Two independent reviewers systematically searched PubMed, the Cochrane Library, Embase, and Scopus, with no restrictions on publication date. For publication language, only articles published in English were considered and included. Moreover, a manual search of reference lists for related reviews and studies was conducted. The search string was built as follows: (“blood pressure” OR hypertension OR hypotension) AND (“cognitive impairment” OR “cognitive decline” OR dementia) AND (stroke). A Supplementary File (Supplementary: Search key terms and strategy) describing the comprehensive search term framework is attached.

The systematic review adhered to the PECOS (P-participants, E-exposure, C-comparisons, O-outcomes, S-study design) framework for identifying potential eligible studies. The inclusion and eligibility criteria were defined as follows:

(1) Type of participants: Studies were included if participants were adults ($\ge$18 years) with cognitive impairment after stroke, including ischemic or hemorrhagic stroke. Participants with other cognitive dysfunction diseases (e.g., Alzheimer’s disease, Parkinson’s disease, psychiatric disorders) were excluded.

(2) Type of exposures: Studies examined systolic or diastolic BP (SBP or DBP), or both, hypertension, or hypotension were included. Studies that focused on BP variability or pulse pressure were excluded.

(3) Type of comparisons: There was no restriction on comparisons.

(4) Type of outcomes: Cognitive impairment should be operationalized or assessed using standardized instruments, including but not limited to the Mini-Mental State Examination (MMSE), Montreal Cognitive Assessment (MoCA), Cambridge Cognition Examination (CAMCOG), etc. [1].

(5) Type of study designs: Observational studies (e.g., prospective or retrospective, and cross-sectional studies) or post-hoc or secondary analyses from randomized controlled trial (RCT) reporting the odds ratio (OR) of PSCI were included.

All retrieved records were initially exported to EndNote X9 (version EndNote X9.3.2, Captivate Analytics, San Francisco, CA, USA) for deduplication. First, two independent reviewers screened the title and abstracts of the records for eligibility. In cases where an abstract was unavailable, records were preserved for a comprehensive full-text review. Second, the same investigators evaluated the complete text of potential studies to ascertain their final inclusion in the review. Finally, any discrepancies would be resolved by consensus with a senior reviewer.

The following data from each included study were extracted by two independent reviewers: First author, publication year, country, source of study, design, sample size, participants characteristics (gender, age, type of stroke), exposure characteristics (type, timing), outcome characteristics (measurements, timing), missing data (loss rate, methods to missing data), and related methods of statistical analysis for association between BP and PSCI. If key information was missing from the study report, a communication would be dispatched to the original authors via email to acquire information. Any inconsistencies were addressed through discussions with a senior reviewer for resolution.

Two reviewers independently evaluated the risk of bias (RoB) for each study using modified iterations of the Newcastle–Ottawa scale [17]. The criteria for overall quality assessment were as follows [18]: (1) Good quality: Receiving 3 or 4 stars in the selection domain AND 1 or 2 stars in the comparability domain AND 2 or 3 stars in the outcome/exposure domain. (2) Fair quality: Receiving 2 stars in the selection domain AND 1 or 2 stars in the comparability domain AND 2 or 3 stars in the outcome/exposure domain. (3) Poor quality: Receiving 0 or 1 star in the selection domain OR 0 stars in the comparability domain OR 0 or 1 stars in the outcome/exposure domain.

Meta-analyses were conducted utilizing RevMan 5.3 software (version 5.3, Cochrane Collaboration, Oxford, UK). The inverse variance method was used to pool effect estimates. Higgins I${}^{\mathrm{2}}$ statistics was used to assess statistical heterogeneity. As described in Higgins’s study, we attributed low, moderate, and high descriptors to I${}^{\mathrm{2}}$ values of 25%, 50%, and 75%, respectively [19]. If I${}^{\mathrm{2}}$ $\ge$50%, a random effect model was applied in meta-analyses; otherwise, a fixed effect model was utilized. In consideration of inconsistency, subgroup analyses were conducted according to different levels of BP. If different levels of BP were reported in one study, we pooled these statics in the same or different subgroup analysis. Sensitivity analysis used the leave-one-out method to assess the robustness of pooled effect estimates. Forest plots were used for graphical representation. If sufficient studies ($\ge$10) were included in the same outcome, we assessed publication bias using a funnel plot. Dichotomous data were provided as the OR and 95% confidence interval (CI). p-values equal to or below 0.05 were deemed statistically significant.

A total of 767 articles were identified in the electronic databases (212 from PubMed, 191 from the Cochrane Library, 200 from Embase, and 164 from Scopus), with five additional articles identified through reference screening. After excluding 22 duplicates, 750 titles, and abstracts were reviewed, and 43 full-text articles were assessed. Finally, after excluding 31 studies for ineligible outcomes [20, 21, 22, 23, 24, 25, 26], participants [27, 28, 29], exposures [13, 30, 31, 32], study designs [33, 34, 35, 36, 37, 38, 39, 40], methodological analyses [4, 14, 15, 41, 42, 43], and published languages [44], 12 articles were identified as eligible and included in this meta-analysis. Fig. 1 illustrates the study selection process flow, including the reasons for exclusion.

Fig. 1.

Flow chart depicting the search procedure. Systematic search of PubMed, the Cochrane Library, Embase, and Scopus was performed and identified 767 articles, with 5 additional articles identified through reference screening. After duplicate removal and title/abstract review, 43 citations were selected for full-text review. A total of 12 studies were included in the meta-analysis, with a total population of 21,732 patients.

Altogether, 12 articles [8, 11, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54] with 21,732 participants were included (Table 1, Ref. [8, 11, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54]). For different metrics of BP, six studies reported hypertension [46, 49, 51, 52, 53, 54], 3 reported SBP [47, 48, 50], and 1 reported DBP [45]. Individuals were classified as hypertensive if their BP exceeded 140/90 mmHg. Some of the included studies also reported the effects of different BP grades on PSCI [8, 11, 46, 47]. The details of BP measurements are presented in Table 2 (Ref. [8, 11, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54]). For the timing of BP recorded, except for 1 study, which measured BP at the follow-up visit [46], the other 11 studies measured BP at baseline or admission [8, 11, 45, 47, 48, 49, 50, 51, 52, 53, 54]. For the tools used to measure PSCI, 11 studies used scales or examinations to test PSCI, and only 1 study reported PSCI by clinical diagnosis [46]. For the timing of PSCI being observed, the most common follow-up duration was 3 months (reported in 5 studies), the shortest was at discharge, and the longest was 4.4 years. Of these, 12 included studies [8, 11, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54], 3 were secondary or post-hoc analyses from RCTs [8, 46, 48], 9 were observational studies [11, 45, 47, 49, 50, 51, 52, 53, 54], and every study used logistic regression for statistical analysis to explore the association between BP and PSCI.

The general quality was assessed as good in five studies, fair in one study, and poor in six. In the included studies, BP was evaluated using dependable methods. For comparability, we considered baseline cognitive function or the National Institutes of Health Stroke Scale (NIHSS) score as key confounders and age, sex, antihypertensive treatment, and education as additional confounders. In the outcome assessment, four studies that reported PSCI were assessed by an independent, experienced neurologist or physician, while the other eight did not. The details of the RoB assessment are presented in Tables 3-1 (Ref. [8, 11, 45, 46, 47, 48, 49, 50, 52, 53, 54]) and 3-2 (Ref. [51]).

A total of six studies were incorporated into the meta-analysis in terms of hypertension and PSCI. A forest plot (Fig. 2) showed that hypertension significantly increased the odds of PSCI in patients following a stroke (OR 1.53, 95% CI, 1.18–1.99; p = 0.001). The statistical heterogeneity was moderate (I${}^{\mathrm{2}}$ = 66%).

Fig. 2.

Forest plot of the association between hypertension and post-stroke cognitive impairment (PSCI). SE, standard error; CI, confidence interval; p-value of 0.05 or less indicated statistical significance.

The meta-analysis included 12 SBPs. Forest plot (Fig. 3) showed that SBP was significantly associated with the incidence of PSCI (OR 1.13, 95% CI, 1.05–1.23; p = 0.002). Furthermore, people with an SBP 120 mmHg, 120–139 mmHg, 140–159 mmHg, or 160–179 mmHg were more likely to have PSCI (OR 1.15, 95% CI, 1.07–1.25; p = 0.0003; OR 1.26, 95% CI, 1.06–1.49; p = 0.010; OR 1.15, 95% CI, 1.00–1.32; p = 0.05; OR 1.02, 95% CI, 1.01–1.04; p = 0.009, respectively). However, when SBP $\ge$180 mmHg, the association between SBP and PSCI became non-significant (OR 1.17, 95% CI, 0.63–2.17; p = 0.62). The overall statistical heterogeneity was moderate (I${}^{\mathrm{2}}$ = 52%).

Fig. 3.

Forest plot of association between SBP and PSCI. SE, standard error; CI, confidence interval; p-value of 0.05 or less indicated statistical significance. SBP, systolic blood pressure; PSCI, post-stroke cognitive impairment.

The meta-analysis included nine DBPs. The forest plot (Fig. 4) showed that the DBP could significantly predict the odds of PSCI (OR 1.38, 95% CI, 1.11–1.72; p = 0.004). In the subgroup analysis, DBP 80 mmHg and 80–99 mmHg did not significantly increase the odds of PSCI (OR 1.22, 95% CI, 0.95–1.58; p = 0.12; OR 1.18, 95% CI, 0.86–1.61; p = 0.30, respectively). While for people with DBP $\ge$100 mmHg, the incidence of PSCI was significantly increased (OR 1.96, 95% CI, 1.51–2.56; p 0.00001). The overall statistical heterogeneity was high (I${}^{\mathrm{2}}$ = 90%).

Fig. 4.

Forest plot of association between DBP and PSCI. SE, standard error; CI, confidence interval; p-value of 0.05 or less indicated statistical significance. DBP, diastolic blood pressure; PSCI, post-stroke cognitive impairment.

The sensitivity analysis did not affect the pooled effect estimates; the overall effect remained significant when we removed the studies one by one. Furthermore, when only studies on ischemic stroke were retained in this meta-analysis, the pooled effect estimates remained unaffected. Considering that only Hilkens et al. [46] measured BP at follow-ups, we removed this study from the meta-analysis to detect the robustness of pooled effect estimates, which remained significant, thereby indicating that the meta-analysis results were robust. Nevertheless, when we removed Gong’s study for the SBP and PSCI outcomes, the overall heterogeneity changed from 52% to 0%, meaning Gong’s study could be the main source of the moderate heterogeneity.

More than 10 studies were incorporated into the SBP and PSCI outcomes meta-analysis. The funnel plot (Fig. 5) showed the existing obvious publication bias, which may be due to some negative results in the unpublished studies.

Fig. 5.

Funnel plot of publication bias by SBP for PSCI. OR, odds ratio; SE, standard error; SBP, systolic blood pressure; PSCI, post-stroke cognitive impairment..

This meta-analysis showed that for post-stroke patients (aged from 46 to 83 years), hypertension, SBP, and DBP measured at baseline or follow-ups are risk factors for PSCI. This finding applies to common subtypes of stroke, such as ischemic and hemorrhagic stroke. Additionally, during the 4-year period after a stroke, the onset of cognitive impairment is closely related to BP, and this effect begins as early as being discharged. Furthermore, SBP or DBP at different intervals have different effects on the occurrence and progression of PSCI. The incidence of cognitive impairment in stroke patients with SBP $\ge$180 mmHg or DBP 100 mmHg may be less influenced by BP, while stroke patients with SBP 120 mmHg, 120–179 mmHg or DBP $\ge$100 mmHg are more likely to develop PSCI. However, due to limited included studies and statistical heterogeneity, the evidence of this meta-analysis may be low, and more research is needed to support our findings.

A previous systematic review presented a qualitative analysis of the association between BP and cognitive impairment for non-stroke patients [55] and highlighted that people with hypertension tended to demonstrate a more pronounced decline in global cognitive functioning compared to normotensive individuals. This finding strengthens the evidence of our work. Hypertension has been shown to be a risk factor for cognitive decline in patients with or without stroke [55, 56]. For patients with stroke, a previous study published in 2009 performed a meta-analysis of predictors for pre- and post-stroke dementia, and an insignificant association was found between hypertension and PSCI [5], contradicting the findings of this meta-analysis. A recent meta-analysis published in 2022 supported that hypertension was a predictor of post-stroke dementia [57]. The main sources of discrepancy are methodological differences and heterogeneity of included studies. More rigorous designed studies are required to detect the predictive significance of hypertension in PSCI.

In addition, our work also found SBP or DBP was significantly associated with an increase in PSCI, and this finding was similar to that of a previous meta-analysis [18], which supported that mean SBP or DBP was a risk factor for cognitive decline in patients without stroke. Furthermore, we found that both SBP 120 mmHg or 120–179 mmHg and DBP $\ge$100 mmHg were predictors of PSCI, which indicated that both normal and abnormal systolic or diastolic BP in the early post-stroke period might influence the development of PSCI; however, this effect may be indirect rather than independent or U-shaped according to previous studies [10, 11, 12]. Therefore, it is unclear whether early or long-term use of antihypertensive treatments after stroke can reduce the incidence of PSCI, while a previous study has also disputed the use of antihypertensive treatments after stroke to prevent PSCI [58]. In brief, SBP or DBP at different intervals significantly increases the incidence of PSCI, although whether this relationship is U-shaped or direct remains to be debated.

The current meta-analysis exhibits various strengths and limitations that warrant acknowledgment. The strengths include the following: First, this was the first meta-analysis that provided evidence of hypertension, SBP, and DBP on PSCI, and a significant association was found between BP and PSCI. Second, we conducted subgroup analysis according to different cut-off values of BP and found that BP at different intervals had different predictive effects of PSCI. These findings provide more theoretical and practical significance for future research. Third, this meta-analysis was performed according to MOOSE report guidelines [16]. Systematic reviews of observational studies were considered to possess significant limitations and methodological intricacies attributed to the observational nature of the studies from which they are derived [59]. Therefore, the MOOSE checklist was selected for this meta-analysis to strengthen the evidence.

As a limitation: First of all, caution should be exercised in interpreting the results of this meta-analysis owing to the restricted number of studies included and its moderate to high statistical heterogeneity. This meta-analysis included different study designs, such as observational studies and secondary analyses from RCTs. Therefore, heterogeneity arising from different study designs tends to be unavoidable, which is similar to previous meta-analyses [56]. This meta-analysis used a random effect model to minimize potential heterogeneity. In addition, heterogeneity between studies (measurement methods and data reporting) is common in observational studies; precisely, if all original studies were from the same study without heterogeneity, the larger the overall estimated variance, the larger the sample size required. Generic inverse variance is a weighting method based on the variance of the effect, reflecting the estimated effect of a pooled sample. Therefore, according to previous studies [56, 60], we used a random-effects model with a generic inverse variance method of log (OR) to conduct a meta-analysis. Furthermore, different outcome measurement tools could explain measurement bias across studies, although these are global standardized tools to define PSCI [1]. Currently, there is no definite recommendation about the sensitivity of different tools measuring PSCI; thus, for future studies, we could include specific tools to measure PSCI to reduce measurement bias.

At first, we found a significant but not U-shaped relationship between BP and PSCI. Subsequent research is required to investigate the association between BP at different intervals and PSCI. Second, high BP impacts cerebral perfusion, inducing adaptive vascular alterations, and these changes render the brain susceptible, potentially predisposing it to the onset of cognitive impairment [61, 62]. However, the etiology of PSCI is complicated; more studies are needed to explore the mechanism of high BP on PSCI. Lastly, multiple co-variables take effect when performing a regression analysis of BP on PSCI, and previous studies reported that baseline cognitive function was also a significant predictor for PSCI [58]. However, few of the included studies adjusted baseline cognitive function; thus, future studies need to take more into account when exploring predictors of PSCI, and multi-level meta-regression analysis is needed in the future to present more important predictors of PSCI.

Through meta-analysis, a notable correlation was identified between hypertension, abnormal or normal BP, and PSCI. Therefore, in individuals who have previously experienced hypertension, the use of antihypertensive treatments can prevent PSCI. However, for non-hypertensive patients with abnormal or normal BP in the early stages post-stroke, the use of antihypertensive treatments may have little significance in preventing PSCI. Additionally, considering the predictive value of BP on PSCI, regular and long-term follow-ups for BP should be performed in patients following a stroke. Finally, the etiology of PSCI is individualized, meaning PSCI is not only affected by BP, therefore, clinicians should consider all important factors of patients in order to make correct decisions on treatment.