear margin in susceptibility-weighted imaging. Symmetric basal ganglia calcification and flow void artifacts of the pial blood vessels were excluded.

Neuropsychological assessments were performed at 3 months through a face-to-face interview with patients after the index stroke by a trained neurologist (HHZ), who was blinded to all clinical data. We did not assess cognitive impairment in the acute phase because, in acute stroke, assessment of cognitive function is not reliable because of cognitive status being unstable, and the severity of disease can preclude lengthy neuropsychological testing (23, 24). We did not assessed the cognitive function used by the Montréal Cognitive Assessment because it is commonly used as a screening tool rather than a diagnostic criteria (25, 26). We used a more comprehensive neuropsychological battery that was a modified version of the vascular dementia battery (VDB) (27, 28), which comprises the following seven cognitive domains: (1) executive functions [frontal assessment battery (29, 30); trail-making test, parts A and B]; (2) attention (digit span: forward and backward) (31), visual memory span (forward and backward) (32) and the auditory detection test (28); (3) language (modified Boston naming test) (33) and verbal fluency (34) (animal and food categories); (4) verbal memory (immediate and delayed word list recall, delayed word list recognition, and immediate and delayed story recall) (35). For this, the tester needs to clearly read 12 related vocabularies or a story to the testee, learn it three times, and then the tester needs to recall the vocabulary or story in real time and then recall the vocabulary or story they just learned 20 min later. During the delayed word list recognition, according to the content just learned, the testee needs to recognize the order of the vocabulary. Next is (5) visual memory (immediate picture recall, delayed picture recall, and delayed picture recognition) (32), visual reproduction I and II and delayed recognition from the Wechsler memory scale III (WMS-III) (32); (6) visuoconstruction [clock drawing test (35), Wechsler adult intelligence scale III [WAIS-III] block design (31) and visual reproduction (copy) from WMS-III (32)]; (7) visuomotor speed [symbol digit modalities test (36), digit cancellation test, and the maze task (28)]. The testee can see the content of the test. In the symbol digit modalities test, he or she needs to simulate different symbols corresponding to different numbers that have been specified, fill in the corresponding numbers in the space below the symbols, and record the number filled in by the tester in 120 s. In the digit cancellation test, the subject needs to cross out the number to be canceled as much as possible within 90 s. The maze task is to calculate the time it takes for the testee to successfully find the exit route of the maze. Apart from the executive domain, impairment of individual cognitive domains was identified if at least 50% of the subtests were below the cutoff points. The cutoff points were derived from a Chinese sample (33) (Supplementary Table 1). The criteria for impairment in any of the seven cognitive components were adjusted for the education level of each patient (28). PSCI refers to a series of syndromes meeting the diagnostic criteria of cognitive impairment within 6 months after an index stroke according to the Diagnostic and Statistical Manual of Mental Disorders (Fifth Edition) criteria (37). We defined PSCI as impairment in at least one cognitive domain; PSCI-ND was defined as impairment in at least one cognitive domain but with normal daily abilities, which were assessed using the Lawton activities of daily living (ADL) scale (38). Cases with simultaneous impairment in at least two cognitive domains and IADL were considered to have PSD. Additionally, we also assessed the anxiety and depression statuses of patients, which were measured using the Hamilton anxiety rating scale (HAMA) and the Hamilton depression rating scale (HAMD), respectively.

The functional dependence assessment at 3 months used the ADL scale. The ADL, which includes the instrumental ADL (IADL) and basic ADL (BADL) scales, examines the current functional level of a patient and can identify improvement or deterioration over time. The eight domains of function measured with the IADL scale include the ability to use a telephone, shopping, food preparation, housekeeping, laundry, mode of transportation, responsibility for own medications, and ability to handle finances. The total IADL score is calculated by summing up the points obtained for each item. The maximum IADL score is 32. The BADL scale features six additional questions that measure different levels of toilet activity, feeding, dressing, grooming, and physical activities. The total ADL score is calculated by summing the IADL and BADL scores together. The lowest possible ADL score is 14 points, which indicates that the patient's abilities are completely normal. If the score is >14 points, there is a reduced function, and the highest possible score is 56 points. The severity of disability was measured with the modified Rankin scale (mRS). We defined a poor disability outcome as mRS > 2 points. Death and recurrence of stroke during the follow-up period were also recorded.

All statistical analyses were conducted using the SPSS Version 24.0 (IBM Corp., Armonk, NY, USA) software. Descriptive data are presented as proportions, means, or medians as appropriate. A univariate analysis was performed to compare putative risk factors between patients with and without PSCI. In the logistic regression analysis, we used a backward elimination procedure. The PSCI served as a dependent variable. Subsequently, risk factors with a value of P < 0.05 were analyzed with a multivariate logistic regression analysis using a backward stepwise selection strategy. Correlation analyses were conducted to test collinearity between candidate independent variables. If the correlation coefficient between any of these putative risk factors was ≥0.50, then variables with a smaller P-value were entered into the logistic regression. As age and prestroke cognitive status were important risk factors for PSCI, we set up two models, respectively. Model 1 included age, and model 2 included IQCODE score with the rest of the putative factors held constant. The odds ratio of any independent risk factor was interpreted as the risk of non-remission of PSCI, when all other risk factors were held constant. For the analysis of the association between cognitive status and mRS, we constructed an ordinal regression model (dependent variable: mRS grading), adjusted for age education level, IQCODE, NIHSS score at admission, and PVH. The level of statistical significance was set at P < 0.05 (two-sided).

The univariate analysis reveals that patients with PSCI were older, had higher IQCODE scores, had more frequent previous stroke history and atrial fibrillation, had higher CHIPS scores, and had more severe WMLs and MTLA. Additionally, patients with PSCI had higher HAMD scores at 3 months (Table 1). Furthermore, PSCI patients had severer impairment in each zone of the CCP in comparison to those with non-PSCI (Supplementary Table 2).

The independent variables with values of P < 0.05 in the univariate analysis results (Table 1) were subjected to Spearman's correlation analysis to check the colinearity, including age, IQCODE score, ischemic stroke history, atrial fibrillation history, infratentorial acute infarction, total CHIPS score, PVH, DWMH, MTLA, and HAMD. Age was significantly correlated with IQCODE score, PVH score, and MTLA score (correlation coefficient r = 0.566, 0.509, and 0.549, respectively). Additionally, PVH score was significantly correlated with DWMH score, CHIPS score, and MTLA score (r = 0.675, 0.586, and 0.557, respectively), and CHIPS score was also significantly correlated with DWMH score and MTLA score (r = 0.713 and 0.529, respectively). There were no significant correlations between any other independent variables (the correlation coefficients were all <0.5). Because both age and IQCODE score are important factors affecting PSCI, we set up two models to analyze them. Model 1 included age, and model 2 included IQCODE score with the rest of the factors held constant; history of ischemic stroke, history of atrial fibrillation, acute infratentorial infarction, CHIPS score, and HAMD score served as independent variables.

The logistic regression of model 1 revealed that age, CHIPS score, and HAMD score were independent risk factors for PSCI at 3 months, whereas model 2 revealed that IQCODE, CHIPS, and HAMD scores were independent risk factors for PSCI at 3 months (Table 2). Compared with model 2, the predictive probability of model 1 was greater (80.6 vs. 74.8%), and the model fit was higher (Cox and Snell R² = 0.377 vs. 0.309; Nagelkerke R² = 0.504 vs. 0.413).

As the history of stroke was an important risk factor for PSCI, we conducted another regression model in patients without stroke history (Supplementary Table 3). The analysis revealed that, despite adjusting for the previous stroke, higher CHIPS scores were also associated with PSCI at 3 months.

Ordinal regression analysis revealed that more severe cognitive dysfunction was significantly associated with higher mRS, adjusted for age, education level, IQCODE, NIHSS score at admission, and PVH (Table 3).