Prognostic value of glycated hemoglobin among patients with ST-segment elevation myocardial infarction: a systematic review and meta-analysis

Search strategies

Relevant citations were identified by searching several databases up to March 2016, including PubMed, Web of Science, Embase, Chinese national knowledge infrastructure (CNKI), Chinese biological medicine (CBM) and Wanfang databases. We used the following terms (‘glycated hemoglobin’, ‘glycosylated hemoglobin’, ‘hemoglobin A1c’ or ‘HbA_1c’), (‘ST-segment elevation myocardial infarction’, ‘ST elevation myocardial infarction’ or ‘STEMI’) and (‘mortality’, ‘death’, ‘outcome’, ‘prognosis’, ‘MACE’, ‘major adverse cardiovascular events’ or ‘major adverse cardiac events’) as our searching strategies. Searching results were further restricted to English or Chinese-language and human studies. The reference lists of candidate articles were carefully reviewed to identify relevant studies.

Criteria for study inclusion and exclusion

Citations were separately screened by two investigators (GXL and YL) for eligibility. Studies were included if they met the criteria as follows: 1) prospective studies that investigated the association between HbA_1c and the prognosis of STEMI patients; 2) reported rates of short- and long-term mortality after the diagnosis of STEMI; 3) provided sufficient data for the calculation of relative risk (RR) and its corresponding 95% confidence interval (CI). Studies were excluded according to the following criteria: 1) if the study subjects were non-STEMI patients; 2) conference articles were also excluded because they did not contain enough data to qualify for a meta-analysis; 3) when there were multiple publications from the same population, only the publication with the largest sample size was included.

Data extraction

Data were independently extracted by two authors (GXL and XWH) using a purpose-designed form. Any discrepancy would be resolved by discussing with the third author (JL). The following information was collected: first author, year of publication, geographic location, mean age, reperfusion strategy, diabetic status, sample size and cut-off value of HbA_1c, the clinical endpoints in low and high HbA_1c group, respectively.

Since a threshold of 6.5% HbA_1c level was most widely used in numbers of studies and was recommended by the International Expert Committee to diagnose diabetes mellitus in 2009 [17], a cut-off value of 6.5% was preferred. Alternative cut-off values were those closest to 6.5%. Clinical endpoints were all-cause mortality at different time points (in-hospital, 30-day and long-term). Seven-day mortality was incorporated into in-hospital mortality. In addition, 1-year mortality was treated as long-term endpoints.

Quality assessment

Newcastle-Ottawa Scale [18], a nine-star system, was applied to perform quality assessment by two researchers, separately. Studies that were awarded with seven or more stars were considered to be of high quality.

Statistical analysis

Summary estimates of the pooled RRs for the associations between HbA_1c level and risk of clinical outcomes among STEMI patients were combined using the inverse variance method. Between-study heterogeneity was checked using Cochran Q and I² statistics. It was considered to be statistically significant when the p-value was lower than 0.05 or I² was higher than 50% [19] Then a random-effect model was used for the calculation of pooled RR. Otherwise, it was thought to be insignificant and a fixed-effect model was chosen. To explain heterogeneity, subgroup analyses were conducted according to geographic location, reperfusion strategy, diabetic status, sample size (<500 or ≥500) and cut-off value for each outcome, if applicable. To assess the influence of individual studies on the pooled RRs, sensitivity analyses were conducted by sequentially excluding studies one by one. To further explore possible source of heterogeneity, meta-regression analyses were performed. The potential publication bias was examined using the adjusted rank correlation test and the regression asymmetry test, respectively [20], [21]. In case of publication bias, Copas selection model was applied to identify and correct the asymmetry in the funnel plot [22]. The main analyses were performed using Stata 12.0 (Stata Corporation, College Station, TX, USA). Copas selection model was conducted using R software (version 2.11.1). All statistical tests were two-tailed and p<0.05 was considered statistically significant.

Eligible studies

According to our initial search strategies, there were 312 citations retrieved from the electronic databases (Figure 1). After scanning the titles, abstracts, reading full texts and hand searching the reference lists of candidate articles, we finally got 19 prospective studies including a total of 35,994 STEMI patients to evaluate the associations between HbA_1c level and their in-hospital, 30-day and long-term mortality. Among these studies, there were 13 studies from Asia [3], [8], [10], [12], [13], [23], [24], [25], [26], [27], [28], [29], [30], 5 from Europe [11], [31], [32], [33], [34] and 1 from North America [35]. Detailed information of the eligible studies was shown in Table 1.

Figure 1:

Flow diagram showing the process for study selection.

Pooled analysis

Of the 19 studies in the current meta-analysis, 12 studies reported data on in-hospital mortality in 29,526 STEMI patients. Our results showed no significant association between HbA_1c level (high vs. low) and in-hospital mortality (RR 1.20, 95% CI 0.95–1.53, p=0.13) (Figure 2A). Data on 30-day mortality were available in eight studies, with 11,704 STEMI patients involved. Pooled effects showed a substantial difference in 30-day mortality between high and low HbA_1c level group (RR 1.25, 95% CI 1.03–1.52, p=0.02) (Figure 2B). Similarly, statistically significant association was observed between high HbA_1c level and risk of long-term mortality (RR 1.45, 95% CI 1.20–1.76, p<0.01), with 9472 STEMI patients from 10 studies enrolled (Figure 2C).

Figure 2:

Forest plots for the associations between HbA_1c level and (A) in-hospital mortality, (B) 30-day mortality and (C) long-term mortality among diagnosed with STEMI (high vs. low).

CI, confidence interval; HbA_1c, glycated hemoglobin; RR, relative risk; STEMI, ST-segment elevation myocardial infarction.

Subgroup analyses

As depicted in Table 2, subgroup analyses for in-hospital mortality based on geographic location, reperfusion strategy, diabetic status, sample size and cut-off value were in accordance with that of pooled analysis.

Subgroup analysis by geographic location showed that HbA_1c level was marginally associated with 30-day mortality among European STEMI patients (RR 1.30, 95% CI 0.99–1.73, p=0.06), but not in Asian STEMI patients (RR 1.25, 95% CI 0.97–1.60, p=0.08). Subgroup analysis categorized by reperfusion strategy did not show any significant association between HbA_1c and 30-day mortality among studies in which reperfusion strategy was unspecified. Similarly, the significant association could not be replicated among STEMI patients without previously known diabetes or from larger sample size studies. Variation in cut-off value might have substantial influence on the association between HbA_1c level and 30-day mortality (5.8 to 6.3%, RR 1.56, 95% CI 1.05–2.33, p=0.03; 6.5%, RR 1.15, 95% CI 0.92–1.45, p=0.22) (Table 2).

As for long-term mortality, results from subgroup analyses according to geographic location, reperfusion strategy and cut-off value were consistent with that in the pooled analyses. Nevertheless, our results indicated that the significant association was slightly affected by diabetic status (for diabetic patients, RR 1.46, 95% CI 0.96–2.22, p=0.07; for non-diabetic patients, RR 1.32, 95% CI 1.10–1.57, p<0.01; for patients with diabetic status unspecified, RR 1.90, 95% CI 0.95–3.80, p=0.07). When stratified by sample size, the association between HAb1c level and long-term mortality turn out to be insignificant among studies with sample size larger than 500 (RR 1.21, 95% CI 0.93–1.58, p=0.15).

Sensitivity analyses and meta-regression analyses

Sensitivity analyses were conducted by deleting studies one by one to assess the robustness of the results. There were no substantial variations in the pooled RRs after the removal of one study each turn, which confirmed that our results were stable and reliable (Figure 3A–C).

Figure 3:

Sensitivity analyses for the associations between HbA_1c level and (A) in-hospital mortality, (B) 30-day mortality and (C) long-term mortality among patients diagnosed with STEMI (high vs. low).

CI, confidence interval; HbA_1c, glycated hemoglobin; STEMI, ST-segment elevation myocardial infarction.

Meta-regression analyses showed that none of the following covariates (age, sex ratio, geographic location, reperfusion strategy, diabetic status, sample size and cut-off value) were significantly responsible for the between-study heterogeneity in the associations between HAb_1c level and in-hospital, 30-day and long-term mortality rates (Data not shown).

Publication bias

Judging from Begg’s correlation tests (in-hospital mortality: p=0.84; 30-day mortality: p=0.90; long-term mortality, p=0.86) and Egger’s regression tests (in-hospital mortality: p=0.47; 30-day mortality: p=0.91; long-term mortality, p=0.24), we found no evidence of publication bias (Table 3).

Study limitations

Some limitations of the current study should be noted. Firstly, as a meta-analysis of observational studies, selection bias could not be avoided. Secondly, though fully adjusted effect sizes were preferred during data collection, many studies reported data in terms of sample size and events and merely crude effect sizes were available. Differences in baseline characteristics (such as age, previous history of CAD and treatment details) between patients with and without a high HbA_1c were not adjusted, which was likely to confound our findings. Thirdly, hazard ratios may be more appropriate for the summary of long term outcomes. However, after carefully examining the included studies, most of them do not offer sufficient data for the calculation of hazardratios (HR) and few studies [33], [34] reported data in forms of hazard ratios. Therefore, we use RR as a substitute for HR in the summary estimate. Perhaps, future studies should report their results in more details and offer hazard ratios as well. Despite the limitations mentioned above, our meta-analysis significantly enhanced the statistical power by pooling substantial data from different studies. Furthermore, the robustness and reliability of our results were further validated by the sensitivity analyses.