Stroke Severity and Sex Shape Hidden Cognitive Impairment Profiles, Study Finds

Cognitive impairment after cerebrovascular disease has long been treated as a single, linear problem: screen the patient, apply a cut-off score, and move on. A new cross-sectional study from West China Hospital of Sichuan University, published in Frontiers in Neurology, argues that this one-size-fits-all view misses something fundamental. Using latent profile analysis (LPA), a person-centered statistical technique, the research team identified three distinct cognitive impairment profiles among 474 patients with cerebrovascular diseases (CVDs), and found that stroke severity measured by the National Institutes of Health Stroke Scale (NIHSS) and patient sex were the factors most strongly associated with which profile a patient fell into.

The stakes are considerable. Cerebrovascular diseases remain a leading cause of death and loss of healthy life in China, with a reported crude mortality rate of 149.6 per 100,000 individuals. Beyond the physical damage, cognitive impairment has emerged as a pervasive complication spanning a spectrum from mild deficits to dementia. Prevalence rates reported across CVD subtypes are striking: roughly 66 percent in stroke and about 54.6 percent in moyamoya disease. Evidence even suggests that among people adhering to optimal lifestyle behaviors, the presence of cognitive impairment significantly elevates the risk of all-cause and cerebrovascular mortality. Yet how impairment manifests across multiple cognitive domains in individual patients has remained poorly understood, often resulting in generalized care plans that fail to meet specific rehabilitation needs.

Traditionally, clinical and nursing research has followed a variable-centered path, relying on global screening instruments such as the Mini-Mental State Examination (MMSE) or the Montreal Cognitive Assessment (MoCA) to categorize patients via dichotomous cut-off scores. While effective for identifying universal risk factors like age, comorbidities, or lesion sites, this approach assumes population homogeneity. It cannot simultaneously analyze individual performance across multiple cognitive domains such as memory, orientation, and language ability. LPA offers a sophisticated alternative by identifying latent subgroups within a population based on response patterns across multiple continuous indicators. Recent studies applying LPA to older adults, stroke survivors, and people with cardiometabolic diseases have consistently identified three cognition profiles: a high-performing group, a severely impaired group, and an intermediate group characterized by domain-specific deficits whose distinguishing features varied across disease populations.

The new study recruited 474 patients from the neurosurgery department of a tertiary referral hospital in southwestern China between November 2023 and September 2025, using convenience sampling and reporting according to the STROBE checklist. Participants were aged 18 or older, diagnosed with CVDs via ICD-10 codes (I60-I68) and confirmed by cranial CT or MRI. Patients with pre-existing dementia attributable to other conditions, severe auditory impairment, psychiatric illness, communication disorders, acute critical medical conditions, or concurrent enrollment in other trials were excluded. The cohort had a median age of 52 years, a near-equal gender split, and was clinically dominated by intracerebral hemorrhage (52.3 percent), moyamoya disease (18.8 percent), and intracranial aneurysms (15.0 percent), with ischemic stroke accounting for only 4.6 percent. The majority underwent surgery (79.7 percent) and received intensive supportive management including analgesia, sedation, and mechanical ventilation.

The methodological core of the study lay in how it prepared the cognitive data. The Chinese version of the MMSE, a 30-item instrument assessing five dimensions of orientation, memory, attention and calculation, recall, and language ability, was administered to all participants. The cohort’s mean total MMSE score was 22.92, indicating overall mild cognitive impairment, with the most significant deficits observed in recall (mean 1.66 out of 3) and orientation (mean 7.72 out of 10), while memory registration and language ability were relatively better preserved. Crucially, the five MMSE domain scores were separately adjusted for age, gender, and years of education using ordinary least-squares linear regression, producing demographically adjusted residual z-scores. These adjusted z-scores, rather than raw scores, served as the continuous indicators in the LPA, ensuring that profiles reflected genuine cognitive differences rather than demographic artifacts.

The analysis, conducted in R using the tidyLPA package, estimated one- through five-profile solutions. Fit statistics told a nuanced story: AIC, BIC, and adjusted BIC decreased as profiles were added, and the Lo-Mendell-Rubin and bootstrapped likelihood ratio tests were significant for the two- through five-profile solutions, meaning the four- and five-profile models showed better statistical fit than the three-profile model. However, the three-profile model showed greater classification certainty, with an entropy of 0.955 compared with 0.859 and 0.815 for the four- and five-profile models. The team retained the three-profile solution as a pragmatic choice based on parsimony and clinical interpretability, and examined local-solution sensitivity using 1,000 randomized hierarchical initializations for each model.

The retained model classified participants into a mild impairment group (82.5 percent, n=391), which performed relatively well across all five cognitive domains; a moderate impairment group (11.4 percent, n=54), which showed lower performance across all domains with the greatest deficits in orientation and attention/calculation while memory was relatively preserved; and a severe impairment group (6.1 percent, n=29), characterized by pervasive deficits across all domains. The three profiles reflected increasing severity of global cognitive impairment. Univariate analyses revealed significant inter-profile differences in gender, educational background, alcohol consumption, disease subtypes, follow-up interval, nutritional risk (NRS-2002), NIHSS score, and length of hospital stay. The severe group had a median MMSE score of just 3, compared with 15 in the moderate group and 27 in the mild group.

Multinomial logistic regression, using the mild impairment group as the reference category, identified two factors independently associated with profile membership. Higher NIHSS scores were associated with higher odds of membership in the moderate impairment profile (OR=1.222, 95 percent CI=1.126-1.326, p<0.001) and in the severe impairment profile (OR=1.397, 95 percent CI=1.254-1.557, p<0.001). Male gender was associated with lower odds of membership in the severe impairment profile (OR=0.218, 95 percent CI=0.067-0.704, p=0.011). The NIHSS association aligns with the scale’s established prognostic utility: a higher score typically correlates with a larger infarct or hematoma volume and a lesion in a more critical location, such as the dominant hemisphere or key cognitive networks, depleting the brain’s cognitive reserve. Population-based evidence from the Oxford Vascular Study suggests severe stroke can bring forward dementia by about 25 years.

The gender finding is more exploratory but mechanistically intriguing. The authors note that females typically demonstrate higher baseline cerebral blood flow than males, potentially rendering them more sensitive to hemodynamic fluctuations following cerebrovascular events. While sex hormones confer neuroprotection in younger physiology, preclinical evidence suggests these effects may become ineffective or even deleterious with advancing age, and 64 percent of the female participants in this study were aged over 50. Sociodemographic factors may also play a role, as women in the cohort had significantly lower educational attainment than men. The authors caution that further studies are needed to assess the biological or social mechanisms underlying this association.

Important limitations temper the conclusions. The cohort’s predominance of hemorrhagic and neurosurgical cases, with ischemic stroke accounting for only 4.6 percent, introduces selection bias that may limit generalizability to broader CVD populations. The cross-sectional design precludes causal inference and cannot characterize individual cognitive trajectories over time. The severe impairment subgroup was small, with an events-per-parameter ratio of 2.4, limiting statistical precision and producing wide confidence intervals. The MMSE itself is a brief screening tool with limited coverage of executive function, processing speed, visuospatial ability, and episodic memory, and demographic adjustment cannot recover information lost to these measurement limitations. The authors call for multicenter, longitudinal studies with larger, more diverse samples to validate the profiles and assess their prognostic utility. Still, the emergence of distinct profiles underscores the multidimensional nature of cognitive impairment captured by the MMSE, challenging the notion of a unitary trajectory of decline and pointing toward targeted nursing assessment and intervention for specific patient subgroups.

Subject of Research: Neurology

Article Title: Cognitive impairment profiles in patients with cerebrovascular diseases: a latent profile analysis

Article References: Yin, Y., Ma, H., Zou, R., Fan, C., Zhu, W., & Zheng, J. (2026). Cognitive impairment profiles in patients with cerebrovascular diseases: a latent profile analysis. Frontiers in Neurology, 17, Article 1840409. https://doi.org/10.3389/fneur.2026.1840409

Image Credits: AI Generated

DOI: 10.3389/fneur.2026.1840409

Keywords: cognitive impairment, cerebrovascular disease, latent profile analysis, stroke, NIHSS, MMSE, neurology, vascular cognitive impairment, Cognitive, impairment, profiles, patients