The systematic literature search was performed by a librarian specialist in two electronic databases, Embase and MEDLINE, to identify studies published in English in the period 1980–2022 in accordance with the PRISMA guidelines. Additional records were identified from recently published literature in 2023. The search strategy used to retrieve the relevant database records published between January 1980 and September 2022 was composed of a conceptual string with the following combination of terms or their analogs or synonyms: “cerebrovascular” AND “ischemic” AND “malnutrition” AND “DHA OR EPA OR uridine OR choline OR folate OR vitamin B6 OR vitamin B12 OR vitamin C OR vitamin E OR selenium OR taurine OR arginine OR CoQ10 OR carnitine” NOT “animal.” Non peer reviewed sources, such as conference abstracts, book chapters, and video-audio media were excluded. The full search strategy is available in the Supplementary Table S1. The review was not registered; no protocol was prepared for registration.

Records were screened based on title and/or abstract and were excluded in case of reviews, case reports, duplicates and when the search criteria were not met. For example, studies that did not present data on any of the nutrients within the scope of this systematic review were excluded. The studies were assessed based on the full-text and excluded if the study population was <18 years old, or if the population included exclusively subjects with haemorrhagic stroke or transient ischemic attack, or it did not include a control group. The inclusion of a study in the meta-analysis required that the blood levels of the nutrients and the number of subjects assessed were available for the stroke and control group. Studies presenting data for more than one nutrient or for independent subgroups were treated independently and constituted the total number of reports. Study selection and data extraction was performed by three independent reviewers and the final dataset was discussed with the review team.

The blood levels of DHA, EPA, uridine, choline, folate, vitamin B6, vitamin B12, vitamin C, vitamin E, selenium, CoQ10, carnitine, arginine, and taurine in stroke cases and controls were extracted. Data on the type of stroke and timing after stroke were recorded. Timing was based on the definition of the stroke phases provided by the Stroke Recovery and Rehabilitation Roundtable Taskforce (Bernhardt et al., 2017), simplified to: acute (<7 days), subacute (between 7 days and 6 months), and chronic (6 months or more). Data on the type of controls were also extracted. Mean age and the number of subjects assessed were collected for both the stroke and control groups.

Risk of bias (high, unclear, or low) was evaluated by two independent assessors for each study on the basis of 4 domains: selection, detection, attrition, and reporting bias. These were selected from the ROBINS-I tool (Sterne et al., 2016) as being most relevant for non-randomized observational studies. In short, each assessor rated the risk per domain as either high, unclear, or low. Then, all assessments were compared and sporadic disagreements were discussed between assessors to come to an agreement on the final score.

In addition, publication bias was assessed with Egger's test if possible (see Section 2.5).

Blood levels of the nutrients were expressed as mean ± standard deviation (SD). In the studies where the mean ± SD was available in subgroups rather than for the whole group of stroke patients or controls, the weighted mean and the pooled SD were calculated. When a study reported the mean value but the SD was missing the SD was estimated based on the standard error of the mean (SEM) or the figures provided in the study, if available. When a study reported median, first and third quartile, the formulas from Wan et al. (2014) were used to impute mean and SD. In case the median was available with the range of the data, the formulas from Hozo et al. (2005) were used to impute mean and SD.

To allow comparison across studies, the standardized mean difference (SMD) estimated by Hedges' g, was used to standardize the results of the studies to the same scale, which expresses the size of the effect in each study relative to the variability observed in that study.

In the meta-analyses where the number of reports was >3, the overall effect was estimated with a random-effect (RE) meta-analysis model with Hartung Knapp (HK) modification for unadjusted data. The analysis was replicated including age as a covariate if the number of reports was >3; age-adjustment was obtained by correcting for centralized age differences between groups. In case the mean age was not available and the median (range) was reported, or the mean age was available only in the stroke or the control group and these groups were age matched, the mean value of the age was imputed. The between-study variance was estimated using the I² statistic, where an I² > 95% indicates considerable amount of heterogeneity.

The publication bias was evaluated with Egger's test (regression test for funnel plot asymmetry) if the number of reports included in the meta-analysis was ≥10. The RE meta-analysis with HK modification and the Egger's test output was generated using Metafor package in R (Viechtbauer, 2010; R Core Team, 2022).

The number of stroke subjects included in the 105 reports was 5,725 and the mean age was 64.1 (range: 27.7–77.2) years. Overall, the main diagnosis was ischemic stroke and the assessment 165 of the nutrient blood levels occurred within the acute (k = 59), the subacute (k = 22), or the chronic (k = 11) 166 phase. For the remaining reports (k = 13), the moment of assessment was not indicated. Two studies with assessments overlapping phase borders, were assigned to the best fitting phase, i.e., 1–10 days (Han et al., 2002) to acute, and 21–42 weeks (Brattstrom et al., 1992) to chronic. The number of controls included in the reports was 16,171 and it primarily included relatively healthy subjects whose mean age was 60.6 (range: 27.0–77.8) years. The results of the 105 reports consisted of the following meta-analysis groups: folate (k = 27), vitamin B12 (k = 23), vitamin E (k = 11), DHA (k = 7), EPA (k = 7), vitamin B6 (k = 6), vitamin C (k = 6), selenium (k = 6), arginine (k = 4). Nutrients with three or fewer available reports, i.e., taurine (k = 2), CoQ10 (k = 2), choline (k = 2), carnitine (k = 1), and uridine (k = 1), were not included in the meta-analysis and were treated separately. A summary of the results is reported in Supplementary Table S2.

Risk of selection bias was present in 16%, and unclear in 38% of the studies. Selection bias was obvious where specific patients with a higher risk of malnutrition due to dysphagia or the use of specific medication were excluded. Similarly, the selection of relatively unhealthy control subjects, such as elderly patients without cerebrovascular diagnosis, was considered a source of bias. Risk of detection bias was present in 9%, and unclear in 32% of the studies. In most cases detection bias was due to a lack of fasting in one or both groups (stroke patients and controls) before blood sampling. In addition, unblinded handling or analysis of samples could have introduced bias. Risk of attrition bias and risk of reporting bias were very low to absent in the selected studies; there were hardly any missing data, and all outcomes were clearly reported. Overall results of the risk of bias assessment are summarized in Figure 2.

Twenty-seven reports with a total of 3,279 stroke cases and 2,679 controls were included in the meta-analysis (Brattstrom et al., 1992; Hultberg et al., 1997; Williams et al., 2001; Han et al., 2002; Karttunen et al., 2002; Kelly et al., 2003; Sachdev et al., 2003; Glew et al., 2004; Kocer et al., 2004; Fang et al., 2005; Liu et al., 2005; Urbanska et al., 2006; Angelova et al., 2008; Mojiminiyi et al., 2008; Kara et al., 2009; Kim et al., 2010; Moghaddasi et al., 2010; Mohapatra and Sarangi, 2010; Osunkalu et al., 2010; Mejia Mohamed et al., 2011; Omrani et al., 2011; Ashjazadeh et al., 2013; Jiang et al., 2014; Cho et al., 2015a,b; Rudreshkumar et al., 2018; van Wijk et al., 2023). The mean number of subjects was 90.2 (range: 20–579) in the stroke groups and 99.2 (range: 20–427) in the control groups. The mean age was 60.8 (range: 27.7–72.7) years in the stroke cases and 58.6 (range: 27.0–71.7) years in the controls. Blood levels of folate were lower in stroke patients compared to controls for unadjusted [t₍₂₆₎ = −3.03, p = 0.006] and age-adjusted [t₍₂₅₎ = −2.97, p = 0.006] model (Figure 3). The Egger's test [t₍₂₅₎ = −0.59, p = 0.56] indicated no evidence of publication bias (Supplementary Figure S1).

Twenty-three reports with a total of 2,075 stroke cases and 1,707 controls were included in the meta-analysis (Brattstrom et al., 1992; Hultberg et al., 1997; Williams et al., 2001; Han et al., 2002; Karttunen et al., 2002; Kelly et al., 2003; Sachdev et al., 2003; Glew et al., 2004; Kocer et al., 2004; Liu et al., 2005; Urbanska et al., 2006; Angelova et al., 2008; Kara et al., 2009; Salemi et al., 2009; Moghaddasi et al., 2010; Mohapatra and Sarangi, 2010; Mejia Mohamed et al., 2011; Omrani et al., 2011; Ashjazadeh et al., 2013; Jiang et al., 2014; Rudreshkumar et al., 2018; Kweon et al., 2021; van Wijk et al., 2023). The mean number of subjects was 90.2 (range: 20–199) in the stroke groups and 74.2 (range: 20–249) in the control groups. The mean age was 61.5 (range: 27.7–72.7) years for stroke cases and 59.7 (27.0–71.7) years for controls. Blood levels of vitamin B12 were lower in stroke patients compared to controls for unadjusted [t₍₂₂₎ = −3.58, p = 0.002] and age-adjusted [t₍₂₁₎ = −3.55, p = 0.002] model (Figure 4). The Egger's test [t₍₂₁₎ = −0.27, p = 0.79] indicated no evidence of publication bias (Supplementary Figure S2).

Eleven reports with a total of 457 stroke cases and 484 controls were included in the meta-analysis (Singh et al., 1995; Daga and Madhuchhanda, 1997; Chang et al., 1998, 2005; Cherubini et al., 2000; Gariballa et al., 2002; Sanchez-Moreno et al., 2004; Mao et al., 2008; van Wijk et al., 2023). Mao et al. (2008) reported three subgroups and one control group. Therefore, the control group was divided into three equal subgroups and using the same mean and SD for each control group. The mean number of subjects was 41.5 (range: 12–110) in the stroke groups and 53.8 (range: 20–202) in the control groups. The mean age was 65.9 (range: 52.5–77.2) years for stroke cases and 62.9 (range: 46.2–77.8) years for controls. The blood levels of vitamin E were lower in stroke patients compared to controls for the unadjusted [t₍₁₀₎ = −3.00, p = 0.013] and for the age-adjusted [t₍₈₎ = −2.56, p = 0.034] model (Figure 5). The Egger's test [t₍₉₎ = 0.93, p = 0.38] indicated no evidence of publication bias (Supplementary Figure S3).

Seven reports with a total of 773 stroke cases and 608 controls were included in the meta-analysis (Park et al., 2009; Kim et al., 2012; Ikeya et al., 2013; Shang et al., 2018; Szczuko et al., 2020; Hutanu et al., 2021; van Wijk et al., 2023). The mean number of subjects was 110.4 (range: 40–248) in the stroke groups and 86.9 (range: 35–215) in the control groups. The mean age was 66.8 (range: 59.0–72.5) years for stroke cases and 65.3 (range: 54.4–75.4) years for controls. Blood levels of DHA were lower in stroke patients compared to controls for both the unadjusted [t₍₆₎ = −3.37, p = 0.015] and age-adjusted [t₍₅₎ = −3.66, p = 0.015] model (Figure 6).

Seven reports with a total of 773 stroke cases and 608 controls were included in the meta-analysis (Park et al., 2009; Kim et al., 2012; Ikeya et al., 2013; Shang et al., 2018; Szczuko et al., 2020; Hutanu et al., 2021; van Wijk et al., 2023). The mean number of subjects was 110.4 (range: 40–248) in the stroke groups and 86.9 (range: 35–215) in the control groups. The mean age was 66.8 (range: 59.0–72.5) years for stroke cases and 65.3 (range: 54.4–75.4) years for controls. Blood levels of EPA were lower in stroke patients compared to controls for both the unadjusted [t₍₆₎ = −4.44, p = 0.004] and age-adjusted [t₍₅₎ = −4.20, p = 0.009] model (Figure 7).

Six reports with a total of 631 stroke cases and 604 controls were included in the meta-analysis (Brattstrom et al., 1992; Williams et al., 2001; Kelly et al., 2003; Angelova et al., 2008; Rudreshkumar et al., 2018; van Wijk et al., 2023). The mean number of subjects was 105.2 (range: 20–180) in the stroke groups and 100.7 (range: 20–249) in the control groups. The mean age was 56.1 (range: 27.7–70.7) years for stroke cases and 53.6 (range: 27.0–68.4) years for controls. Blood levels of vitamin B6 were not different in stroke patients compared to controls; neither for the unadjusted [t₍₅₎ = −0.69, p = 0.52], nor for age-adjusted [t₍₄₎ = −0.62, p = 0.57] model (Figure 8).

Six reports with a total of 248 stroke cases and 342 controls were included in the meta-analysis (Hume et al., 1982; Sharpe et al., 1994; Singh et al., 1995; Cherubini et al., 2000; El Kossi and Zakhary, 2000; Sanchez-Moreno et al., 2004). The mean number of subjects was 41.3 (range: 10–110) in the stroke groups and 57.0 (range: 12–202) in the control groups. The mean age was 65.0 (range: 52.5–77.2) years for stroke cases and 62.0 (range: 46.2–77.8) years for controls. Blood levels of vitamin C were significantly lower in stroke patients compared to controls, with statistical significance for the unadjusted [t₍₅₎ = −3.36, p = 0.020] and a trend in the age-adjusted [t₍₃₎ = −2.49, p = 0.088] model (Figure 9).

Six reports with a total of 581 stroke cases and 12,030 controls were included in the meta-analysis (Chang et al., 1998; Angelova et al., 2008; Hu et al., 2019; Mironczuk et al., 2021; van Wijk et al., 2023). The mean number of subjects was 96.8 (range: 36–202) in the stroke groups and 2,005 (range: 21–6,983) in the control groups. The mean age was 63.0 (range: 46.9–70.8) years for stroke cases and 52.9 (range: 46.0–63.1) years for controls. Blood levels of selenium were lower in stroke patients compared to controls, for both unadjusted [t₍₅₎ = −3.47, p = 0.018] and age-adjusted [t₍₄₎ = −3.19, p = 0.033] model (Figure 10).

Four reports with a total of 358 stroke cases and 158 controls were included in the meta-analysis (Rashid et al., 2003; Hosinian et al., 2016; Szpetnar et al., 2016; Ercan et al., 2019). The mean number of subjects was 89.5 (range: 18–228) in the stroke groups and 39.5 (range: 12–60) in the control groups. The mean age was 71.0 (range: 64.0–75.3) years for stroke cases and 65.7 (range: 60.1–71.3) years for controls. Blood levels of arginine were not significantly different in stroke patients and controls; neither for unadjusted [t₍₃₎ = −0.10, p = 0.93], nor for age-adjusted [t₍₂₎ = −0.12, p = 0.92] model (Figure 11).

For five of the preselected nutrients, i.e., taurine, CoQ10, choline, carnitine, and uridine, an insufficient number of reports was found to perform the meta-analysis.

For taurine, two reports with a total of 188 stroke cases and 97 controls were selected (Castillo et al., 1996; Ghandforoush-Sattari et al., 2011). The mean number of subjects was 94.0 (range: 60–128) in the stroke groups and 48.5 (range: 43–54) in the control groups. The mean reported age was 68.0 years for stroke cases and 56.0 years for controls. In both studies blood levels of taurine in stroke patients were elevated and reached statistical significance in one out of two (p < 0.0001 vs. p = 0.170) studies.

For CoQ10, two reports with a total of 157 stroke cases and 117 controls were selected (Simani et al., 2018; van Wijk et al., 2023). The mean number of subjects was 78.5 (range: 76–81) in the stroke groups and 58.5 (range: 34–83) in the control groups. The mean age was 63.6 (range: 63.0–64.2) years for stroke cases and 62.0 (range: 60.8–63.1) years for controls. Both studies reported lower blood levels of CoQ10 in stroke patients compared to controls (p < 0.05 and p < 0.001).

For choline, two reports with a total of 145 stroke cases and 148 controls were selected (Lee et al., 2019; van Wijk et al., 2023). The mean number of subjects was 72.5 (range: 65–80) in the stroke groups and 74.0 (range: 65–83) in the control groups. The mean age was 63.7 (range: 63.2–64.2) years for stroke cases and 62.2 (range: 61.3–63.1) years for controls. Both studies reported lower blood levels of choline in stroke compared to controls (p < 0.002 and p < 0.008).

Only one report (van Wijk et al., 2023) showed that blood levels of uridine was 4.01 ± 1.28 μmol/L in 78 stroke cases and 4.17 ± 1.24 μmol/L in 82 controls (p > 0.1). The blood levels of total carnitine reported in the same study for 81 stroke cases and 82 controls were 49.2 ±13.6 and 51.3 ± 9.4 μmol/L (p > 0.1), respectively.