Three main electronic databases including PubMed, Web of Science, Embase, Wanfang, and China National Knowledge Infrastructure (CNKI) were used for literature search with a predefined combined search term including (1) “stress hyperglycemia” OR “stress hyperglycaemia” OR “stress-induced hyperglycemia” OR “stress induced hyperglycemia” OR “stress-induced hyperglycaemia” OR “stress induced hyperglycaemia” OR “admission hyperglycemia” OR “admission hyperglycaemia” OR “admission glucose” OR “stress hyperglycemia ratio” OR “stress hyperglycaemia ratio” OR “glycemic ratio” OR “stress-hyperglycaemia ratio” OR “stress-hyperglycemia ratio” OR “glycemic gap” OR “relative hyperglycemia” OR “acute-to-chronic glycemic ratio”; (2) “heart failure” OR “cardiac failure” OR “cardiac dysfunction”; and (3) “mortality” OR “death” OR “hospitalization” OR “rehospitalization” OR “prognosis” OR “survival” OR “major adverse cardiovascular events” OR “MACE”. Only studies with human subjects and published in English or Chinese were included. A manual screening of the citations from the pertinent articles and reviews (cross-references) was also carried out for possible eligible studies. The conclusive database search took place on February 12, 2024.

The inclusion criteria were made according to the PICOS principle

We excluded reviews, meta-analyses, studies with SHR analyzed as continuous only, or studies without outcomes of interest. In cases where there was potential overlap in patient population across multiple studies, only the study with the largest sample size was included in this analysis.

Two separate authors conducted a thorough search of academic literature, performed data collection and analysis, and independently assessed the quality of the studies. Any discrepancies that arose were resolved by involving the corresponding author in discussion for final decision-making. Data on study information, design, diagnosis of the patients, sample size, age, sex, and diabetic status of the patients, the cutoffs of SHR, follow-up durations, outcomes evaluated, and variables adjusted in the regression model for studying the association between SHR and clinical outcomes of patients with HF were gathered. The assessment of study quality was carried out using the Newcastle-Ottawa Scale (NOS) (35), which involved scoring based on criteria including participant selection process, comparability among groups, and validity of outcomes. This scale utilized a rating system ranging from 1 to 9 stars; higher stars indicated better study quality.

An association between SHR and the clinical outcomes of patients with HF was presented using risk ratio (RR) and corresponding 95% confidence interval (CI), compared between HF patients with high versus low SHR at baseline. For studies reporting odds ratio (OR), data were converted to RRs for the meta-analysis (RR=OR/([1−pRef]+[pRef×OR]), where pRef is the prevalence of the outcome in the reference group (a low SHR group) (36). Data of RRs and standard errors were calculated based on the 95% CIs or p values, followed by a logarithmical transformation to ensure stabilized variance and normalized distribution (34). We combined the log RR or log hazard ratios (HR) and corresponding standard errors by the inverse variance approach. The heterogeneity among studies was assessed using the Cochrane Q test and I² statistic (37, 38), with I² > 50% indicating significant statistical heterogeneity. A random-effects model was used for result aggregation considering the influence of heterogeneity (34). For the primary outcome of all-cause mortality, the sensitivity analysis by omitting one study at a time was performed to evaluate the robustness of the finding. For characteristics presented as the continuous variables, such as sample size, mean age, proportion of men, proportion of diabetic patients, cutoff of SHR, follow-up duration, and study quality scores, a univariate meta-regression analysis was also performed (34) to explore the influence of these variables on the outcome of the meta-analysis. For the continuous variables, medians were directly retrieved as means in the meta-regression analysis. Additionally, multiple subgroup analyses were performed to evaluate the influences of study characteristics on the results, such as in acute or chronic HF, in diabetic or non-diabetic patients, as well as subgroup analyses according to the cutoffs of SHR and follow-up durations of the included studies. Medians of continuous variables were selected as the cutoff values for defining subgroups. Publication bias estimation involved constructing funnel plots initially evaluated through visual inspection for symmetricity before being analyzed using Egger’s regression test (39), where a p < 0.05 indicates statistical significance. These analyses were conducted using the RevMan Version 5.1 (Cochrane Collaboration, Oxford, UK) and Stata software version 17 (Stata Corporation, College Station, TX).

The process of selecting relevant studies for inclusion in the meta-analysis is depicted in Figure 1 . Initially, 416 potentially pertinent records were identified through thorough searches of five databases. Among these, 79 were removed due to duplication. Subsequent screening based on the titles and abstracts resulted in the exclusion of an additional 310 studies that did not align with the aim of the meta-analysis. The full texts of the remaining 27 records underwent independent review by two authors, leading to the removal of a further 17 studies for various reasons detailed in Figure 1 . Ultimately, ten cohort studies remained (22–31) and were considered suitable for subsequent quantitative analyses.

Table 1 presents the summarized characteristics of the included studies. Overall, one prospective cohort (23) and nine retrospective cohort studies (22, 24–31) were included in the meta-analysis. These studies were published between 2021 and 2024, and performed in China, Spain, Portugal, and the United States. All of the studies included adult populations with HF. The mean ages of the patients were 58.2 to 83.0 years. Methods for defining the cutoff of SHR varied among the included studies, such as the use of the median (22), tertiles (23, 25, 26), quintiles of SHR (28), receiver operating characteristic analysis derived cutoffs (24, 27, 30, 31), or arbitrarily determined cutoff (29). The cutoff values for defining a high SHR varied from 0.82 to 1.75 among the included studies. The follow-up durations varied from within hospitalization to 48 months. The primary outcome of all-cause mortality was reported in ten cohorts (1984 events) (22–31), while the secondary outcomes of CV death (28, 30) were reported in two studies (227 events) (28, 30), HF rehospitalization in two studies (346 events) (28, 30), and MACE in three studies (360 events) (24, 25, 27), respectively. Multivariate analyses were used in all of the included studies when the association between SHR and the clinical outcomes of patients with HF was reported, and variables such as age, sex, hemodynamic parameters, comorbidities, ejection fraction (EF), and concurrent medications were adjusted to a varying extent among the included studies. The data of OR was reported in five studies (22, 24, 25, 27, 29), and the data of HR was reported in the other five studies (23, 26, 28, 30, 31). The NOS of the included studies were seven to nine stars, suggesting overall good study quality ( Table 2 ).

Since two studies reported the outcome according to the diabetic status of the patients (23, 26), and another study reported the outcome according to the EF of the patients (28), these datasets were included in the meta-analysis independently. Overall, pooled results of 14 datasets from ten cohort studies showed that compared to patients with lower SHR at baseline, those with a higher SHR were associated with an increased risk of all-cause mortality during follow-up (RR: 1.61, 95% CI: 1.17 to 2.21, p = 0.003; Figure 2A ) with significant heterogeneity (I² = 82%).

Sensitivity analysis by excluding one dataset at a time did not significantly change the results (p all < 0.05; Figure 2B ).

Further meta-regression analysis suggests that different in the cutoff of SHR significantly modified the association between SHR and all-cause mortality of patients with HF (coefficient = 1.22, p = 0.02; Table 3 ; Figure 2C ), but not for the other variables such as sample size, mean age, proportion of men, percentile of diabetic patients, follow-up duration, or study quality scores (p all > 0.05; Table 3 ).

Subgroup analysis suggested consistent association between a high SHR and an increased risk mortality in patients with acute HF and chronic/overall HF patients (p for subgroup difference = 0.14; Figure 3A ), and in patients with and without diabetes (p for subgroup difference = 0.86; Figure 3B ). Consistent to the results of meta-regression analysis, the subgroup analysis suggested a more remarkable association between SHR and all-cause mortality in studies with cutoff of SHR ≥ 1.05 than those with cutoff of SHR < 1.05 (RR: 2.29 versus 1.08, p for subgroup difference < 0.001; Figure 4A ). Subgroup analysis showed similar results in studies with follow-up duration < 12 months and ≥ 12 months (p for subgroup difference = 0.38; Figure 4B ).

Pooled results of three datasets from two (28, 30), two (28, 30) and three studies (24, 25, 27) respectively showed that a high SHR at baseline was related to the incidence of CV death (RR: 2.19, 95% CI: 1.55 to 3.09, p < 0.001; I² = 0%; Figure 5A ), HF-rehospitalization (RR: 1.83, 95% CI: 1.44 to 2.33, p < 0.001; I² = 0%; Figure 5B ), and MACE (RR: 1.54, 95% CI: 1.15 to 2.06, p = 0.004; I² = 74%; Figure 5C ) of patients with HF during follow-up.

The funnel plots for the meta-analysis of the association between SHR and all-cause mortality of HF are shown in Figure 6 . The symmetrical nature of the funnel plots suggested the low likelihood of publication bias. The Result of the Egger’s regression test also showed a low risk of publication bias (p = 0.98). The publication biases underling the meta-analyses for the three secondary outcomes could not be determined because only three datasets were included.