Prospective analysis of long-term trajectories of functional scores before and after a diagnosis of stroke Authors: Xianwen Shang, Lei Liu, Xueli Zhang, Yu Huang, Xiayin Zhang, Danli Shi, Yanxian Chen, Zhuoting Zhu, Zongyuan Ge, Wei Wang, Honghua Yu, Mingguang He Published: 2026 ABSTRACT BackgroundLimited data exist on long-term trajectories of functional scores before and after stroke diagnosis.MethodsWe analyzed 1325, 754, 3907, and 2902 incident strokes from China Health and Retirement Longitudinal Study (CHARLS), English Longitudinal Study of Ageing (ELSA), Health and Retirement Study (HRS), and Survey of Health, Ageing and Retirement in Europe (SHARE), respectively. Two non-stroke participants were matched to each stroke case by age and gender. Linear mixed models estimated trajectories of functional scores over time in individuals with and without stroke.ResultsHere we show that individuals with stroke have fewer memory but more depressive symptoms than those without stroke in most years from stroke onset with the largest difference several years after stroke onset. A steeper decline in memory is found among stroke patients for several years before and after stroke onset (adjusted β (95% CI) for annual change in CHARLS: −0.062 (−0.074, −0.049), ELSA: −0.024 (−0.03, −0.018), HRS: −0.026 (−0.03, −0.023)). Stroke participants have greater limitations in activities of daily living (ADL) than non-stroke participants, starting several years before stroke onset, with the difference increasing for five years after stroke onset. After that, the difference remains stable or decreases. Similar results are seen in instrumental ADL and the frailty index. Women have greater increases in CESD, mobility limitations, and frailty index compared with men.ConclusionsOur analysis of data from multiple countries identifies the period when related functional scores show the greatest change, highlighting a critical window for stroke prevention and management. Introduction Stroke was the third leading cause of global mortality, accounting for 7.3 million deaths, and the fourth leading cause of disability, responsible for 160.5 million disability-adjusted life years in 20211. It was estimated that there were 93.8 million prevalent and 11.9 million incident strokes in 20211. There will be a rapid increase in the incidence of stroke due to population growth and ageing and the increasing prevalence of vascular risk factors1,2. The estimated direct and indirect economic costs of stroke ranged from 746 billion to 1.08 trillion US dollars in 2017 and are projected to increase to between 880 billion and 2.31 trillion US dollars in 20503. Stroke has posed a tremendous burden on both health and economic systems worldwide. Therefore, the prevention and management of stroke are of great importance to reduce this increasing burden. The loss of functionality in cognition and activities of daily living (ADL) are primary concerns for rehabilitation following a stroke2,4–6. Frailty and depressive symptoms are also commonly observed among patients with stroke7,8. Studying the trajectories of function loss is important for informing advance care planning9. Previous studies have investigated changes in cognition, mobility, and depressive symptoms in individuals following a stroke10–12. A pooled analysis of nine longitudinal hospital-based cohorts from seven countries found that cognitive decline was significantly faster in patients with stroke compared with controls12. A study of 355 stroke survivors aged 75 years or older revealed a period of cognitive stability followed by rapid decline before a diagnosis of dementia10. Data from the Health and Retirement Study reported a relative stability of limitations in ADLs and Instrumental ADLs (IADLs) 4 to 6 years post-stroke, followed by a progressive increase in limitations11. More recent studies have examined the trajectories of these functional indices before and after stroke12–19, considering that the loss of these functional indices might be both risk factors for and consequences of stroke. The long-term trajectory of frailty before and after stroke is less well understood17. It remains unclear which mental and physical function scores are more sensitive to stroke onset and when substantial changes in these scores occur in the years before and after stroke. The timing and nature of functional changes around stroke are clinically critical. Pre-stroke trajectories can pinpoint a high-risk period where accelerating decline in memory, mood, mobility, or frailty indicates mounting vascular burden and impending stroke, guiding preventive measures like intensified risk factor control or early depression screening. Post-stroke trajectories define a recovery window where evidence-based care most effectively alters long-term outcomes in these domains. However, a significant gap remains in understanding the concurrence and relative sensitivity of different functional domains. Most literature focuses on single domains or single-country cohorts, leaving it unclear whether pre-stroke acceleration is a universal phenomenon or domain-specific. Furthermore, the precise ‘critical windows’—the exact years before and after onset where specific domains undergo their most rapid transformation—have not been delineated across international populations. This study fills these gaps by identifying which functional scores serve as the most sensitive early indicators of stroke and characterizing their non-linear trajectories across four distinct healthcare and socioeconomic systems (China Health and Retirement Longitudinal Study (CHARLS), English Longitudinal Study of Ageing (ELSA), HRS, and Survey of Health, Ageing and Retirement in Europe (SHARE)), comparing stroke cases with age- and sex-matched controls. Overall, data from four large cohorts demonstrate that stroke patients experience accelerated memory decline, alongside more rapid increases in depressive symptoms and functional limitations, compared to those without stroke. These trajectories are evident both prior to and following diagnosis, with the most pronounced changes occurring in the years immediately surrounding the stroke event. These findings identify a critical window—the peri-stroke period—during which cognitive and physical functions decline most acutely. Methods Participants The current analysis utilized four publicly available sister survey datasets: CHARLS, ELSA, HRS, and SHARE20–23. Access to all individual datasets was obtained through online registration and application. Written informed consent was obtained from all participants. The CHARLS study was approved by the Ethical Review Committee at Peking University. ELSA received approval from the London Multi-Centre Research Ethics Committee, the National Hospital for Neurology and Neurosurgery & Institute of Neurology Joint Research Ethics Committee, the Berkshire Research Ethics Committee, and the South Central—Berkshire Research Ethics Committee. HRS was approved by the Institutional Review Board at the University of Michigan, and SHARE by the Ethics Committee of the University of Mannheim. Additional Institutional Review Board approval was not required for this study because it involved the secondary analysis of de-identified, publicly accessible data. Details regarding population selection can be seen in the Supplementary Methods. The China health and retirement longitudinal study CHARLS is a nationally representative survey that began in 2011, tracking adults aged 45 and older in China. Follow-up surveys were conducted in 2013, 2015, 2018, and 202020. Face-to-face computer-assisted personal interviews were conducted with 10,257 households, achieving a response rate of 80.5% at baseline. The English longitudinal study of ageing ELSA is a representative cohort study of adults aged 50 and older living in England22. The study commenced in 2002 and was followed up every 2 years until 2020. Data on demographics, physical and psychosocial health, and cognitive function were collected using face-to-face interviews and self-completion questionnaires. The original sample consisted of 11391 members ranging in age from 50 to 100 years, with a response rate ranging from 73 to 82% across waves. The health and retirement study HRS is a nationally representative longitudinal survey that includes over 37,000 individuals aged 50 and older, spanning 23,000 households across the USA21. The survey has been conducted every two years since 1992, with a baseline response rate of 81.6%. Data from 1994 to 2022 were used in the analysis. The Survey of Health, Ageing and Retirement in Europe SHARE is a multinational, representative health survey of community-dwelling adults aged 50 years or older from 27 countries (covering most of the European Union and Isarel) across waves23. Data at least one wave from 137,356 participants were collected. Data were collected in 2004, 2006/2007, 2008/2009, 2011, 2013, 2015, 2017, 2019/2020, and 2021/2022. Data on demographics, physical health, behavioural risks, and mental health were collected using face-to-face interviews on a laptop. As shown in Figs. S1–S4, for each of 1325 (CHARLS), 754 (ELSA), 3907 (HRS), and 2902 participants (SHARE) who had incident stroke during follow-up, two controls were randomly selected from those free of incident stroke matched by age ( ± 1 year) and gender. To ensure valid comparisons, controls were required to be alive and stroke-free at the index date (the time of the matched case’s diagnosis). To provide sufficient data for robust trajectory estimation via linear mixed models, we included only those participants with at least two assessments in CHARLS (out of 5 possible) or at least three assessments in the other cohorts (up to 10 in ELSA, 14 in HRS, and 7 in SHARE). This criterion ensures that estimated slopes reflect long-term functional changes rather than random fluctuations between only two time points. Other information regarding individual studies is shown in the Supplementary Methods. Ascertainment of stroke In all four cohorts, stroke was identified through self-reported physician-diagnosed cases. During each survey wave, participants were asked, “Have you been informed by a doctor that you have been diagnosed with a stroke?” Those who confirmed a stroke diagnosis were then asked to provide the age or the specific year and month when the stroke was first diagnosed. The number of years since the stroke diagnosis was determined by subtracting the age at diagnosis from the age at the time of the survey wave. This diagnosis age was then assigned to the matched control group. Negative and positive values indicated years before and after the stroke, respectively. Assessment of memory For all four cohorts, the immediate and delayed recall of a 10-word list was used to test the memory (one out of four-word lists was randomly selected)24. The total score for immediate recall ranged from 0 to 10 with each correctly recalled word assigned a score of 1 and so did delayed recall. The total memory score was summed by immediate and delayed scores and ranged from 0 to 20. Other cognitive screening tasks were not analysed given they are not available across all cohorts. Activities of daily living ADL was assessed by the difficulty with doing each of the everyday activities because of a health or physical problem. The CHARLS, ELSA, HRS, and SHARE used 9, 10, 10, and 10 ADL items, respectively (Tables S1–S4). IADL was assessed by the difficulty with doing daily activities because of a physical, mental, emotional or memory problem. IADL was assessed using 11, 13, 6, and 13 items for CHARLS, ELSA, HRS, and SHARE, respectively (Tables S5–S8). Depressive symptoms Depressive symptoms were measured using the Centre for Epidemiologic Studies Depression Scale (CESD)25. Participants were asked to report the frequency they had experienced symptoms for most of the past week. In the CHARLS, the total score of depressive symptoms was calculated by summing all the 10 items, with each item scoring 0 (Rarely or none of the time ( < 1 day)), 1 (some or a little of the time), 2 (occasionally or a moderate amount of the time (3–4 days)), and 3 (most or all of the time (5–7 days)). The score ranges from 0 to 30, with higher scores indicating more depressive symptoms (Table S9). In both ELSA (Table S10) and HRS (Table S11), 8 item-version CESD was used to measure depressive symptoms. In the SHARE, 12 item-version CESD was used to measure depressive symptoms (Table S12). Frailty index Frailty index was calculated based on co-morbidity, function, cognition, depression, health attitudes and practices and physical performance26. We included 29 items (8 chronic diseases, 3 self-reported rated health conditions, 7 ADL items, 9 IADL items, 1 memory item, and 1 depression item) for the calculation of frailty index. Recoding procedures were applied for each item where absence of a deficit was assigned to 0, and presence of the deficit to 1. The total score was calculated by summing all the 29 items. The score ranged from 0 to 29 with a higher score indicating a higher level of frailty (Table S13). Covariates Data on age, gender, education, physical activity, smoking, alcohol consumption, and chronic diseases were collected using questionnaires. The highest educational attainment was categorized as low, intermediate, and high levels. Smoking was classified as never, former, and current smokers in CHARLS, ELSA, and SHARE, and as never/former and current smokers in HRS. In all cohorts, the frequency of moderate or vigorous physical activity was self-reported by participants. In CHARLS and ELSA, body mass index (BMI) was computed as measured weight in kilograms divided by the square of measured height in metres. In HRS and SHARE, BMI was calculated based on self-reported height and weight. Statistical analysis All cognitive and functional scores were rescaled to Z-scores by subtracting the mean and dividing by the standard deviation. Linear mixed models were used to estimate the trajectories of each score over time in individuals with and without incident stroke. Recognizing that while physical function declines with age, stroke may accelerate this deterioration, we hypothesized that these trajectories would involve both linear and non-linear patterns. To test this, we first established overall linear trends using fixed-effect coefficients. We then employed likelihood ratio tests to compare these linear models against non-linear alternatives using restricted cubic splines, allowing us to identify the precise timing of accelerated decline. The optimal number of knots was determined using the Akaike and Bayesian information criteria (Tables S14–S17). We tested models with two to eight knots for the trajectories of individual scores. The annual changes in each score for each time period between knots were analyzed using linear mixed models with marginal means. This dual approach—combining splines with period-specific marginal means—was designed to provide a continuous visual representation of functional change alongside quantifiable, clinician-friendly estimates of ‘annual change’ during specific windows. Consequently, we were able to confirm the non-linearity of the trajectories while quantifying the rate of decline during critical periods. Two models were tested: (1) unadjusted (model 1); (2) adjusted for age, gender, education, smoking, physical activity, alcohol consumption, BMI, hypertension, and diabetes (model 2). Within-participant correlations were accounted for by including correlated random intercepts and random slopes for time. Splines were incorporated into the models to capture potential variations in trajectories of individual scores over time. The mean difference in individual scores between those with and without incident stroke for each two-year period from stroke diagnosis was examined using linear mixed models. The trajectories of each score over time by gender were also fitted. The potential moderation of the association between incident stroke and changes in scores over time by gender was then tested. Given the multiple comparisons, false discovery rate–adjusted P values for the interaction by gender were reported. Sampling weights were applied to obtain estimates of the differences in individual scores between individuals with and without stroke, as well as the changes in individual scores over time. Multiple imputations for missing data were conducted under the assumption that data were Missing at Random (MAR). Under MAR, the probability of missingness is related to observed variables (such as age, baseline functional score, and chronic conditions) already included in the models. We employed the fully conditional specification (FCS) method for the imputation of outcomes and covariates. This method imputes each variable with missing values using a regression model conditional on the other variables, iterating through all variables until convergence. It is flexible and easy to implement, allowing variable-wise models and accommodating different types of variables. All covariates were included in the imputation models to create 10 imputed datasets. To improve the validity of the MAR assumption, we included all longitudinal outcomes, baseline covariates, and mortality status in the imputation models, as recommended by Hughes et al. to reduce bias in longitudinal settings27. Data analyses were conducted using SAS 9.4 for Windows (SAS Institute Inc.) and all P values were two-sided with statistical significance set at <0.05. Results Baseline characteristics There were 1325, 754, 3907, and 2902 incident stroke cases in the CHARLS, ELSA, HRS, and SHARE studies, respectively (Figs. S1–S4). The median follow-up periods were 4 years (interquartile range, 2–7), 10 years (4–15), 13 years (6–20), and 9 years (3–12) for CHARLS, ELSA, HRS, and SHARE, respectively. Tables S18–S21 show the characteristics of the incident stroke cases and the age- and gender-matched individuals. The median ages were 60 years (range, 44–92), 65 years (46–87), 62 years (40–96), and 67 years (40–95) for CHARLS, ELSA, HRS, and SHARE, respectively. Stroke and changes in memory There was a significant decline in memory among both individuals with and without stroke in all cohorts. Both linear and non-linear relationships were observed among those with stroke (Fig. 1). A steeper decline was found among individuals with stroke in the peri-stroke period (adjusted β (95% CI) for annual change from year −4 to 4 relative to stroke onset: −0.062 (−0.074, −0.049), P value < 0.0001) in CHARLS. The decline slowed down starting from year 4 after stroke onset (−0.018 (−0.047, 0.012), P value = 0.24). Similar results were seen in HRS (from year −11 to 4 relative to stroke onset: −0.026 (−0.03, −0.023)) and ELSA (from year −11 to 11 relative to stroke onset: −0.024 (−0.03, −0.018), Fig. S5). In both unadjusted and adjusted models, individuals with stroke had lower memory scores than those without stroke in most years before and after stroke onset. The largest differences were observed several years after stroke onset (Figs. S6–S9).Fig. 1Trajectories of memory before and after the diagnosis of stroke.CHARLS, China Health and Retirement Longitudinal Study; ELSA, English Longitudinal Study of Ageing; HRS, Health and Retirement Study; SHARE, Survey of Health, Ageing and Retirement in Europe. The year from stroke onset for individuals with incident stroke was assigned to the corresponding matched controls. The optimal number of knots was determined with the use of the Akaike and Bayesian information criteria. The shading in the figure represents the 95% confidence interval. *refers to significance for linear relationships (P value < 0.05). Linear relationships were tested using the fixed effect coefficient in a linear mixed model. †refers to significance for non-linear relationships (P value < 0.05). Non-linear relationships were tested by comparing models with and without quadratic/spline terms using likelihood ratio tests. Trajectories of memory before and after the diagnosis of stroke. CHARLS, China Health and Retirement Longitudinal Study; ELSA, English Longitudinal Study of Ageing; HRS, Health and Retirement Study; SHARE, Survey of Health, Ageing and Retirement in Europe. The year from stroke onset for individuals with incident stroke was assigned to the corresponding matched controls. The optimal number of knots was determined with the use of the Akaike and Bayesian information criteria. The shading in the figure represents the 95% confidence interval. *refers to significance for linear relationships (P value < 0.05). Linear relationships were tested using the fixed effect coefficient in a linear mixed model. †refers to significance for non-linear relationships (P value < 0.05). Non-linear relationships were tested by comparing models with and without quadratic/spline terms using likelihood ratio tests. Stroke and changes in depressive symptoms Both linear and non-linear relationships were found among individuals with stroke. A significant increase in CESD scores was observed in individuals both with and without stroke, with a more rapid escalation among those with stroke (Fig. 2). There was a sharp increase in CESD scores around the time of stroke onset in all cohorts (adjusted β (95% CI) for annual change from −5 to 3 years relative to stroke onset for CHARLS: 0.049 (0.037, 0.061); from −7 to 4 years for ELSA: 0.022 (0.013, 0.032); from −18 to 6 years for HRS: 0.023 (0.021, 0.026); from −5 to 7 years for SHARE: 0.008 (0.001, 0.015), Figure S10). Individuals with stroke had higher CESD than those without stroke in most years following stroke onset. The largest differences were observed several years after stroke onset, and these associations remained significant after adjustment for covariates (Figs. S11–S14).Fig. 2Trajectories of depressive symptoms before and after the diagnosis of stroke.CHARLS, China Health and Retirement Longitudinal Study; ELSA, English Longitudinal Study of Ageing; HRS, Health and Retirement Study; SHARE, Survey of Health, Ageing and Retirement in Europe. The year from stroke onset for individuals with incident stroke was assigned to the corresponding matched controls. The optimal number of knots was determined with the use of the Akaike and Bayesian information criteria. The shading in the figure represents the 95% confidence interval. *refers to significance for linear relationships (P value < 0.05). Linear relationships were tested using the fixed effect coefficient in a linear mixed model. †refers to significance for non-linear relationships (P value < 0.05). Non-linear relationships were tested by comparing models with and without quadratic/spline terms using likelihood ratio tests. Trajectories of depressive symptoms before and after the diagnosis of stroke. CHARLS, China Health and Retirement Longitudinal Study; ELSA, English Longitudinal Study of Ageing; HRS, Health and Retirement Study; SHARE, Survey of Health, Ageing and Retirement in Europe. The year from stroke onset for individuals with incident stroke was assigned to the corresponding matched controls. The optimal number of knots was determined with the use of the Akaike and Bayesian information criteria. The shading in the figure represents the 95% confidence interval. *refers to significance for linear relationships (P value < 0.05). Linear relationships were tested using the fixed effect coefficient in a linear mixed model. †refers to significance for non-linear relationships (P value < 0.05). Non-linear relationships were tested by comparing models with and without quadratic/spline terms using likelihood ratio tests. Stroke and changes in mobility Linear relationships between years from stroke onset and ADL were found in CHARLS and SHARE, while both linear and non-linear relationships were observed in ELSA and HRS (Fig. 3). A steeper increase in ADL was seen in individuals with stroke (adjusted β (95% CI) for annual change: 0.080 (0.072, 0.088)) compared to those without stroke (0.019 (0.013, 0.025)) in CHARLS. Similar results were found for SHARE (stroke: 0.028 (0.024, 0.031); non-stroke: 0.007 (0.004, 0.010)). A steeper increase in ADL was observed in the years surrounding stroke onset in ELSA (from −5 to 3 years relative to stroke onset: 0.043 (0.032, 0.053)) and HRS (from −10 to 8 years: 0.058 (0.054, 0.061), Fig. S15). Individuals with stroke had higher ADL scores than those without stroke in the adjusted model, especially in all years following stroke onset. The largest difference was observed in the last year after stroke onset (Figs. S16–S19). Similar results were found for IADL (Figs. S20–S25).Fig. 3Trajectories of activities of daily living before and after the diagnosis of stroke.CHARLS, China Health and Retirement Longitudinal Study; ELSA, English Longitudinal Study of Ageing; HRS, Health and Retirement Study; SHARE, Survey of Health, Ageing and Retirement in Europe. The year from stroke onset for individuals with incident stroke was assigned to the corresponding matched controls. The optimal number of knots was determined with the use of the Akaike and Bayesian information criteria. The shading in the figure represents the 95% confidence interval. *refers to significance for linear relationships (P value < 0.05). Linear relationships were tested using the fixed effect coefficient in a linear mixed model. †refers to significance for non-linear relationships (P value < 0.05). Non-linear relationships were tested by comparing models with and without quadratic/spline terms using likelihood ratio tests. Trajectories of activities of daily living before and after the diagnosis of stroke. CHARLS, China Health and Retirement Longitudinal Study; ELSA, English Longitudinal Study of Ageing; HRS, Health and Retirement Study; SHARE, Survey of Health, Ageing and Retirement in Europe. The year from stroke onset for individuals with incident stroke was assigned to the corresponding matched controls. The optimal number of knots was determined with the use of the Akaike and Bayesian information criteria. The shading in the figure represents the 95% confidence interval. *refers to significance for linear relationships (P value < 0.05). Linear relationships were tested using the fixed effect coefficient in a linear mixed model. †refers to significance for non-linear relationships (P value < 0.05). Non-linear relationships were tested by comparing models with and without quadratic/spline terms using likelihood ratio tests. Stroke and changes in frailty index Linear relationships between years from stroke onset and the frailty index were found among individuals in SHARE, while both linear and non-linear relationships were observed in CHARLS, ELSA, and HRS (Fig. 4). A sharp increase in the frailty index around the time of stroke onset was observed in CHARLS, ELSA, and HRS. A more rapid increase in the frailty index was found in individuals with stroke compared to those without stroke in SHARE (Fig. S26). Individuals with stroke had higher frailty index than those without stroke in most years following stroke onset, and a larger difference was seen in those with longer stroke duration (Fig. S27–S30).Fig. 4Trajectories of frailty index before and after the diagnosis of stroke.CHARLS, China Health and Retirement Longitudinal Study; ELSA, English Longitudinal Study of Ageing; HRS, Health and Retirement Study; SHARE, Survey of Health, Ageing and Retirement in Europe. The year from stroke onset for individuals with incident stroke was assigned to the corresponding matched controls. The optimal number of knots was determined with the use of the Akaike and Bayesian information criteria. The shading in the figure represents the 95% confidence interval. *refers to significance for linear relationships (P value < 0.05). Linear relationships were tested using the fixed effect coefficient in a linear mixed model. †refers to significance for non-linear relationships (P value < 0.05). Non-linear relationships were tested by comparing models with and without quadratic/spline terms using likelihood ratio tests. Trajectories of frailty index before and after the diagnosis of stroke. CHARLS, China Health and Retirement Longitudinal Study; ELSA, English Longitudinal Study of Ageing; HRS, Health and Retirement Study; SHARE, Survey of Health, Ageing and Retirement in Europe. The year from stroke onset for individuals with incident stroke was assigned to the corresponding matched controls. The optimal number of knots was determined with the use of the Akaike and Bayesian information criteria. The shading in the figure represents the 95% confidence interval. *refers to significance for linear relationships (P value < 0.05). Linear relationships were tested using the fixed effect coefficient in a linear mixed model. †refers to significance for non-linear relationships (P value < 0.05). Non-linear relationships were tested by comparing models with and without quadratic/spline terms using likelihood ratio tests. Trajectories of functional cores before and after stroke onset A larger annual change in ADL/IADL and memory was found five years before and after stroke onset compared to other scores across four cohorts (Fig. 5). The annual changes in ADL adjusted β (95% CI) in CHARLS: 0.064 (0.044, 0.083) before stroke, 0.044 (0.012, 0.077) after stroke; ELSA: 0.035 (0.018, 0.052) before stroke, 0.024 (0.002, 0.045) after stroke; HRS: 0.085 (0.077, 0.093) before stroke, 0.024 (0.011, 0.037) after stroke; SHARE: 0.034 (0.019, 0.049) before stroke, 0.033 (0.003, 0.062) after stroke. The frailty index changes within five years before stroke onset were more sensitive to stroke (Fig. S31).Fig. 5Trajectories of individual scores before and after the diagnosis of stroke among individuals with stroke.CHARLS China Health and Retirement Longitudinal Study, ELSA English Longitudinal Study of Ageing, HRS Health and Retirement Study, SHARE Survey of Health, Ageing and Retirement in Europe. The shading in the figure represents the 95% confidence interval. Trajectories of individual scores before and after the diagnosis of stroke among individuals with stroke. CHARLS China Health and Retirement Longitudinal Study, ELSA English Longitudinal Study of Ageing, HRS Health and Retirement Study, SHARE Survey of Health, Ageing and Retirement in Europe. The shading in the figure represents the 95% confidence interval. Changes in functional cores before and after the diagnosis of stroke by gender Similar trajectories of memory were found in both men and women across all cohorts (Fig. S32). The strength of the inverse association between stroke and memory differed by sex over time: it was initially stronger in men, but later became stronger in women (Figs. S33–S36). Although women had higher CESD scores than men, the trajectories of CESD were similar for both genders (Fig. S37). The positive association between stroke and CESD was stronger in women than in men for most years following stroke onset (Figs. S38–S41). Individuals with stroke had a higher rate of increase in ADL compared to those without stroke, with similar trends observed in both women and men (Fig. S42). The association between stroke and ADL was stronger in women than in men (Figs. S43–S46). Similar results were observed for IADL (Figs. S47–S51). There was an increasing trend in the frailty index with years from stroke onset. Individuals with stroke had higher frailty index scores than those without stroke in both women and men for most years following stroke onset (Fig. S52). The association between incident stroke and the frailty index was stronger in women than in men for most years following stroke onset (Figs. S53–S56). Sensitivity analysis The analysis of individuals with complete data showed results consistent with the main analysis (Figs. S57–S62). Discussion Our analysis based on four large longitudinal cohorts from China, UK, European Union, and the USA, showed a higher decline rate in memory and a higher increase rate in CESD, ADL/IADL and frailty index in individuals with incident stroke than those without stroke. Both linear and non-linear relationships were observed between functional scores and years from stroke onset, with a steeper decline or increase rate in the several years before and after stroke onset. Individuals with stroke had lower memory but higher mobility limitations and frailty index than those without stroke in most years from stroke onset with a largest difference achieved in several years after stroke onset. Larger annual changes in ADL and frailty index within five years before stroke onset whereas larger changes in CESD within five years after stroke onset were seen. The potential adverse effect of stroke on changes in mobility limitations and frailty index was larger in women than in men. Taken together, these patterns support a model in which a stroke occurs against a background of progressive vascular and systemic damage, manifesting as accelerated deterioration in memory, mobility, depressive symptoms, and frailty before the clinical event. The marked divergence from non‑stroke controls in the years immediately before and after the stroke suggests that these trajectories are not simply due to normal aging but likely reflect subclinical cerebrovascular burden and its interaction with acute brain injury and subsequent adaptation. Our study is consistent with previous research showing a significant decline in memory among both stroke and non-stroke participants12,18,28, with a steeper decline observed in stroke patients. We also found there was no significant difference in memory between stroke and non-stroke participants until several years before stroke onset, whilst a larger difference was observed with more years after the stroke. Likely, the Rotterdam Study showed that stroke patients had higher decline rate in cognition compared with non-stroke individuals28. A meta-analysis also demonstrated that stroke patients experienced higher cognitive decline than that of non-stroke controls from 1 to 3 years after onset12. In contrast, data from the SHARE with 2,362 incident strokes between 2004 and 2019 revealed an acute decrease in memory after stroke but similar decline between stroke and non-stroke participants14. Our research reveals a steeper memory decline beginning several years before stroke onset and continuing afterward in stroke patients than that in non-stroke participants in CHARLS, ELSA and HRS, which may help to identify critical periods for addressing memory loss in individuals with stroke. We found stroke patients had higher CESD scores than non-stroke participants starting around five years before stroke onset. This difference increased as the time to stroke onset decreased and continued to grow until five years after the stroke, then became stable or even decreased until around ten years after the stroke. Our findings are consistent with a meta-analysis demonstrating that the association between depression and risk of stroke was stronger among studies with shorter follow-up duration29. Another meta-analysis has shown that there was a higher incidence of depression among stroke patients than non-stroke controls30. This suggests there may be a bidirectional relationship between depression and stroke. Although previous research has examined the changes in depressive symptoms before and after the first stroke17, the trajectories of depressive symptoms throughout the whole period have not been tested yet. The sharp increase in CESD scores around the time of stroke and the sustained higher level thereafter indicate that depression is not only a risk marker for stroke but also a treatable complication that may influence subsequent recovery and disability. This underscores the need for systematic depression screening and timely treatment both in individuals at high vascular risk and in stroke survivors during the early post‑stroke years. Our study reveals that stroke participants had larger ADL scores than non-stroke participants starting from five years before stroke onset, and this difference increased until five years after stroke onset. Similar results were seen for IADL. In the Rotterdam Study with a mean follow-up duration of 12.5 years, the differences between the trajectories of stroke and non-stroke participants were significant at 2.2 years before stroke for ADL and 3.0 years for IADL28. This indicates functional deterioration may start years before a first-ever stroke. We also found the difference in ADL and IADL between stroke and non-stroke participants remained stable or even decreased 5 years after stroke. This pattern could potentially be attributed to attrition bias if individuals with more severe impairment were disproportionately lost to follow-up. However, sensitivity analyses using imputed data produced results consistent with those from the complete-case analysis, suggesting that attrition bias is unlikely to be the primary driver of the observed trajectory. The observation that ADL and IADL trajectories begin to diverge several years before a stroke and show the largest differences in the first five years after a stroke suggests two potential intervention windows. The first is a pre‑stroke window, where emerging limitations in daily function may signal heightened stroke risk and justify intensified vascular risk-factor management and potential targeted interventions for frailty or mobility. The second is an early post‑stroke window, where aggressive rehabilitation, environmental adaptation, and support services may have the greatest impact in preventing long-term disability. The trajectories of the frailty index over the years following a stroke are similar to those of ADL/IADL in our study. This may be attributed to the fact that the calculation of the frailty index involves components such as memory, depressive symptoms, and ADL/IADL, with ADL/IADL accounting for a significant weight26. We found that changes in ADL/IADL were most sensitive to stroke onset following changes in memory and CESD. ADL/IADL and memory had steeper changes before stroke onset whereas CESD steeper changes after stroke onset. It is of great importance to control depression symptoms after a stroke, and all these scores in the several years before a stroke are predictors of stroke. The adverse effect of stroke on CESD, mobility, and frailty index was greater in women, suggesting that potential intervention priorities should be gender-specific. Women, on average, experience their first stroke at an older age than men (not shown in our study) and tend to have more severe strokes31. These factors may partly explain their worse outcomes following stroke. To our knowledge, this is the first study involving four prospective cohorts from different ethnicities, with large sample sizes and long follow-up durations, to examine the trajectories of multiple functional scores before and after stroke onset. Admittedly, our study has several potential limitations. Firstly, only first strokes were taken into consideration in our analysis, recurrent strokes that might influence the changes in functional scores should be tested in future research32. Secondly, stroke patients who were older at baseline, especially those diagnosed in the second or third waves, might have died during follow-up later, potentially affecting the trajectories of functional scores. Thirdly, although we included multiple covariates in our analysis, unknown confounders might still impact the associations. Fourth, the trajectories of certain scores (memory and frailty) in CHARLS and SHARE differed from those in ELSA and HRS, while ELSA and HRS exhibited similar trajectories across all scores. These discrepancies may reflect differences in age, ethnicity, and follow-up duration across cohorts. Fifth, the median age of included stroke participants was somewhat lower than that of typical stroke cohorts, which may be due to stroke patients being more likely to be lost to follow-up because of higher mortality risk (only those with two or more surveys were included in the analysis). It may also be attributable to the increasing incidence of stroke among young adults in recent decades globally33. However, the lower median age is unlikely to substantially influence the identified trajectory patterns, especially given age matching and age adjustment. Sixth, although stroke was confirmed if the year or age at diagnosis was reported, or if it was reported again in a subsequent survey wave, reliance on self-reported physician diagnosis may have resulted in measurement error. Misclassification was more likely to involve false positives, which would tend to bias the associations between stroke and trajectories of functional scores toward the null. In addition, stroke severity, type, and anatomical location were unknown in the current analysis, which may have influenced the observed associations. Lastly, although the four cohorts included were nationally representative samples, there may be selection bias since we only included stroke patients and matched controls in our analysis. We also observed expected differences in mortality and assessment frequency between cases and controls. Such imbalances can introduce attrition bias, where ‘healthier’ survivors are over-represented in later years. We addressed this by utilizing linear mixed models, which are robust to data missing at random, and by performing multiple imputation sensitivity analyses that included mortality as a predictor. The consistency between these results and our primary analysis suggests that the fundamental shape of functional decline is not solely driven by the loss of the most severely affected participants. Conclusions Stroke patients show greater declines in memory, and greater increases in CESD, ADL/IADL and frailty index, compared to non-stroke individuals. These changes are most pronounced in the years immediately before and after the stroke. Men experience a larger decline in memory, while women have greater increases in mobility limitations and frailty index. These findings highlight the need for potential early and gender-specific interventions to address the functional impacts of stroke. While the overall international patterns are consistent, cohort-specific contrasts in the magnitude and shape of trajectories suggest that national context—measurement, care pathways, and social supports—may modulate functional changes around stroke. Supplementary information Transparent Peer Review file Supplemental Information Transparent Peer Review file Supplemental Information