Post-stroke cognitive decline among women in midlife

Abstract

Background:

Stroke is a significant risk factor for cognitive impairment. Women face a heightened risk due to their longer life expectancy and the greater prevalence of stroke-related disability.

Objective:

To compare the trajectory of cognitive decline in women with and without a stroke diagnosis and assessed demographic differences.

Design:

The study employed a longitudinal, observational cohort design.

Methods:

The Study of Women’s Health Across the Nation interviewed women in their 40s and 50s from seven U.S. cities approximately yearly to collect information on their physical, biological, psychological, and social health. Cognitive assessments were conducted between 2000 (Wave 4) and 2008 (Wave 10) that were designed to examine information processing speed, working memory, immediate memory, and delayed memory. Adjusting for baseline age, race/ethnicity, income, education, marital status, comorbidities, and insurance, generalized linear mixed models were used to compare cognitive decline between women who had experienced a stroke and those who had not.

Results:

Among the 3302 women in the sample, age was 49.52 (SD = 2.64) years old at baseline and 53.74 (SD = 3.94) in the final wave. In Wave 4, only 0.67% (N = 22) reported having been diagnosed with a stroke, but 8.39% (N = 277) had been diagnosed by Wave 10. Compared to stroke-free women, the multivariable-adjusted changes in cognitive performance were −0.18 information processing, −0.09 working memory, −0.15 immediate memory, and −0.17 delayed memory. Compared to White women, Black women who suffered a stroke saw multivariable-adjusted annual changes in information processing, working memory, immediate memory, and delayed memory of −0.08, −0.03, −0.08, and −0.09, respectively.

Conclusion:

These findings underscore the persistent impact of stroke on multiple cognitive domains in midlife women, with a more significant decline observed among Black women. Targeted prevention and rehabilitation efforts are needed to address both the cognitive consequences of stroke and the demographic differences in post-stroke outcomes.

Introduction

Cognitive decline is a natural part of the aging process that affects various cognitive domains, including memory, attention, processing speed, and executive function. ¹ Research indicates that cognitive aging can begin in midlife, ² with notable declines in specific cognitive domains such as processing speed and memory. ³ As the brain ages, various structural and functional changes contribute to this decline. One of the most significant changes is the reduced size of the hippocampus, a region crucial for memory formation. ⁴ This shrinkage and a general loss of neurons and synapses lead to decreased brain volume and impaired cognitive function. ⁵ In addition to these structural changes, reduced blood flow to the brain and the accumulation of abnormal protein clumps, such as beta-amyloid plaques and tau tangles, play critical roles in cognitive decline. ⁶ These clumps disrupt normal brain function, leading to memory loss and other cognitive impairments.

Age-related cognitive decline is a multifaceted and highly variable process influenced by many factors that all have a differential impact on individuals experiencing the decline. ⁷ This variability in cognitive decline can be observed across different cognitive domains, such as memory, attention, processing speed, and executive function. ⁸ One of the primary reasons for this variation is the heterogeneity in brain aging. ⁹ Some individuals experience significant declines in cognitive function, while others maintain relatively stable cognitive abilities well into old age. ¹⁰

The current literature suggests that genetic, environmental, and lifestyle factors influence the decline. ¹¹ Genetic predisposition can accelerate the onset and progression of cognitive impairments. ¹² Similarly, lifestyle choices such as activity level, diet, smoking, and alcohol consumption can either accelerate or delay cognitive decline and/or the onset of cognitive impairments. ¹³ In addition, chronic health conditions, such as diabetes, hypertension, and cardiovascular disease (CVD), can also negatively impact brain health. ¹⁴ Finally, mental health issues like depression, anxiety, and chronic stress can contribute to cognitive decline by affecting brain function and overall cognitive health. ¹⁵ Effective management of these conditions is crucial for reducing the risk of cognitive decline.

Research exploring cognitive decline also suggests that some population groups experience accelerated cognitive aging due to associated chronic health-related conditions such as stroke.^{16 –19} Stroke, a leading cause of disability, is also a significant risk factor for cognitive impairment. ¹⁷ Post-stroke cognitive decline occurs in a substantial proportion of stroke survivors. ²⁰ Studies show that women experience more significant cognitive decline after stroke when compared to men, with some differences attributed to pre-stroke characteristics such as widowhood.^21,22 A least one study showed that although differences between men and women did not exist in post-stroke cognitive decline; however, the affected domains differed between men and women. ²³

There is evidence that indicates that stroke survivors often experience declines in attention, working memory, and executive function, even after accounting for pre-stroke cognitive function and vascular risk factors. ²⁴ However, less is known about the association between stroke and cognitive decline despite reports that stroke survivors often experience declines in attention, working memory, and executive function, even after accounting for pre-stroke cognitive function and vascular risk factors. ²⁴ Therefore, it remains unclear whether stroke accelerates age-related cognitive decline beyond what is expected in normal aging, particularly in women transitioning through midlife. This issue is of interest because previous studies of cognitive aging (in the absence of stroke) suggest women experience cognitive decline associated with menopause and midlife. ²⁵ Study of Women’s Health Across the Nation (SWAN) offers a unique opportunity to explore the relationship between stroke and cognitive decline in midlife women by allowing for a nuanced examination of post-stroke cognitive trajectories and how those trajectories differ among women from diverse backgrounds. ²⁶

The objective of this study was to explore the trajectory of cognitive decline in women with and without a history of stroke. This study utilized panel date from SWAN to assess the longitudinal association between cognitive decline and stroke among women. To complete the study, we employed longitudinal mixed models to account for the inter-respondent correlation of repeated individual-level measurements over time, allows for different individual levels at baseline, and controls for time-invariant individual characteristics.^27,28 A critical component of the study was to examine whether the association between stroke and cognitive decline varies by race and ethnicity. Understanding this issue is important because prior research has shown that Black and Hispanic women experience earlier cognitive aging and a higher burden of vascular risk factors.^29,30 Therefore, the approach will also explore whether stroke disproportionately accelerates cognitive decline among these populations.

Methods

This study employed a longitudinal, observational cohort design using data from the SWAN. Cognitive function was assessed annually from Wave 4 (2000) through Wave 10 (2008), providing up to 8 years of follow-up for participants. This study was reviewed and approved by the University of Florida Institutional Review Board as exempt under federal guidelines for the secondary analysis of public datasets. Written consent is not required for study approved as exempt. This manuscript followed the STROBE checklist for observational studies.

Data

The study sample consisted of 3302 women enrolled in the SWAN, a multiethnic, longitudinal, community-based study of women’s midlife aging. SWAN is a multi-site, longitudinal, epidemiological study that collects data on women’s physical, biological, psychological, and social health in their middle years to understand how midlife experiences affect health during aging. ³¹ SWAN began in 1996–1997 by enrolling women aged 42–52 from seven sites: Boston, Chicago, Detroit, Los Angeles, Pittsburgh, Oakland, and Newark. Each site recruited at least 450 participants, with targeted proportions being non-Hispanic White and from one other racial/ethnic background: African American, Chinese, Hispanic, and Japanese. Women were eligible if they had an intact uterus, at least one ovary, at least one menstrual period, and no use of reproductive hormones affecting ovarian or pituitary function in the past 3 months; were not currently pregnant or breast-feeding; and self-identified as non-Hispanic White or a member of the site-designated minority group.

Sampling procedures were approved separately for each site. In three sites, random digit dialing of numbers likely to be households and not commercial firms was used, supplemented by contacting women on lists with broad coverage, such as voter registration lists. In four sites, women were contacted from broad coverage lists: city census, utility customers, and health maintenance organization subscription list. ³¹ All sites used computer-assisted telephone interviews that were standardized to determine eligibility. Telephone numbers were unavailable for a proportion of women at one site; thus, interviewers directly contacted those households without listed telephone numbers for face-to-face interviews. The response rate was 50.7% overall, with less educated and smokers less likely to participate. Additional details regarding the SWAN study design, recruitment, and protocol have been published elsewhere. ³¹ Participants were assessed annually with interviewer- and self-administered questionnaires and physical measurements (e.g., weight, height, and blood pressure). They self-identified from five options for racial/ethnic backgrounds: Caucasian/White non-Hispanic (N = 1552, 47.00%), Black/African American (N = 934, 28.29%), Japanese/Japanese American (N = 281, 8.51%), Hispanic (N = 285, 8.63%), and Chinese/Chinese American (N = 250, 7.57%). In addition to reporting their age, marital status, household income, and educational attainment, they also provided comprehensive information regarding medical history, medication use, and lifestyle behaviors. ³²

Inclusion and exclusion criteria

Women were eligible for inclusion if they participated in at least one wave of cognitive assessment between Wave 4 (2000) and Wave 10 (2008). Additionally, they were required to have valid cognitive test scores for at least one cognitive domain and available information on stroke status during follow-up. Women were excluded if they had missing cognitive outcome measures, missing stroke status, or incomplete data on key covariates included in the multivariable models (baseline age, race/ethnicity, income, education, marital status, comorbidities, and insurance coverage).

Cognitive assessment

Cognitive assessments were conducted in the 4th (2000–2002), 6th (2002–2004), 7th (2003–2005), 8th (2004–2006), 9th (2005–2007), and 10th (2006–2008) follow-up visits. ³³ Since the first cognitive assessment available in the public SWAN dataset was conducted in the fourth follow-up visit, it is considered baseline for this analysis. Participants were able to choose between English and an alternative language in which the cognitive test battery was administered at the following sites: Los Angeles (Japanese), Davis (Chinese), and New Jersey (Spanish).

Cognitive function is assessed using four well-established measures: the Symbol Digit Modalities Test (SDMT), ³⁴ Digit Span Backwards (DSB), ³⁵ the East Boston Memory Test (EBMT), ³⁶ and Immediate Recall and the EBMT Delayed Recall. These measures were utilized to identify cognitive changes in the respective areas of information processing speed, working memory, immediate memory, and delayed memory. Information processing speed was assessed using the SDMT, a task in which participants are provided a master key from which they are to match as many numbers to symbols as possible within 90 s. ³⁴ The score reflects the total number of accurate matches, ranging from 0 to 110. Working memory was measured by the DSB task, in which participants read strings of single-digit numbers and were asked to repeat the string backward, with two trials at each string length increasing from 2 to 7. The test is discontinued after errors in both trials at any string length. ³⁷ The primary outcome for the DSB is the number of correct trials, with a score range of 0–12. Verbal episodic memory (immediate and delayed) was assessed using the EBMT. ³⁸ Participants are read a list of 15 words and asked to recall as many of the words from the list as possible immediately (immediate recall) and again after a delay of approximately 20 min (delayed recall). The primary outcome is the number of words recalled immediately and after the delay; the score range is 0–15. Higher scores on all the cognitive measures indicate better cognitive performance. We calculated Z-scores standardized to visit 4 for each test to allow direct comparisons among cognitive tasks. We subtracted the mean score at Wave 4 from the participant’s test score at each wave and divided it by the standard deviation (SD) of the Wave 4 score. ³⁹ Therefore, a Z-score equal to 1 would describe a cognitive function 1 SD above the mean score at visit 4. The Z-scores were used in regression analyses to compare regression coefficients across cognitive tests.

Stroke diagnosis

At the baseline visit (1997–1997), all women were asked whether a doctor, nurse practitioner or other health care provider had told them that they had a stroke. In each visit thereafter (1997–2008), women were asked if they had been told that they had a stroke since their last visit.

Charlson comorbidity index

The Charlson comorbidity index (CCI) is a widely used comorbidity index that weights conditions based on their association with 1-year mortality. The CCI’s inter-rater reliability was excellent, with a high agreement between self-report and medical charts. ⁴⁰ It has been shown to predict long-term mortality in different clinical populations when derived from both medical records and self-reported diagnoses of 16 distinct illnesses, including stroke and hemiplegia. ⁴¹ In stroke studies, a modified version of the CCI is used, excluding stroke and hemiplegia. ⁴² It is often used to reduce potential confounding in alignment with secondary data analyses.^{43 –45} This analysis used grouped CCI categories ⁴⁶ to capture comorbid disease burden. Since a modified CCI score above two is associated with poor outcomes among female stroke patients. ⁴⁷ The two categories of the CCI included zero or one comorbidity and two or more comorbidities. Methodologically, adjusting for CCI enhances the validity of the finding, ensuring that vascular risk disparities do not confound cognitive outcomes. ⁴⁸

Covariates

Covariates included age, race/ethnicity, education (less than college, college or above), insurance (insured, not insured), marital status (married, not married, and income (<$50,000; ⩾$50,000).^49,50 The last observation carried forward was used to account for missing values in time-varying covariates.^51,52

Imputation

Missing data occurs due to non-response, data entry errors, or equipment malfunctions during data collection or interviews. ⁵³ Whether intended or unintended, missing data can be classified to describe the type of missingness, determine whether considerations should be taken to mitigate or minimize missing values, and allow for decisions on handling missing data. ⁵⁴ If data are missing at random (MAR), imputation is a valid method for handling missingness. It helps maintain the integrity of the dataset by allowing for more accurate and reliable statistical analyses. ⁵⁵ Multiple imputation techniques, such as fully conditional specification (FCS) and joint modeling, are particularly effective for longitudinal, repeated measures data, as they account for the complexity of the time-dependent covariates.^56,57 Following the methodology of Tan et al., ⁵⁶ missing values were imputed using FCS with all model covariates. Ten datasets were imputed using chained equations. Model results were combined using the MIANALYZE procedure of SAS version 9.4. ⁵⁸

Statistical analysis

Stroke status was treated as a time-varying exposure, allowing participants to contribute person-time to the stroke-free group until the occurrence of stroke and to the post-stroke group thereafter. As a result, the number of observations classified as post-stroke varies across waves and reflects both the timing of incident stroke and intermittent missing cognitive outcome data rather than participant re-entry or selective attrition. Given that SWAN was designed to follow women through the menopausal transition and that cognitive fluctuations may occur during perimenopause, all models adjusted for baseline age and included time since baseline to account for normative aging and menopause-related cognitive change occurring across the cohort. Because participants were of similar age at enrollment and progressed through the menopausal transition during overlapping calendar periods, these effects are expected to influence both stroke and stroke-free women comparably. Stroke status was modeled as a time-varying exposure, allowing estimation of cognitive changes associated with stroke over and above age- and time-related trends.

Longitudinal generalized linear mixed models were used to estimate within-person changes in cognitive trajectories over time, leveraging repeated observations to improve statistical efficiency and reduce reliance on baseline group sizes. These models included random intercepts and slopes to evaluate the association between cognitive decline and stroke while accounting for the within-subject correlation across repeated time-dependent observations. ⁵⁹ Model fit was evaluated using Bayesian information criterion and Akaike information criterion values, and sensitivity to covariate inclusion was evaluated using inverse probability weighting as an alternative approach to address potential residual confounding.^60,61

Models controlled for age, race, ethnicity, education, income, and insurance and were used to compare the baseline level and rate of change in cognitive decline among women with stroke to those without stroke. In this approach, we estimated the mean change in each group, conditional on covariates, as in standard fixed-effects repeated measures models. In addition, the mixed-effects models included random coefficients, which provided estimates of individual differences from the group. Thus, each person was assumed to follow the average path of the group except for random effects that caused the baseline level of cognition to be lower or higher and the rate of change in cognition to be faster or slower. The variance-covariance matrix for the random coefficients was not assumed to be of a restricted form. We assumed that residual error was identically, customarily distributed, and independent of the random effects. A significant strength of this approach is the ability to model all data available for each person, regardless of the length of follow-up and spacing of evaluations.

The core mixed-effects model included terms for time since baseline, stroke diagnosis, and the interaction between stroke diagnosis and time since baseline. Time since baseline indicates the mean annual rate of decline in the cognitive measure. The term for stroke indicates the mean difference in the baseline level of cognitive function between women with and without stroke, and the interaction between stroke and time since baseline indicates the mean difference in the rate of cognitive decline between women with and without stroke. In subsequent models, we added terms for the interactions of race and ethnicity to test whether the baseline level or rate of change in cognitive decline varied along demographic lines. ⁶² An alpha of 0.05 was used for all analyses. Analyses were conducted using SAS version 9.4. ⁵⁸

Sensitivity analysis

Most women in the sample reported that they had never been diagnosed with a stroke, creating a significant imbalance in the two groups being compared. Propensity score matching (PSM) was applied to strengthen causal inference and mitigate selection bias. PSM helps to reduce selection bias by creating a matched sample where the distribution of observed covariates is similar between the treated and control groups. ⁶³ This ensures that differences in baseline characteristics do not confound the estimated treatment effect. ⁶⁴ Propensity scores were estimated using a logistic regression model with stroke status as the dependent variable and age, race/ethnicity, socioeconomic status, education, and health status as independent variables. The predicted probabilities from this model are the propensity scores. ⁶⁵ A nearest-neighbor matching algorithm with caliper restriction was employed to maximize balance while minimizing residual confounding. Standardized mean differences (SMDs) were examined to confirm the covariate balance between matched groups. ⁶⁶ SMDs were used to compare the distribution of covariates between the matched groups. A SMD of less than 0.1 was considered a good balance. ⁶⁷ Mixed-effects models with average treatment weights were then applied to analyze the matched longitudinal data. These models can handle the correlation between repeated measures and allow for the inclusion of time-varying covariates and interaction terms to evaluate potential racial/ethnic differences in post-stroke cognitive change. ⁶⁸

Results

At baseline (fourth follow-up visit), the mean age of women in the sample was 49.5 (SD = 2.63) years and 0.67% (N = 22) reported having been diagnosed with a stroke (Table 1). About 8% and 9% of women were Black and Hispanic, respectively, and 62% were married. Less than 10% of women were uninsured, and about one-third had an education equal to or beyond a college degree. About half (54%) lived in households earning $50,000 or less annually, and less than 25% had two or more comorbidities.

Table 1.

Baseline sample characteristics of women with and without a stroke diagnosis.

	Full sample (N = 3302)		No stroke diagnosis (N = 3280), 99.33%		Stroke diagnosis (N = 22), 0.67%		t-statistic	p-value
Variable	Mean	SD	Mean	SD	Mean	SD
Age (45–57)	49.53	2.64	49.53	2.64	48.59	2.36	1.67	0.0947
	N	%	N	%	N	%	χ2 statistic	p-value
Black	250	7.57	248	7.56	2	9.09	0.0731	0.7869
Hispanic	283	8.51	281	8.57	2	9.09	2.0601	0.1512
Married	2054	62.2	2043	62.29	11	50	1.4032	0.2362
Married	1248	37.8	1237	37.71	11	50
Income > $50,000	1511	45.76	1508	45.98	3	13.64	9.2082	0.0024
Income ⩾ $50,000	1791	54.24	1772	54.02	19	86.36
Education ⩾ college	2251	68.17	2242	68.35	9	40.91	7.5859	0.0059
Education < college	1051	31.83	1038	31.65	13	59.09
Insured	3029	91.73	3010	91.77	19	86.36	0.8417	0.3589
No insurance	273	8.27	270	8.23	3	13.64
0–1 comorbidities	2523	76.41	2516	76.71	7	31.82	24.4288	<0.0001
2+ comorbidities	779	23.59	764	23.29	15	68.18

CI: confidence level; SE: standard deviation.

At baseline, there was no statistically significant difference in the age of women who had been diagnosed with a stroke (M = 48.59, SD = 2.36) and those who had not (M = 49.53, SD = 2.64; t = 1.67, p = 0.094) nor was there a difference in the portion of women with and without a stroke diagnosis who were Black (χ² = 0.07, p = 0.70), Hispanic (χ² = 2.06, p = 0.15), married (χ² = 1.40, p = 0.24), or uninsured (χ² = 0.84, p = 0.359). However, 86% of women diagnosed with stroke earned $50,000 or less annually compared to only 54% of women who had not been diagnosed with stroke (χ² = 9.21, p = 0.00) and only 40.91% of women with a stroke diagnosis had graduated from college compared to 68.35% of their counterparts (χ² = 7.59, p = 0.01). Nearly 70% of women with a stroke diagnosis had two or more comorbidities compared to less than 25% of women without a stroke diagnosis (χ² = 24.43, p < 0.0001).

While only 0.67% (N = 22) of the sample reported having been diagnosed with a stroke at baseline, the number of self-reported strokes was 15, 30, 37, 33, 53, and 87 in Waves 5 through 10, respectively; thereby demonstrating an increase of 64 individual with stroke over the 8-year period. Figure 1 shows the annual and cumulative incidence of stroke from baseline to Wave 10. By the final wave, 8.39% of the sample reported having been diagnosed with a stroke.

Figure 1.

Annual and cumulative stroke diagnoses.

Table 2 lists average cognitive assessment Z-scores for women with and without a stroke diagnosis by wave. A statistically significant difference in the mean between women with and without a stroke was tested using a t-test. Despite higher mean cognition scores among women with no stroke diagnosis, few statistically significant differences appeared in Waves 4 to 6. However, all four cognitive measures showed statistically significant variations in Waves 7, 8, 9, and 10. The non-stroke cohort scored significantly higher than the stroke cohort, and the gap between the two groups widened over time.

Table 2.

Average cognition Z-score by wave and stroke diagnosis.

Wave	Non-stroke			Stroke			Mean differential
	N	Mean	SD	N	Mean	SD	t-value	Pr > |t|
	Information processing (SDMT)
4	3280	0.001461	0.997431	22	−0.21782	1.341207	1.03	0.31
5	3287	0.026656	1.021507	15	−0.21034	1.086998	0.90	0.37
6	3272	0.105161	0.96738	30	−0.11807	1.418648	1.25	0.21
7	3265	0.152539	1.001006	37	−0.22894	1.450248	2.29	0.02
8	3269	0.113524	1.068724	33	−0.44566	1.155516	2.99	0.00
9	3249	0.090212	1.030197	53	−0.54318	1.368525	4.41	<0.0001
10	3215	0.109642	1.045278	87	−0.42377	1.188652	4.68	<0.0001
	Working memory (DSB)
4	3280	0.003544	0.998691	22	−0.5283	1.077935	2.49	0.01
5	3287	0.012054	1.037211	15	−0.35326	1.202774	1.36	0.17
6	3272	0.062468	1.027779	30	−0.24164	0.904662	1.61	0.11
7	3265	0.089666	1.003178	37	0.029108	1.153168	3.36	0.05
8	3269	0.079854	1.034911	33	−0.49025	0.931689	3.15	0.00
9	3249	0.020681	1.013648	53	−0.12265	1.084055	2.14	0.04
10	3215	0.066871	1.023134	87	−0.2147	1.042611	2.53	0.01
	Immediate recall (EBMT immediate)
4	3280	0.000719	0.99964	22	−0.10722	1.071384	0.50	0.61
5	3287	−0.1181	1.158605	15	−0.46048	1.340457	1.14	0.25
6	3272	0.037763	1.036447	30	−0.02149	0.989586	0.31	0.76
7	3265	0.082763	0.984706	37	−0.38226	1.31328	2.84	0.00
8	3269	0.047982	1.233741	33	0.0426	1.0845	2.29	0.02
9	3249	−0.11831	1.229677	53	−0.59791	1.420246	2.81	0.01
10	3215	0.027997	1.065966	87	−0.84396	1.24516	4.68	<0.0001
	Delayed recall (EBMT delayed)
4	3280	0.000247	0.999234	22	−0.03682	1.133735	0.17	0.86
5	3287	−0.10967	1.152781	15	−0.38825	1.48224	0.93	0.35
6	3265	0.085921	0.99256	37	−0.48508	1.173104	0.13	0.89
7	3272	0.045631	1.029297	30	0.020321	1.163571	3.47	0.00
8	3269	0.047096	1.078632	33	−0.07968	1.173198	2.18	0.04
9	3249	−0.10679	1.188218	53	−0.70313	1.339912	3.62	0.00
10	3215	0.061769	1.034523	87	−0.78707	1.331279	7.49	<0.0001

Bold indicates significant at 95% confidence level. The number of participants classified in the stroke group varies across waves due to the time-varying definition of stroke status and availability of cognitive assessments at each wave. Counts represent wave-specific observations rather than a fixed cohort. SD: standard deviation; SDMT: Symbol Digit Modalities Test; DSB: Digit Span Backwards; EBMT: East Boston Memory Test.

Bold indicates significant at 95% confidence interval.

Table 3 and Figure 2 show that after adjusting for race, ethnicity, education, income, insurance, and marital status, women who experienced a stroke had −0.183 (95% CI: −0.30, −0.07) lower information processing speed scores, −0.0925 (95% CI: −0.20, −0.01) lower working memory scores, −0.1491 (95% CI: −0.27, −0.03) lower immediate recall scores, and −0.1732 (95% CI: −0.30, −0.05) lower delayed recall scores. In the time since baseline, stroke survivors showed lower information processing speed scores (−0.0221, 95% CI: −0.03, −0.01), working memory scores (−0.0218, 95% CI: −0.02, −0.02), immediate recall scores (−0.0235, 95% CI: −0.02, −0.02), and delayed recall scores (−0.0151, 95% CI: −0.03, −0.01).

Table 3.

GLMM estimation of cognitive decline among women.

	Information processing (SDMT)
Variable	Estimate	SE	95% CI		Z	Pr > |Z|
Intercept	0.09	0.01	0.07	0.12	6.92	<0.0001
Stroke	−0.18	0.06	−0.30	−0.07	−3.05	0.00
Time since baseline	0.02	0.00	0.01	0.02	10.39	<0.0001
No insurance	−0.52	0.03	−0.57	−0.46	−18.21	<0.0001
Black	0.14	0.02	0.10	0.18	6.31	<0.0001
Hispanic	0.45	0.02	0.41	0.49	22.40	<0.0001
Education < college	−0.05	0.01	−0.07	−0.02	−3.11	0.00
Income < $50,000	−0.14	0.01	−0.16	−0.12	−15.43	<0.0001
Married	0.08	0.01	0.07	0.10	9.63	<0.0001
2+ comorbidities	−0.15	0.01	−0.18	−0.13	−11.87	<0.0001
Time since baseline × stroke	−0.02	0.01	−0.03	−0.01	2.79	0.04
	Working memory (DSB)
Intercept	0.13	0.01	0.10	0.15	9.02	<0.0001
Stroke	−0.09	0.06	−0.20	0.00	−2.68	0.03
Time since baseline	0.01	0.00	0.01	0.01	6.42	<0.0001
No insurance	−0.29	0.02	−0.34	−0.24	−11.75	<0.0001
Black	−0.04	0.02	−0.08	0.00	−1.83	0.07
Hispanic	−0.12	0.02	−0.16	−0.08	−5.86	<0.0001
Education < college	−0.14	0.01	−0.16	−0.11	−9.82	<0.0001
Income < $50,000	−0.11	0.01	−0.13	−0.09	−10.41	<0.0001
Married	0.04	0.01	0.03	0.06	5.62	<0.0001
2+ Comorbidities	−0.12	0.01	−0.14	−0.09	−9.39	<0.0001
Time since baseline × stroke	−0.02	0.01	−0.03	−0.01	2.22	0.02
	Immediate recall (EBMT immediate)
Intercept	0.07	0.01	0.04	0.09	5.17	<0.0001
Stroke	−0.15	0.06	−0.27	−0.03	−2.50	0.01
Time since baseline	0.01	0.00	0.01	0.02	6.03	<0.0001
No insurance	−0.26	0.02	−0.30	−0.21	−10.83	<0.0001
Black	−0.12	0.02	−0.16	−0.09	−6.51	<0.0001
Hispanic	−0.01	0.02	−0.05	0.03	−0.56	0.58
Education < college	−0.02	0.01	−0.05	0.00	−1.74	0.08
Income < $50,000	−0.22	0.01	−0.24	−0.20	−22.20	<0.0001
Married	0.05	0.01	0.04	0.07	6.42	<0.0001
2+ comorbidities	−0.10	0.01	−0.12	−0.08	−7.98	<0.0001
Time since baseline × stroke	−0.02	0.01	−0.02	−0.02	4.08	0.03
	Delayed recall (EBMT delayed)
Intercept	0.05	0.01	0.03	0.08	3.88	0.00
Stroke	−0.17	0.06	−0.30	−0.05	−2.74	0.01
Time since baseline	0.01	0.00	0.01	0.02	7.47	<0.0001
No insurance	−0.24	0.02	−0.29	−0.20	−10.35	<0.0001
Black	−0.08	0.02	−0.11	−0.04	−4.38	<0.0001
Hispanic	0.00	0.02	−0.03	0.04	0.20	0.84
Education < college	−0.01	0.01	−0.04	0.01	−0.90	0.37
Income < $50,000	−0.21	0.01	−0.23	−0.19	−21.67	<0.0001
Married	0.06	0.01	0.05	0.08	7.52	<0.0001
2+ comorbidities	−0.11	0.01	−0.14	−0.09	−8.73	<0.0001
Time since baseline × stroke	−0.02	0.01	−0.02	−0.02	−5.04	0.04

GLMM: generalized linear mixed models; SDMT: Symbol Digit Modalities Test; DSB: Digit Span Backwards; EBMT: East Boston Memory Test; CI: confidence interval.

Figure 2.

Predicted mean change in cognitive Z scores (SD) among women with and without stroke.

Table 4 and Figure 3 show cognitive decline by racial-ethnic background in those with stroke compared to their stroke-free counterparts. White women who experienced a stroke had −0.200 (95% CI: −0.32, −0.08) lower information processing scores, −0.102 (95% CI: −0.21, −0.01) lower work memory scores, −0.155 (95% CI: −0.27, −0.04) lower immediate recall scores and −0.172 (95% CI: −0.30, −0.05) lower delayed recall scores when compared to White women without stroke. Black women who experienced a stroke, however, had −0.371 (95% CI: −0.76, −0.02) lower information processing scores, −0.204 (95% CI: −0.29, −0.10) lower working memory scores, 0.294 (95% CI: −0.948, −0.061) lower immediate recall, and −0.456 (95% CI: −0.99, −0.36) lower delayed recall than Black women who did not experience a stroke. Finally, since baseline, Black stroke survivors had greater declines in −0.080 (95% CI: −0.16, −0.01) information processing, −0.033 (95% CI: −0.21, −0.06) working memory, −0.0801 (95% CI: −0.16, −0.01) immediate recall, and −0.092 (95% CI: −0.19, −0.01) delayed recall than their White counterparts with stroke. Neither pre-, post-, or stroke-free Hispanic women showed any statistically significant differences in this model specification.

Table 4.

GLMM estimation of demographic differences in cognitive decline among women.

	Information processing (SDMT)
Variable	Estimate	SE	95% CI		Z	Pr > |Z|
Intercept	0.100	0.014	0.073	0.127	7.160	<0.0001
Stroke	−0.200	0.062	−0.322	−0.077	−3.200	0.001
Time since baseline	0.016	0.002	0.012	0.019	8.500	<0.0001
No insurance	−0.516	0.028	−0.571	−0.460	−18.210	<0.0001
Black	0.101	0.029	0.045	0.158	3.500	0.001
Hispanic	0.419	0.025	0.369	0.468	16.620	<0.0001
Education < college	−0.046	0.015	−0.075	−0.017	−3.110	0.002
Income < $50,000	−0.142	0.009	−0.160	−0.124	−15.450	<0.0001
Married	0.084	0.009	0.067	0.101	9.630	<0.0001
2+ comorbidities	−0.153	0.013	−0.178	−0.127	−11.880	<0.0001
Time since baseline × stroke	0.012	0.011	−0.010	0.033	1.040	0.299
Time since baseline × Black	0.011	0.006	0.000	0.021	1.930	0.054
Time since baseline × Hispanic	0.008	0.005	−0.001	0.017	1.740	0.082
Stroke × Black	−0.371	0.199	−0.760	−0.018	−1.870	0.041
Stroke × Hispanic	0.303	0.279	−0.243	0.850	1.090	0.277
Time since baseline × stroke × Black	−0.080	0.038	−0.155	−0.006	−2.110	0.035
Time since baseline × stroke × Hispanic	−0.032	0.054	−0.136	0.073	−0.590	0.556
	Working memory (DSB)
Intercept	0.136	0.014	0.108	0.164	9.500	<0.0001
Stroke	−0.102	0.057	−0.213	0.009	−1.800	0.072
Time since baseline	0.009	0.002	0.005	0.012	4.700	<0.0001
No insurance	−0.292	0.025	−0.341	−0.243	−11.740	<0.0001
Black	−0.097	0.033	−0.161	−0.033	−2.980	0.003
Hispanic	−0.169	0.028	−0.224	−0.115	−6.050	<0.0001
Education < college	−0.136	0.014	−0.164	−0.109	−9.820	<0.0001
Income < $50,000	−0.107	0.010	−0.127	−0.087	−10.450	<0.0001
Married	0.044	0.008	0.029	0.060	5.610	<0.0001
2+ comorbidities	−0.120	0.013	−0.145	−0.095	−9.410	<0.0001
Time since baseline × stroke	0.013	0.010	−0.007	0.033	1.320	0.188
Time since baseline × Black	0.016	0.007	0.002	0.030	2.200	0.028
Time since baseline × Hispanic	0.013	0.006	0.002	0.024	2.290	0.022
Stroke × Black	−0.204	0.252	−0.289	−0.098	−4.810	0.042
Stroke × Hispanic	0.122	0.341	−0.547	0.791	0.360	0.720
Time since baseline × stroke × Black	−0.033	0.045	−0.121	−0.055	−3.730	0.046
Time since baseline × stroke × Hispanic	0.002	0.074	−0.142	0.146	0.030	0.974
	Immediate recall (EBMT immediate)
Intercept	0.081	0.014	0.054	0.107	5.980	<0.0001
Stroke	−0.155	0.060	−0.273	−0.036	−2.560	0.010
Time since baseline	0.008	0.002	0.004	0.012	3.920	<0.0001
No insurance	−0.257	0.024	−0.304	−0.211	−10.830	<0.0001
Black	−0.230	0.032	−0.292	−0.167	−7.230	<0.0001
Hispanic	−0.065	0.032	−0.128	−0.001	−1.990	0.046
Education < college	−0.022	0.013	−0.047	0.003	−1.740	0.082
Income < $50,000	−0.220	0.010	−0.240	−0.201	−22.210	<0.0001
Married	0.054	0.008	0.038	0.071	6.420	<0.0001
2+ comorbidities	−0.101	0.013	−0.125	−0.076	−8.020	<0.0001
Time since baseline × stroke	0.003	0.011	−0.019	0.025	0.240	0.809
Time since baseline × Black	0.029	0.007	0.015	0.042	4.020	<0.0001
Time since baseline × Hispanic	0.015	0.007	0.001	0.028	2.170	0.030
Stroke × Black	−0.294	0.334	−0.948	−0.061	−3.880	0.038
Stroke × Hispanic	0.693	0.666	−0.613	1.999	1.040	0.298
Time since baseline × stroke × Black	−0.076	0.055	−0.183	−0.031	−2.390	0.016
Time since baseline × stroke × Hispanic	−0.188	0.122	−0.428	0.051	−1.540	0.124
	Delayed recall (EBMT delayed)
Intercept	0.064	0.013	0.038	0.090	4.790	<0.0001
Stroke	−0.172	0.065	−0.299	−0.046	−2.660	0.008
Time since baseline	0.010	0.002	0.006	0.014	5.090	<0.0001
No insurance	−0.244	0.024	−0.290	−0.198	−10.350	<0.0001
Black	−0.181	0.030	−0.239	−0.122	−6.070	<0.0001
Hispanic	−0.060	0.031	−0.120	0.001	−1.930	0.053
Education < college	−0.011	0.013	−0.036	0.014	−0.890	0.371
Income < $50,000	−0.211	0.010	−0.230	−0.192	−21.700	<0.0001
Married	0.064	0.009	0.047	0.080	7.510	<0.0001
2+ comorbidities	−0.110	0.013	−0.135	−0.086	−8.740	<0.0001
Time since baseline × stroke	0.000	0.012	−0.024	0.024	0.000	0.999
Time since baseline × Black	0.028	0.007	0.015	0.041	4.220	<0.0001
Time since baseline × Hispanic	0.017	0.007	0.005	0.030	2.640	0.008
Stroke × Black	−0.456	0.275	−0.996	−0.660	−0.084	0.010
Stroke × Hispanic	0.547	0.505	−0.443	1.537	1.080	0.279
Time since baseline × stroke × Black	−0.092	0.049	−0.189	−0.005	2.860	0.042
Time since baseline × stroke × Hispanic	−0.129	0.103	−0.330	0.072	−1.260	0.210

Bold indicates significant at 95% confidence level. GLMM: generalized linear mixed models; SDMT: Symbol Digit Modalities Test; DSB: Digit Span Backwards; EBMT: East Boston Memory Test; CI: confidence interval; SE: standard error.

Figure 3.

Predicted mean change in cognitive Z scores (SD) among women with and without stroke by race/ethnicity.

Balanced panel sensitivity analysis

The results of sensitivity analyses using the balanced panel data were like those of the main analyses, although most associations were slightly stronger. As shown in Appendix Tables 1 and 2, including more women with stroke in the sample, confirmed the robustness of these findings.

Discussion

Three key findings emerged in this study of cognitive decline among women in midlife. First, the cognitive performance of women with a history of stroke was negatively impacted, as lower cognitive scores in all four areas (information processing speed, working memory, immediate recall, and delayed recall) were observed when compared to women without stroke. Second, the magnitude of cognitive decline in years 5–10 was greater among women with a history of stroke compared to those without a history of stroke. Third, Black women experienced more significant cognitive decline after stroke when compared to White women with stroke. Understanding these interactions is essential for informing targeted interventions to mitigate disparities in cognitive health outcomes.

Lower cognitive performance after stroke and greater decline over time

The finding of lower cognitive performance at baseline between midlife women with a history of stroke compared to those without stroke is supported by prior literature. Cognitive decline is a common consequence of stroke and is associated with worse functional outcomes. ⁶⁹ Deficits can emerge in attention, memory, and executive functioning that negatively impact other non-cognitive skills, such as motor recovery after stroke. ⁶⁹ Consequently, understanding the nature and impact of cognitive issues after stroke is critical to the development of intervention strategies that address the range of functional deficits that occur after stroke. Compounding the post-stroke (new) cognitive deficits are cognitive declines known to exist during normal aging. ²⁵ Of more significant concern is the magnitude of cognitive decline that can increase after stroke. In a study of over 9000 adults, of which 5% (471) had a stroke, Zheng et al. ¹⁹ found that the rate of global cognitive decline after stroke was steeper than the decline experienced before stroke. The steepest decline rates were observed in memory, although declines were observed in semantic fluency and temporal orientation. The authors concluded that cognitive decline after stroke should be carefully considered because of the accelerated decline of cognitive skills. Consequently, primary stroke prevention strategies should start earlier in life for individuals at risk for stroke to reduce the impact of post-stroke cognitive impairments. There are additional concerns about the impact of stroke on cognitive decline, concerning other disease conditions. Delagado et al. ¹⁸ found that stroke survivors who developed dementia within 3 years of their stroke had a substantially steeper cognitive decline. This greater cognitive decline pattern suggests that stroke may predispose stroke survivors to dementia and the cognitive issues associated with dementia.

Racial differences in magnitude of cognitive decline over time after stroke

The finding of a greater magnitude of cognitive decline among Black women aligns with previous work designed to examine differences in neurocognitive outcomes between non-Hispanic Black and White stroke survivors. In a study of 120 stroke survivors (91 Black, 79 White), Johnson et al. found that Blacks are at greater risk for neurocognitive decline after stroke, and this decline was associated with worse stroke outcomes. ⁷⁰ Similarly, Zha et al. ⁷¹ found that Black stroke survivors had a higher probability of cognitive decline 5 years after stroke when compared to White stroke survivors. Similar patterns of racial differences in post-stroke cognitive decline have also been observed based on self-reports of stroke survivors. ⁷² Therefore, racial differences have been reported based on objective and subjective measures of cognitive decline. Black stroke survivors have a higher risk of recurrent stroke. ⁷³ Recurrent stroke may be one such pathway where racial differences in cognitive decline are occurring. Recurrent strokes increase the risk of functional disability, post-stroke complications, and mortality. ⁷⁴ In other words, more recurrent strokes among Black women may impact cognitive decline. ⁷⁴ Alternatively, cognitive decline is also a risk factor for recurrent stroke, ¹⁹ and differences in cognitive decline may be due to racial differences in the number of strokes. Ultimately, these collective findings highlight the need to understand the specific mechanisms contributing to racial differences in risk of neurocognitive decline, which will in turn facilitate the development and implementation of interventions designed to reduce disparities in risk of poststroke cognitive decline. They may also point to structural and social determinants of health, including differential access to healthcare, exposure to cumulative stressors, and disparities in post-stroke rehabilitation.^17,19 Additionally, they may shape the extent of cognitive decline in minoritized groups, particularly given prior findings that Black and Hispanic women experience earlier cognitive aging and a higher burden of vascular risk factors.^16,18

Limitations

The following study should be interpreted under the following limitations. Stroke diagnosis is self-reported and may be inaccurate. ⁷⁵ Furthermore, stroke was not a primary outcome of SWAN and information on the type of stroke, severity of stoke, and post-stroke treatment/intervention was not collected. In turn, we have limited information on the stroke profiles such as post-stroke morbidities and this study did not focus on CVD risk factors that may have influenced the likelihood of stroke and subsequent cognitive decline. Second, the relatively small number of women who reported having a stroke (0.67% at Wave 4 and 8.39% by Wave 10) may limit the statistical power to detect differences in cognitive decline between those with and without a stroke diagnosis. Third, FCS was used to impute missing data points. While this method has proven to be valid with longitudinal survey data, it assumes MAR. If this assumption is violated, the imputed values may be biased. ⁷⁶

Fourth, despite adjustments for various demographic and health-related factors, there may still be unmeasured confounders that could influence the observed associations between stroke and cognitive decline such as detailed clinical information regarding stroke etiology (e.g., ischemic vs. hemorrhagic), severity (e.g., NIH Stroke Scale scores), lesion location, or acute treatment. Cognitive trajectories following stroke vary substantially according to these clinical characteristics, and their absence limits the ability to draw mechanistic or lesion-specific inferences. Consequently, the findings should be interpreted as reflecting average cognitive trajectories associated with a history of stroke in midlife women rather than outcomes linked to specific stroke subtypes or severity levels. Despite this limitation, self-reported stroke has demonstrated reasonable validity in large epidemiologic cohorts, and the longitudinal design with repeated cognitive assessments allows for robust estimation of within-person change following stroke occurrence. Fifth, cognitive assessments were conducted over a period of 8 years, and changes in cognitive function could be influenced by factors not accounted for in the study, such as lifestyle changes, interventions, or other health events. Additionally, the SWAN study focuses on women during their menopause transition, which itself can impact cognitive function. This could confound the relationship between stroke and cognitive decline. Sixth, repeated cognitive testing can lead to practice effects, where participants improve simply due to familiarity with the tests rather than actual changes in cognitive function. While the relative stability of cognitive performance among stroke-free women is consistent with expected practice effects from repeated testing, the decline observed among women with stroke may suggest that stroke could attenuate or eliminate the typical learning-related gains seen in longitudinal cognitive assessments. However, observational data cannot fully disentangle practice effects from true cognitive change.

Seventh, because this study was a secondary analysis of an established longitudinal cohort, the sample size was not determined based on a study-specific a priori power calculation. As a result, the study may have been underpowered to detect more minor differences, particularly in subgroup analyses such as race-stratified estimates among women with stroke. However, the use of repeated cognitive assessments and longitudinal mixed-effects models improves statistical efficiency and supports the robustness of observed associations.

Furthermore, this study should be interpreted in the context of the SWAN cohort’s focus on the menopausal transition. Cognitive fluctuations related to perimenopause may contribute to within-person variability in cognitive performance. However, because menopausal aging occurs broadly across the cohort and was accounted for through adjustment for age and time, the observed divergence in cognitive trajectories following stroke is unlikely to be fully explained by hormonal aging alone. Nonetheless, residual confounding related to unmeasured hormonal factors cannot be completely ruled out and represents a limitation of the study.

Finally, a common limitation of the publicly available SWAN data is the survey design does not allow for examination of geographical differences such as rurality which impacts stroke outcomes. Future research would benefit of analysis of variation in cognitive decline in stroke survivors across differing stroke profiles. Finally, many factors most likely intersect to influence cognitive decline among women and stroke survivors. Exploration regarding perceived societal gender roles may provide additional insight into the mechanisms of cognitive decline in stroke, particularly among women.

Conclusions

By leveraging a diverse, longitudinal dataset and employing a robust quasi-experimental design, this study contributes to the growing literature on stroke and cognitive aging in women. Findings will have important implications for healthcare policy and clinical practice, mainly in designing interventions to reduce cognitive decline following stroke. Moreover, this research underscores the importance of addressing racial and ethnic disparities in both stroke prevention and post-stroke care to promote cognitive health equity in aging populations.