Our systematic review adheres to the methodological standards outlined in the Cochrane Handbook for Systematic Reviews,16 PRISMA guidelines,17 and AMSTAR 2 criteria.18 The study protocol was registered in PROSPERO (CRD42024549613). We conducted a comprehensive search of major databases, including MEDLINE (PubMed), Scopus, EMBASE, Web of Science, Science Direct, and the Cochrane Library from inception to 22 mayo 2024. Our search strategy incorporated both controlled vocabulary (e.g., MeSH, Emtree) and free terms, combined with Boolean operators, to implement our PECOS strategy (Population: adult patients with traumatic brain injury; Exposition: Reverse shock index multiplied by Glasgow Coma Scale (rSIG) < 14 points; Comparator: Reverse shock index multiplied by Glasgow Coma Scale (rSIG) > or = 14 points; Outcome: in-hospital mortality; Study design: Observational studies, case control, cohort, transversal studies). Keywords were carefully selected to encompass relevant exposures ("Reverse shock index multiplied by Glasgow Coma Scale” OR “rSIG") AND “traumatic brain injury”. A detailed outline of the search strategy is provided in the Supplementary Materials (refer to Table 1SA in supplementary material).

All articles identified during the primary and secondary screenings were cataloged using Zotero® 6.0.15. After removing duplicates, the documents were transferred to the Rayyan® platform. Two authors (GAVT and MCC) independently and blindly screened the titles and abstracts. Studies were selected by mutual agreement, with a third researcher (EDMR) mediating any disagreements. The selected articles then underwent a thorough full-text review to assess their eligibility. Additionally, reference lists and citations of the included studies were manually reviewed to identify any further relevant publications. For clarity, the selection process is outlined in Fig. 1.

Figure 1.

PRISMA flow diagram of the selection process of the primary studies included.

(0.48MB).

We included observational studies (case-control, cohort, or cross-sectional studies) that evaluated rSIG as a predictor of in-hospital mortality. The studies enrolled adult patients (≥18 years old) of both sexes. We considered articles published up to May 22, 2024, without restrictions on language or publication date. We excluded case reports, systematic reviews, and duplicates. Additionally, studies that did not report association measures, such as odds ratio (OR), but rather presented sensitivity, specificity, positive predictive value, and negative predictive value, were included.

The primary outcome was in-hospital mortality.

Two independent researchers collected and extracted relevant information from each included study using a standardized, blinded spreadsheet. This information included authors’ names, country and year of publication, clinical and epidemiological characteristics of the population, number of participants and cases, measures of association, confounding factors, and outcomes. For dichotomous variables (mortality), adjusted odds ratios (OR) with 95% confidence intervals (CI 95%) were collected based on events in exposed and non-exposed groups. Missing data were reported when appropriate. For continuous variables, we gathered the mean and standard deviation (SD) for both exposed and non-exposed groups. When studies presented these variables as medians and interquartile ranges (IQR), we converted them to means and SDs using appropriate methods.19,20 Some studies reported the strength of association, such as odds ratio (OR), while others provided diagnostic accuracy metrics like sensitivity (Se), specificity (Sp), positive likelihood ratio (LR+), negative likelihood ratio (LR−), and ROC curves for mortality prediction. However, we opted to conduct the meta-analysis on both the studies reporting OR and those presenting diagnostic accuracy metrics.

We employed the Inverse Variance (IV) method21 in the meta-analysis to combine adjusted odds ratios (ORs) and their 95% confidence intervals (CIs). This analysis was performed using R® 4.2.226 software. Forest plots were used to summarize the quantitative synthesis, utilizing the meta library, and the metabin22 (for dichotomous variables) and metacont23 (for continuous variables) functions. The IV method with DerSimonian-Laird (DL) for tau2 was applied. Heterogeneity among studies was assessed using Cochran’s Q test and Higgins I2 statistic. In cases of statistically significant heterogeneity (I2 statistics > 40%), a random effects model without Hartung-Knapp (HK) adjustment was employed. Sensitivity analysis was conducted using Influence Analysis with the InfluenceAnalysis function, and Graphic Display of Heterogeneity (GOSH) with the gosh.diagnostics function.24 Meta-regression was performed using variables that may clinically influence heterogeneity, conducted by “cutoff values” of rSIG and risk of bias (RoB) of studies. The metareg function was utilized, employing mixed-effects with REML as the method of estimation.25

We assessed the potential risk of bias using Newcastle-Ottawa Scale (NOS).26 Two researchers (GAVT and MCC) independently evaluated the certainty of the evidence (CoE) for each outcome based on the Grading of Recommendations Assessment, Development, and Evaluation (GRADE) criteria.27,28 Any discrepancies between reviewers were resolved through discussion with the lead researcher (EDMR).

Eight observational studies were included in the qualitative evaluation and meta-analysis. All studies were conducted in various countries and continents, including Brazil,29 China,30 Iran,31 Korea,14 South Korea,32 Taiwan33 and varios Asian countries.33–35 Approximately 430 000 patients with TBI were evaluated, among whom there were 6,417 deaths (1.5%). Of the nine studies included, six14,29–32,34 reported adjusted association measures (OR) for the meta-analysis, and other six29,30,32–35 too reported rSIG values as a numerical variable in both the mortality and survival groups due to TBI. Additional demographic characteristics of the study population are detailed in Table 1. The patients' site of care was in the intensive care unit (ICU) and emergency department (ED). Chen35 and Park's32 studies were based on data reports (PATOS and EDIIS), contributing a large number of patients to this meta-analysis. Additional demographic characteristics of the study population are detailed in Table 1.

The Newcastle-Ottawa Scale (NOS) was used to assess the risk of bias in the included observational studies. Of the eight studies included, only one, Carteri et al.,29 presented a moderate risk of bias, while the rest showed a low risk of bias.14,30–35 For further details, see Table 2.

The initial meta-analysis of six studies indicates that patients with TBI and low rSIG have a higher risk of mortality, with a 24% increase in the risk of death (OR 1.24; 95% CI 1.12–1.38; I²: 96%) (Fig. 2A); however, there was very high heterogeneity. When performing a sensitivity analysis using the InfluenceAnalysis function and GOSH, we identified Lin et al.30 and Wan-Ting et al.34 as outliers. After excluding these studies, the increase in the risk of death among those with elevated rSIG remained significant (OR 1.17; 95% CI 1.08–1.27; I²: 67%), with a reduction in heterogeneity to a moderate level (Fig. 2B).

Figure 2.

(A) Forest plot of low rSIG and mortality in adults with traumatic brain injury (TBI). All studies were included in the initial meta-analysis, which demonstrated high heterogeneity. After identifying outliers through Influence Analysis and GOSH, Lin et al.30 and Wan-Ting et al.34 were eliminated. (B) Forest plot after exclusion of outliers, illustrating the effect of low rSIG on mortality in adults with TBI, showing moderate heterogeneity.

(0.68MB).

When evaluating subgroups based on the assessed rSIG cutoff points (<14, <16.5 or <18) across all studies, we observed an increase in the risk of mortality only in cutoff rSIG <18 subgroup; however, high heterogeneity persisted (Fig. 3A). Conversely, when evaluating by subgroups according to risk of bias (RoB) and excluding the outliers, we found a reduction in heterogeneity within the low RoB subgroup, accompanied by an increased risk of mortality (OR 1.21; 95% CI 1.20–1.22; I²: 0%) (Fig. 3B). Meta-regression based on the cutoff variable did not reveal it as an explanation for the heterogeneity (cutoff β: −1.7; p: 0.05); however, it did demonstrate a significant role when evaluated by RoB (RoB β: −0.23; p: 0.03) (Figs. 1S and 2S of Supplementary Material)

Figure 3.

Forest plot by subgroups: (A) According to the rSIG cutoff point, heterogeneity remains high; however, statistical significance persists only in the group with rSIG <18. (B) According to risk of bias and outliers excluded, it shows that in the low risk of bias subgroup, there is an association between low rSIG and mortality in patients with TBI, with nule heterogeneity observed.

(1.01MB).

When evaluating rSIG as a continuous variable in both the deceased and survivor groups, we found that the survivor group had a higher mean difference (MD) compared to the deceased (MD 7.72; 95% CI 1.86–13.58; I²: 99%) (Fig. 4).

Figure 4.

Forest plot of rSIG as a numerical variable. It shows that patients with TBI who survive have higher values compared to those who died.

(0.37MB).

When performing meta-regression based on subgroups according to rSIG cutoff point (<14 vs. <18) and risk of bias (moderate vs. low), we found that these variables did not explain the heterogeneity (cutoff β: 4.7; 95% CI −8.3–17.2; p > 0.05 and RoB β: −8.04; 95% CI −23.8–7.7; p > 0.05, respectively)

Seven of the reported studies present results based on sensitivity (Se), specificity (Sp), likelihood ratio + (LR+) and likelihood ratio (LR−). When meta-analyzing these results, we found that rSIG, in predicting mortality in TBI, has a pooled sensitivity of 0.59 (95% CI 0.42–0.75; I2: 96%), a pooled specificity of 0.81 (95% CI 0.75–0.86; I²: 100%), a pooled LR+ of 3.179 (95% CI 2.178–4.63), a pooled LR− of 0.499 (95% CI 0.33–0.756), a pooled diagnostic OR of 6.365 (95% CI 3.05–13.282), and a false positive rate (FPR) of 0.187 (95% CI 0.136–0.251). Finally, the pooled diagnostic performance of rSIG, based on the area under the ROC curve, is 0.85. (Fig. 5).

Figure 5.

Forest plot showing the pooled sensitivity (A), specificity (B), and area under the ROC curve (C) for the predictive ability of low rSIG in mortality among TBI patients.

(0.34MB).

The subgroup evaluation of the various indices according to rSIG cutoff point shows better performance when the cutoff point is rSIG <14 vs. 18 points (pooled S, E, and diagnostic OR for rSIG <14 are: 0.55, 0.77, and 4.14, respectively, and for rSIG <18 they are 0.65, 0.86, and 10.86, respectively). However, in meta-regression, using this cutoff point does not explain the level of heterogeneity (estimate for Se: 1.17, p: 0.6, and estimate for Sp: 1.11, p: 0.1) (Fig. 2S of Supplementary Material).

We used GRADE to assess the certainty of evidence (CoE) regarding the association between rSIG and in-hospital mortality in TBI. Publication bias was not assessed due to the limited number of studies. Table 3 shows a 24% increase in the risk of death in TBI patients with low rSIG, with a moderate certainty of evidence.

Our study has several strengths. First, we implemented a comprehensive search strategy that encompassed six essential databases, facilitating a robust epidemiological evaluation of cause and effect. Second, we employed a rigorous methodology for our review and meta-analysis, which included a thorough quality assessment of the studies and a statistical analysis to address heterogeneity. Third, our results are both robust and reliable, as demonstrated by our careful evaluation of heterogeneity and identification of outliers, with consistent findings across individual studies. Fourth, this is the first systematic review and meta-analysis of sufficient quality that incorporates the most recent observational studies on rSIG based on Asian databases, enrolling a large number of patients.

However, it is important to acknowledge several limitations in our study. First, the primary limitation is that the studies included use different cutoff points for rSIG and mortality. To address this, we analyzed the pooled OR by subgroups and also examined it as a continuous numerical variable in both the mortality and survival groups. Second, only a limited number of completed studies directly addressed our PECO question, with the majority focusing on Asian populations, which may affect the generalizability of our findings. Third, the high heterogeneity observed among the included studies can largely be attributed to the risk of bias inherent in each study. Furthermore, the diverse clinical characteristics of patients and various confounding factors identified in the studies contribute to the overall high level of heterogeneity. Nevertheless, the finding of an increased risk of mortality remains consistent within the low risk of bias subgroup.