Systemic inflammation response index as a clinical outcome evaluating tool and prognostic indicator for hospitalized stroke patients: a systematic review and meta-analysis

The systematic review and meta-analysis followed the PRISMA guidelines [18] and was registered on PROSPERO with the identifier CRD42023405221 (https //www.​crd.​york.​ac.​uk/​PROSPERO/​) [19]. Additional file 1: Table S1 contains the PRISMA checklist. PubMed was searched using the keywords (“Systemic inflammation response index” OR “System inflammation response index” OR “Systemic inflammatory response index” OR “SIRI”) AND (“Patients”). We used the identical search approach for Embase, Cochrane Library, Web of Science, and Scopus. Furthermore, we conducted a manual search in Chinese databases, such as China national Knowledge Infrastructure (CNKI), WanFang, VIP, and China Biology Medicine (CBM). To minimize selection bias, articles in both English and Chinese were taken into account during the search, which spanned from the beginning to February 12, 2023. Additional file 1: Table S2 presents the detailed search strategy.

We included studies that satisfied the following PICO criteria: (1) Population: individuals who have experienced a stroke, including IS and HS (ICH and SAH); (2) Intervention: mechanical thrombectomy, intravenous thrombolysis, surgical procedures (coiling or clipping), conservative treatment, or no treatment; (3) Comparisons: low SIRI vs. high SIRI; evaluating different SIRI values at different endpoints; (4) Outcomes: functional outcomes (measured by modified Rankin Scale [mRS] or Glasgow Outcome Score [GOS] at follow-up), mortality, predictive value of SIRI, SIRI values between poor and good outcomes, stroke-associated pneumonia (SAP) and non-SAP, early neurological deterioration (END) and non-END; SAH-associated clinical parameters between high SIRI and low SIRI, including Hunt-Hess Scale (HHS), modified Fisher Scale (mFS), delayed cerebral ischemia (DCI), vasospasm, and acute hydrocephalus (AHC). We did not include reviews, editorials, commentaries, case reports, letters to the editor, systematic reviews and meta-analyses, notes, replies, and conference abstracts because these types of records are insufficient for data.

Both reviewers (H Y-W and Z Y) individually examined the titles and abstracts of all the records that were obtained. Two reviewers independently assessed the relevant studies in their entirety and made decisions on article inclusion or exclusion according to the eligibility criteria. In case of discordance, the corresponding authors (L Z-P and Y X-S) would adjudicate.

Data were independently extracted into separate Excel spreadsheets by two reviewers, namely F C and A Y-H. To ensure accuracy, the source material and the spreadsheets were cross-checked with each other. Data collection included the first author's name, year of publication, country, study design, sample size, age, range, gender, stroke type, intervention type, SIRI cutoff (× 10⁹/L), primary and secondary endpoints, as well as the duration of follow-up. If any discrepancies were found, they were resolved by the corresponding author (L Z-P and Y X-S).

The primary outcome of this study was the assessment of functional outcomes, as measured by the mRS or GOS at follow-up. The definition of mRS and GOS is presented in Additional file 1: Table S3. The secondary outcomes included mortality, the predictive value of SIRI, SIRI values between poor/good outcomes, the SAP/non-SAP, and END/non-END. Additionally, the study analyzed the differences in HHS, mFS, DCI, vasospasm, and AHC between patients with low SIRI and high SIRI.

Two independent reviewers (H Y-W and F C) assessed the risk of bias of the included studies using the Newcastle–Ottawa Scale (NOS) tool [20] in a blind manner. The risk of bias summaries was then cross-checked, and any unresolved discrepancies were resolved by the corresponding author (LZ-P and YX-S).

We computed odds ratios (ORs) and their corresponding 95% confidence intervals (CIs) for binary variables. Continuous variables were used to calculate the mean difference (MD) along with their corresponding 95% CIs. If there is a substantial difference in the values of continuous variables, we employed the standard mean difference (SMD) for conducting meta-analysis. We extracted ORs and their corresponding 95% CIs from studies that had adjusted for confounding factors. The mean and standard deviation (SD) were estimated by utilizing the sample size, median, and interquartile range. These estimates were obtained using the optional estimation techniques described in McGrath et al.’s publication [21], which can be accessed at https://​smcgrath.​shinyapps.​io/​estmeansd/​. To consider the variation in clinical characteristics, we performed meta-analyses and subgroup analyses utilizing the random-effects approach if the heterogeneity exceeds 50%, or the fixed-effects approach if the heterogeneity is less than 50% [22]. When there were more than five studies included, subgroups analyses were conducted based on the sub-type stroke. Significant heterogeneity was assessed by conducting the Cochrane Q test (P < 0.1 or I² > 50%) [23]. Statistical significance was determined using a significance level of P < 0.05. Funnel plots were utilized to evaluate publication bias. The statistical analyses were conducted using Review Manager software (version 5.3.3), which can be found at https://​training.​cochrane.​org/​online-learning/​core-softwarecochrane​-reviews/​revman.

We acquired a total of 2435 publications using the search method on June 30, 2023. After eliminating 796 duplicates, we evaluated the remaining 1644 publications by their article type, title, and abstracts and we excluded 1620 publications that were not relevant. We thoroughly reviewed the remaining 24 publications for potential eligibility [9, 10, 14‐16, 24‐42]. Two studies [15, 40] shared almost the same data and were from the same author; thus, we combined the data and treated them as a single study. The exclusion of a study [35] was based on the absence of sufficient endpoints. In this systematic review and meta-analysis (Fig. 1), a total of 22 studies [9, 10, 14‐16, 24‐34, 36‐39, 41, 42] were ultimately incorporated.

The 22 studies included in this systematic review and meta-analysis were published between 2020 and 2023. Among them, 5 articles were prospective studies [25, 29, 31, 33, 42], and the remaining 17 articles [9, 10, 14‐16, 24, 26‐28, 30, 32, 34, 36‐39, 41] were retrospective studies. The studies were conducted in China (n = 19, two studies were from the MIMIC database), Italy (n = 1), and Korea (n = 2), with a total of 12,931 patients included. Two research studies employed a 1:1 propensity score matching (PSM) technique to equalize the impact of potential confounders, leading to the incorporation of 12,737 individuals in the analysis. 11 studies [10, 16, 24‐26, 28‐30, 32, 34, 39] focused on AIS, 11 studies focused on HS including 5 studies [9, 31, 33, 36, 42] focused on ICH, and 6 studies [14, 15, 27, 37, 38, 41] focused on SAH. The SIRI cutoff range was between 0.77 and 6.48 (× 10⁹/L), while the duration of follow-up varied from hospitalization to one year post-discharge. Table 1 provides a summary of the findings from the studies that were included.

Two studies [15, 38] reported functional outcomes assessed by GOS. The meta-analysis showed that individuals with high SIRI had a 3.17-fold higher risk of poor outcomes compared to those with low SIRI (odds ratio [OR] 3.17, 95% confidence interval [CI] 1.51–6.65, P = 0.002, I² = 0%, Fig. 2A), and the SIRI value was 0.72 higher in those with poor outcomes compared to those with good outcomes (standard mean difference [SMD] 0.72, 95%CI 0.47–0.97, P < 0.00001, I² = 42%, Fig. 2B). The predictive value of SIRI for poor outcome was 0.72 with a 95%CI of 0.63 to 0.82, P < 0.00001, and I² = 54% (Fig. 2C). After combining with clinical data, the predictive value for poor outcome was 0.88 with a 95%CI of 0.83 to 0.94, P < 0.0001, and I² = 55% (Fig. 2D), indicating that SIRI had a reasonably good predictive accuracy and a potential predictive ability. The results are summarized in Table 2.

Eight studies [9, 14, 16, 24, 25, 29, 39, 41] reported the SIRI values between good and poor outcome group, and the SIRI values were found to be 0.61 higher than that in good outcome with a 95% CI of 0.52 to 0.69, P < 0.00001, and I² = 60% (Fig. 3A). 12 studies [9, 14, 16, 24, 25, 27‐29, 31, 34, 39, 41] assessed functional outcomes using the mRS scale and reported the ORs and 95% CIs for SIRI and poor outcome, with 2 studies [24, 41] considering SIRI as both a continuous and dichotomous variable. The meta-analysis of 7 studies [9, 24, 25, 27, 29, 39, 41] considering SIRI as a continuous variable showed that for each standard deviation increase in SIRI, the risk of poor outcome increased by 20% (OR 1.20, 95% CI 1.07–1.34, P = 0.001, I² = 66%, Fig. 3B). The meta-analysis of 7 studies [14, 16, 24, 28, 31, 34, 41] considering SIRI as a dichotomous variable showed that high SIRI was associated with a higher risk of poor outcome compared to low SIRI (OR 3.01, 95% CI 2.00–4.54, P < 0.0001, I² = 74%, Fig. 3C). The predictive value of SIRI for poor outcome was 0.72 with a 95% CI 0.69 to 0.76, P < 0.00001, and I² = 78% (Fig. 3D). The results are summarized in Table 2.

In summary, despite the use of different assessment tools for poor outcome, it was consistently found that high SIRI was strongly associated with poor outcomes. In other words, there was a significant correlation between high SIRI and poor outcome.

Four studies [9, 10, 30, 31] reported mortality rates ranging from in-hospital to 1 year after discharge. The meta-analysis showed that a high SIRI was associated with a 1.68-fold increased risk for in-hospital mortality (OR 1.68, 95% CI 1.43–1.97, P < 0.00001, I² = 0%, Fig. 4A), a 1.50-fold increased risk for 1-month mortality (OR 1.50, 95% CI 1.14–1.98, P = 0.004, I² = 85%, Fig. 4B), a 1.77-fold increased risk for 3-month mortality (OR 1.77, 95% CI 1.53–2.04, P < 0.00001, I² = 0%, Fig. 4C), and a 1.65-fold increased risk for 1-year mortality (OR 1.65, 95% CI 1.43–1.92, P < 0.00001, I² = 1%, Fig. 4D) when compared to those with low SIRI. The results are summarized in Table 2.

Three studies [32, 33, 42] reported the SAP. The SIRI value of SAP was increased by 3.24 than non-SAP with 95% CI 1.56 to 4.91, P = 0.0002 and I² = 88% (Fig. 5A). 4 studies [32, 33, 38, 42] reported the ORs and 95CIs for SAP, in which one study [32] regarded the SIRI values as continuous variable and dichotomous variable. Three studies [32, 33, 42] regarded the SIRI value as continuous variable and the meta-analysis showed that for each standard deviation increase in SIRI, the risk of SAP increased by 11% (OR 1.11, 95% CI 1.05–1.18, P = 0.0006, I² = 66%, Fig. 5B). Two studies [32, 38] regarded the SIRI value as dichotomous variable and the meta-analysis showed that high SIRI had 2.89-folds risk for SAP comparing low SIRI (OR 2.89, 95% CI 2.23–3.75, P < 0.00001, I² = 0%, Fig. 5C). One study [33] randomized patients into the training and validation cohorts, and the two cohorts were regarded as two independent studies. The predictive value of SIRI for SAP was 0.81 with 95%CI ranged from 0.74 to 0.89, P < 0.00001, I² = 90% (Fig. 5D). The results are summarized in Table 2.

Two studies [26, 36] provided data on END. The SIRI value of END was found to be 0.37 higher than that of non-END with a 95% CI of 0.34 to 0.40, P < 0.00001 and I² = 0% (Fig. 6A). However, the meta-analysis revealed that high SIRI did not significantly increase the risk of END compared to low SIRI (OR 1.78, 95% CI 0.95–3.34, P = 0.07, I² = 85%, Fig. 6B). The results are summarized in Table 2.

Five studies [14, 15, 27, 38, 41] investigated the association between SIRI and SAH-related clinical parameters. The meta-analysis indicated that high SIRI was usually associated with higher scores for HHS (OR 2.70, 95% CI 1.45–5.01, P = 0.002, I² = 67%, Fig. 7A), mFS (OR 2.99, 95% CI 1.57–5.70, P = 0.0009, I² = 77%, Fig. 7B), increased risk of DCI (OR 3.09, 95% CI 2.16–4.43, P < 0.00001, I² = 0%, Fig. 7C), and vasospasm (OR 1.67, 95% CI 1.28–2.17, P = 0.0001, I² = 79%, Fig. 7D) compared to low SIRI. However, the risk of AHC (OR 1.90, 95% CI 0.84–4.29, P = 0.12, I² = 81%, Fig. 7E) was not statistically significant between the two groups. It is noteworthy that HHS, mFS, DCI, vasospasm, and AHC are all indicators of SAH severity, indicating that high SIRI was associated with more severe SAH. In regions with limited medical resources and where CT scans are not readily available, this simple index may prove valuable in predicting SAH severity and patient stratification. The results are summarized in Table 2.

Subgroup analyses were conducted based on the sub-type of stroke (IS and HS) for (i) the difference in SIRI values between the poor outcome group and the good outcome group, (ii) predicting poor outcome when SIRI was regarded as a continuous variable or dichotomous variable, and (iii) the predictive value of SIRI for poor outcome. Subgroup analysis demonstrated that the SIRI values were higher in the poor outcome group than in the good outcome group for both IS (SMD: 0.62; 95% CI 0.49–0.75, P < 0.00001, I² = 14%) and HS (SMD: 0.65; 95% CI 0.33–0.67, P < 0.00001, I² = 84%) (Fig. 8A). When SIRI was regarded as a continuous variable, subgroup analysis demonstrated that for each standard deviation increase in SIRI, the risk of poor outcome increased by 19% for IS (OR: 1.19; 95% CI 1.04–1.37, P = 0.01, and I² = 50%), whereas no statistically significant difference was found for HS (OR: 1.28; 95% CI 0.94–1.74, P = 0.12, and I² = 83%) (Fig. 8B). Similarly, when SIRI was regarded as a dichotomous variable, subgroup analysis demonstrated that the risk of a poor outcome at a high SIRI level was 3.73 times greater than that at a low SIRI level for IS (OR: 3.73; 95% CI 2.19–6.34, P < 0.00001, and I² = 74%) and 2.04 times greater for HS (OR: 2.04; 95% CI 1.50–2.77, P < 0.00001, I² = 0%) (Fig. 8C). Lastly, the predictive value of SIRI for poor outcomes was 0.72 for IS (AUC: 0.72; 95% CI 0.67–0.76) and 0.73 for HS (AUC: 0.73; 95% CI 0.67–0.80) (Fig. 8D). The results are summarized in Table 3.

The NOS has assessed and awarded a median of 8 stars to all the research, with an inter-quartile range of 5 to 9 stars. The methodological quality of the studies included can be found in Additional file 1: Table S4. Additionally, the probability of publication bias was evaluated through funnel plot results, which are displayed in Additional file 1: Figure S1.

While our study provides important insights into the association between SIRI and stroke patient outcomes, it is important to acknowledge several limitations. Firstly, due to the nature of inflammation response in stroke, most of the existing literature on this topic comprises retrospective studies, which may introduce limitations in terms of sample size, confounding variables, and selection bias. Secondly, with the exception of four prospective studies, the majority of studies included in our analysis were retrospective, resulting in considerable heterogeneity in data reporting and follow-up protocols. Therefore, further high-quality prospective studies are needed to confirm the validity and generalizability of our findings. Thirdly, based on our systematic review, the majority of included studies (86%, 19 out of 22 studies) were carried out in China, with two studies from the MIMIC database. As we know, the MIMIC database was established by the Beth Israel Deaconess Medical Center (Boston, MA, USA), and the population consisted mainly of US citizens. Therefore, these two studies reflected the relationship between SIRI and clinical outcomes in Americans. But the existing literature still lacks related studies in Europe or Africa. The broader applicability of SIRI as a predictive tool for stroke outcomes should be identified further in other ethnicities and countries. Fourthly, the high heterogeneity observed in some of our endpoints could influence the robustness of our results. Fifthly, some results are not mirrored to the total population of our studies selected, for each variable evaluated a different lesser number of studies were included. Hence, some findings are less robust. Despite these limitations, our meta-analysis provides valuable preliminary findings that could assist clinicians in making informed treatment decisions for stroke patients. Future research should aim to address these limitations and provide further insights into the association between SIRI and stroke outcomes.