The study was performed in conformity with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines (Supplementary Data Sheet 1). A systematic and comprehensive literature search was conducted in PubMed, Embase, Web of Science, and Cochrane Library, from their inception to 31 August 2022. We used two keywords, namely, “ischemic stroke” and “cognitive reserve” that were cross-searched by two independent reviewers. The phrase “cognitive reserve” as this term is sometimes used interchangeably with education, occupation, IQ, bilingualism, leisure activities, and socioeconomic status (14, 19, 20). In addition, our study was designed to focus on the prognosis of ischemic stroke. Keywords such as “prognosis” or “stroke outcome” were used for retrieval. No language limitations were used. The complete list of keywords for each literature search is available in Supplementary Data Sheet 2 (Table S1).

Two authors (YY and XY) independently identified the article, abstract, and keywords of each article and evaluated the eligibility. Any discrepancies were discussed and resolved by a third referee (ZS). Studies that met the following criteria were included in this systematic analysis: (1) human study (participants age ≥18 years old); (2) cohort study (prospective cohort study or retrospective cohort study); (3) CR proxies as the exposure of interest (i.e., education, occupation, IQ, bilingualism, leisure activities, and socioeconomic status); (4) mortality and functional outcomes (i.e., post-stroke cognitive impairment or post-stroke depression) as the end-point of interest; (5) minimum follow-up period ≥3 months. The following studies were excluded: studies focusing on the transient ischemic attack (TIA) or with a combined record of patients with hemorrhagic and ischemic stroke, not first-ever patients with stroke, conference abstracts, letters, comments, editorials, and case reports. We also excluded systematic reviews and/or meta-analysis, but their reference lists were searched to identify primary studies.

Data from the studies included were independently extracted by three authors (TC, WS, and WZ) through a standardized electronic form. We collected the following data elements for this study: study characteristics (first author, publication year, country, journal, and study design), demographic data (age and proportion of women), population recruitment interval, stroke types, the length of follow-up, the number of patients in the cohorts/number of participants with poor outcome (n total/n outcomes), cognitive reserve indicators, outcome definition and assessment, the type of statistical model, main findings, and the relationship between cognitive reserve and the outcome.

The prespecified primary outcome of interest was unfavorable functional outcomes, including post-stroke cognitive impairment or dementia, disability, and psychological dysfunction (depression or anxiety). Similarly, we considered the secondary outcome as death after a first-ever stroke. The modified Rankin Scale (mRS), the Barthel Index (BI), the Mini-Mental State Examination (MMSE), and the Hospital Anxiety and Depression Scale (HADS) were the common instruments to assess disability, motor impairment, cognitive dysfunction, and psychological disorders after ischemic stroke.

To critically appraise and evaluate the methodological quality of included studies, the Quality in Prognosis Studies (QUIPS) tool which is an optimal assessment tool was used, allowing an evaluation of risk bias and consideration of different CR proxies as prognostic factors (21). Accordingly, we evaluated the potential bias of each study in terms of study participation, study attrition, prognostic factor measurement, outcome measurement, study confounding, as well as statistical analysis and reporting. The outcomes were divided into low-, medium-, and high-risk biases.

The selection procedure for the systematic review is illustrated in Figure 1. The search yielded 2,526 articles from four databases (PubMed: 331, Web of Science: 1,194, Embase: 270, and Cochrane: 731). A total of five additional studies were identified from the references of relevant reviews. Among the 2,531 studies, we eliminated 453 duplicated articles and retained 44 articles for the full-text review after rigorously screening titles and abstracts. Ultimately, 28 articles met the eligibility criteria and were included in this systematic review. Given that the result data of meta-analyses demonstrated significant heterogeneity between studies, we attempted to present the synthesis of the studies in a narrative review format.

The basic characteristics of the 28 included studies are summarized in Table 1. Studies in this review encompassed a total of 1,14,212 patients with stroke and were published between 2004 and 2022, with an average or median age varied from 39.6 to 77.2 years. The follow-up period ranged from 3 months to 36 years. A total of 12 studies were conducted in China (22–24, 35, 37, 38, 40, 42, 43, 45, 47, 48), three studies in the USA (30, 32, 49), two studies in the UK (33, 39), and one study each in Italy (25), Australia (26), Sweden (27), Germany (28), Finland (29), Brazil (31), India (34), Korea (36), Spain (41), France (44), and Czech Republic (46). A total of 15 studies (22, 24, 26, 28, 29, 32–34, 37, 39, 40, 42, 45–47, 49) involving 21,517 patients were pooled for an evaluation of functional outcome, eight of which (22, 26, 28, 39, 40, 42, 47, 49) adopted prospective designs. A total of 11 studies (23, 25, 27, 30, 31, 35, 36, 41, 43, 44) reported the outcome of mortality, of which four studies (23, 31, 43, 44) were designed as prospective studies, and the rest were retrospective studies. In addition, data from two prospective studies (29, 38) were pooled for an assessment of mortality and functional outcomes. Furthermore, the data of two studies (35, 37) were from the same retrospective cohort study but reported different outcomes of stroke, so we included them both. A total of 23 studies (22–24, 26, 28, 29, 32–35, 48) enrolled patients with first-episode ischemic stroke, and five studies (25, 27, 30, 31, 44) selected patients with ischemic and hemorrhagic stroke but reported separately. In addition, 16 studies (23, 25, 27, 28, 30, 32, 35–38, 41, 42, 44, 49) defined cognitive reserve indicators as socioeconomic status, eight studies (22, 24, 29, 31, 40, 43, 45, 47) used education attainment, two studies (26, 39) measured premorbid IQ, and one study each estimated bilingualism (34) and occupation (48).

Detailed information about the judgment of each “risk of bias” domain is tabulated in Supplementary Data Sheet 2 (Table S2). Based on the QUIPS tool, no studies were excluded. In short, five (27, 29, 31, 32, 34) out of 28 studies were rated as having a high risk of bias, eight (22, 24, 26, 39, 45–48) as moderate, and 15 (23, 25, 28, 30, 33, 35–38, 40–44, 49) as low. Overall, the risk of six domains was judged to vary from low to high-risk bias, respectively, in “study participation” (low risk: 60.71%, moderate risk: 32.13%, and high risk: 7.14%), “study attrition” (low risk: 39.29%, moderate risk: 53.57%, and high risk: 7.14%), “prognosis factor measurement” (low risk: 71.43%, moderate risk: 25.00%, and high risk: 3.57%), “outcome measurement” (low risk: 92.86% and moderate risk: 7.14%), “study confounding” (low risk: 71.43%, moderate risk: 25.00%, and high risk: 3.57%), and “statistical analysis and reporting” (low risk: 46.43% and moderate risk: 53.57%) (Figure 2).

A total of 17 studies reported the functional outcome at the end of follow-up, with the endpoint interest of cognitive dysfunction, disability, motor impairment, depression, or anxiety after ischemic stroke (22, 24, 26, 28, 29, 32–34, 37–40, 42, 45–47). A summary of the findings in the included literature that identified the association between CR and functional outcomes is demonstrated in Supplementary Data Sheet 2 (Table S3).

Out of the seven studies using the modified Rankin Scale (mRS), the association between post-stroke disability and cognitive reserve was identified (24, 32, 37, 38, 42, 46, 49). Liu et al. (24) adopted a multivariate logistic regression model and found a significant relationship between poor outcomes and lower educational levels. Among the included literature, the definition of SES varied widely. A total of two studies used a composition of education, occupation, and income (32, 37). Only one study additionally included medical insurance, neighborhood status (38), and another study included marital status and place of residence (46). Another study measured a combination of education, income, caregiver, and insurance (42), and one study evaluated median household income and the area deprivation index (ADI), which combined 17 weighted census indicators (e.g., measures of education, employment, housing quality, and poverty) (49). Regardless of the definition of SES and the type of regression model used, a lower SES significantly increased the risk of adverse clinical outcomes among patients with ischemic stroke.

A total of two studies used the Barthel Index (BI) to confirm the relationship between SES and motor impairment after ischemic stroke (28, 33). Grube et al. (28) defined SES measured by education as CR proxies and found that the lower the education attainment, the worse the functional outcome. In addition, the index of multiple socioeconomic deprivations based on patient postcodes as an SES indicator was found to be strongly associated with short- and long-term motor impairment after stroke (33). Both studies adopted the multivariate logistic regression model.

A total of six studies used various cognitive testing scales to examine the correlation between CR proxies and the post-stroke cognitive outcome (22, 29, 34, 39, 40, 45). The included studies most frequently used education level as CR indicators (n = 4) and found that the higher the education level, the better the cognitive outcome (22, 29, 40, 45). Among the four studies, three of them adopted the logistic regression model (22), binomial logistic regression model and multivariate logistic regression model respectively, whereas the other study did not provide the method used (29). Using bilingualism as an indicator of CR, it is found that bilingualism contributes to better cognitive outcomes, based on the logistic regression model (34). Makin et al. (39) defined composition of premorbid IQ and education as CR indicators and found that both of them were stronger predictors of post-stroke cognition through the logistic and linear regression model.

A study, using a combination index of the mini-mental state examination (MMSE) scores, the activity of daily living (ADL) scores, and the instrument activity of daily living (IADL) scores, identified the association between premorbid IQ and the clinical outcome after ischemic stroke. Based on the multivariate logistic regression model, the study found that better outcomes after stroke had significantly higher premorbid IQ (26).

A total of 13 studies reported the mortality outcome at the end of the follow-up (23, 25, 27, 29–31, 35, 36, 38, 41, 43, 44, 48). A total of three studies defined education as CR proxies and found that educational history was associated with lower mortality after ischemic stroke through the Cox regression model and Cox proportional hazard model (31, 43). One each study regarded income or occupation as CR indicators, and both of them used the Cox proportional hazard model, but the results were different (27, 48). The association of lower income with the increase in the risk of stroke mortality was found in the study by Toivanen et al. (27). Nevertheless, Zhu et al. (48) failed to find any relationship between occupation and mortality after ischemic stroke. A total of eight studies assessed whether SES as a proxy for CR was significantly associated with mortality after stroke, based on the various definitions of SES (23, 25, 30, 35, 36, 38, 41, 44). Pan et al. (35) defined a composition indicator of education, occupation, and income as SES, and Zhou et al. (23) additionally included housing space as one of the indicators. Brown et al. (30) included the value of the housing unit, and Yan et al. (38) also included medical insurance. Similarly, it was concluded that lower SES had a negative impact on the outcome of first-ever stroke. Moreover, Shin et al. (36) calculated the Carstairs deprivation index score and individual income as SES indicators and found that low individual incomes and living in high-SES regions enhanced higher mortality among patients with stroke. Vivanco-Hidalgo et al. (41) considered the Primary Care Service Area Socioeconomic (PCAS) index scores and drug dispensation as SES indicators, finding that an individual's SES was related to survival in patients with ischemic stroke. SES measured by socioeconomic deprivation was strongly correlated with delayed mortality in patients with ischemic stroke (44). Of the eight articles, three studies adopted Cox proportional hazard models (23, 30, 36), and each opted for the logistic regression model (25), multivariate-adjusted logistic regression model (35), and mixed-effects logistic and survival model (41) and multivariate Cox model (38, 44), respectively. Supplementary Data Sheet 2 (Table S4) shows detailed information on the association between CR and mortality after stroke.