Changes over Time in Hemoglobin A1C (HbA1C) Levels Predict Long-Term Survival Following Acute Myocardial Infarction among Patients with Diabetes Mellitus

Frequent fluctuations of hemoglobin A1c (HbA1C) values predict patient outcomes. However, data regarding prognoses depending on the long-term changes in HbA1C among patients after acute myocardial infarction (AMI) are scarce. We evaluated the prognostic significance of HbA1C levels and changes among diabetic patients (n = 4066) after non-fatal AMI. All the results of HbA1C tests up to the 10-year follow-up were obtained. The changes (∆) of HbA1C were calculated in each patient. The time intervals of ∆HbA1C values were classified as rapid (<one year) and slow (≥one year) changes. The outcome was all-cause mortality. The highest mortality rates of 53.8% and 35.5% were found in the HbA1C < 5.5–7% and ∆HbA1C = −2.5–(−2%) categories. A U-shaped association was observed between HbA1C and mortality: adjOR = 1.887 and adjOR = 1.302 for HbA1C < 5.5% and ≥8.0%, respectively, as compared with 5.5–6.5% (p < 0.001). Additionally, ∆HbA1C was associated with the outcome (U-shaped): adjOR = 2.376 and adjOR = 1.340 for the groups of <−2.5% and ≥2.5% ∆HbA1C, respectively, as compared to minimal ∆HbA1C (±0.5%) (p < 0.001). A rapid increase in HbA1C (but not decrease) was associated with a greater risk of mortality. HbA1C values and their changes are significant prognostic markers for long-term mortality among AMI-DM patients. ∆HbA1C and its timing, in addition to absolute HbA1C values, should be monitored.


Introduction
The global burden of diabetes mellitus (DM), a major cardiovascular risk factor, has increased dramatically throughout the last two decades [1,2]. Acute myocardial infarction (AMI) is a leading cause of increased mortality among patients with DM [3][4][5][6][7]. Furthermore, DM is a significant risk factor for short-and long-term morbidity and mortality among patients presenting with AMI [8,9]. Glucose and glycated hemoglobin (HbA 1C ) levels upon admission with AMI were associated with long-term mortality among patients with and without DM [10]. Thus, one of the cornerstones of treatment is glycemic monitoring and control, usually monitored by serial HbA 1C levels [11], based on evidence from several trials reporting improved outcomes with lowering of HbA 1C levels [4,12,13].
Recently, glycemic variability emerged as an additional and possibly even superior predictor of diabetic complications than mean HbA 1C , with frequent fluctuations independently associated with poor prognosis. In contrast, more stable HbA 1C control may benefit patients [14][15][16][17][18][19][20][21]. In addition, Lee et al. recently reported that long-term mean HbA 1C was significantly associated with long-term survival following AMI [22]. However, data regarding long-term HbA 1C changes following AMI and the association with the survival

Data Sources and Classifications
The baseline data were obtained from the electronic medical records of SUMC and included: patient demographics, comorbidities, and results of laboratory tests and echocardiographic, angiographic and revascularization procedures, as previously reported for the Soroka Acute Myocardial Infarction (SAMI) project [23]. The diagnosis of AMI was based on the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) codes: ST-segment elevation AMI (STEMI) 410.0 *-410.6 * and non-ST-segment elevation AMI (NSTEMI) 410.7 *-410.9 *. The DM definitions (and characteristics) were as follows: ICD-9-CM 250 * and/or HbA 1C ≥ 6.5% throughout six months before or after hospitalization. The diagnosis of anemia was based on low hemoglobin blood levels (for males < 13 g/dL; for females < 12 g/dL) during the index hospitalization. Patients were defined as having significant renal failure if they were either on hemodialysis or had an estimated glomerular filtration rate (eGFR) < 30 mL/min/1.73 m 2 . Either the ICD-9-CM code defined dyslipidemia, or this was given if the low-density lipoproteins (LDLs) were ≥100 mg/dL. Severe left ventricular (LV) dysfunction was defined as an ejection fraction < 30%.

Follow-Up and the Study Endpoint
Follow-up started after discharge and continued up to 10 years after that, or until 1 July 2020. The outcome was all-cause mortality. The survival status or death date for each patient was obtained from the Ministry of the Interior's population registry.

Statistical Analysis
Statistical analysis was performed using IBM SPSS 26 (SPSS Inc., Chicago, IL, USA) software. Patient characteristics were presented as mean and standard deviation (SD) for continuous variables and numbers (n) and as percentages (%) for the categorical data. A visual inspection of the continuous variables showed that their distribution was "normal". Comparison of the baseline characteristics and the outcomes was performed using the chisquare test, the chi-square test for linear trend of categorical variables and Student's t-test for continuous variables. A parametric test was used based on the central limit theorem and large sample size. The measure of association between HbA 1C levels during the follow-up and the risk of mortality was assessed using generalized estimating equation (GEE), binary logistic type, which considered the repeated measures of HbA 1C for each patient. Three different models were built: (1) HbA 1C levels alone, (2) ∆HbA 1C levels alone and (3) HbA 1C and ∆HbA 1C levels, with the time from hospital discharge and the baseline characteristics as the potential confounders, in order to assess the disparity of the associations between ∆HbA 1C values and outcomes by the groups of time intervals of ∆HbA 1C (rapid vs. slow change). This analysis included a multivariate model (GEE), which included the interactive variable of ∆HbA 1C by time interval. The reference categories for these models were: HbA 1C 5.5-6.5% and the minimal category of HbA 1C change (−0.5-0.5%). The results of the models are presented as adjusted odds ratios (adjORs) and 95% confidence intervals (CIs) for ORs. For each test, p < 0.05 was considered statistically significant.

Results
A total of 4066 eligible DM patients were analyzed in the current study, with a mean age of 66.4 ± 11.9 years; 36% were females. The median follow-up (after the first year) was 6.1 years, with 1802 (44%) patients dying throughout the follow-up period. The baseline characteristics of the study cohort and comparisons (at the index date) between patients who survived and those that died are presented in Table 1. Patients who died were older and more likely to be women, with higher rates of prior AMI, prior coronary artery bypass graft (CABG), congestive heart failure, renal failure and peripheral arterial disease than the survivors. Furthermore, they had an increased rate of DM-associated target-organ complications compared with the survivors. However, these patients had a lower rate of dyslipidemia, obesity, smoking and family history of early coronary artery disease (CAD) than the survivors. Besides, patients who died had a higher prevalence of non-cardiovascular comorbidity and were more likely to present with NSTEMI (rather than STEMI). Those who died had severe left ventricular dysfunction and were less likely to undergo revascularization due to the index AMI compared to the survivors. The data are presented as n (%) unless otherwise stated. Abbreviations: AMI-acute myocardial infarction, AV-Atrioventricular, CABGcoronary artery bypass graft, CAD-coronary arteries disease, CHF-congestive heart failure, CIHD-chronic ischemic heart disease, COPD-chronic obstructive pulmonary disease, HbA 1C -hemoglobin A1C, IHD-ischemic heart disease, GI-Gastrointestinal, LM-left main (coronary artery), LV-left ventricular, MI-myocardial infarction, PCI-percutaneous coronary intervention, PVD-peripheral vascular (arterial) disease, SD-standard deviation, s/p-status post (history of), STEMI-ST-segment elevation myocardial infarction.
A total of 35,609 HbA 1C test results were documented among the study cohort during the follow-up. HbA 1C values ranged between 3.7% and 18.6%, with a mean of 7.81 ± 1.74%. The mean HbA 1C was lower among patients who died vs. survivors (7.75 ± 1.82% vs. 7.84 ± 1.69%, respectively, p < 0.001). The mean values of HbA 1C throughout the followup (by follow-up year) among patients who died and among survivors are presented in Figure A1 (Appendix A). A trend of decrease in HbA 1C was observed among patients who died, while values remained similar or even increased over time among survivors; changes were most evident after five years of follow-up.
The mean HbA1C was lower among patients who died vs. survivors (7.75 ± 1.82% vs. 7.84 ± 1.69%, respectively, p < 0.001). The mean values of HbA1C throughout the follow-up (by follow-up year) among patients who died and among survivors are presented in Figure  A1 (Appendix A). A trend of decrease in HbA1C was observed among patients who died, while values remained similar or even increased over time among survivors; changes were most evident after five years of follow-up.
Regarding ∆HbA1C (Figure 1b), the risk of mortality increased with negative ∆HbA1C and remained similar with positive ∆HbA1C. A decrease of >2.5% in HbA1C was associated with a higher risk of mortality, as compared with the group with the minimal change (±0.5%): 35.3% vs. 26.7%, adjOR 1.495 (95%CI: 1.355-1.650, p < 0.001). Following multivariable adjustment for potential confounders, the association of HbA1C and mortality presented a U-shape-like association, with increasing risk with both higher and lower HbA1C (more prominent; Figure 2a). A similar trend was found between ∆HbA1C and mortality (Figure 2b). Regarding ∆HbA 1C (Figure 1b), the risk of mortality increased with negative ∆HbA 1C and remained similar with positive ∆HbA 1C . A decrease of >2.5% in HbA 1C was associated with a higher risk of mortality, as compared with the group with the minimal change (±0.5%): 35.3% vs. 26.7%, adjOR 1.495 (95%CI: 1.355-1.650, p < 0.001).
Following multivariable adjustment for potential confounders, the association of HbA 1C and mortality presented a U-shape-like association, with increasing risk with both higher and lower HbA 1C (more prominent; Figure 2a). A similar trend was found between ∆HbA 1C and mortality (Figure 2b). Figure 1. (a) The rates and the relative risks of mortality by HbA1C (adjusted odds ratios and 95% CI) categories; (b) The rates and the relative risks of mortality (adjusted odds ratios and 95% confidence intervals) by HbA1C categories. (a,b) are the results of separate models. Abbreviations: adjOR-adjusted odds ratio (adjusted for repeated measures), CIconfidence interval, HbA1C-hemoglobin A1C, ∆HbA1C-the difference between two HbA1C tests (later-earlier).
Following multivariable adjustment for potential confounders, the association of HbA1C and mortality presented a U-shape-like association, with increasing risk with both higher and lower HbA1C (more prominent; Figure 2a). A similar trend was found between ∆HbA1C and mortality (Figure 2b).  The mortality rates according to the HbA1C changes in the rapid-and slow-changing groups are presented in Figure A2 (Appendix A). Figure 3 displays the adjORs for mortality according to ∆HbA1C in the rapid-and slow-changing groups (adjusted to baseline characteristics, HbA1C levels and time lag from hospital discharge). It is evident that when HbA1C increases rapidly, it carries a greater risk of mortality than a slow increase. However, the rate of decreasing HbA1C has the same risk of mortality irrespective of the time of change.   The mortality rates according to the HbA 1C changes in the rapid-and slow-changing groups are presented in Figure A2 (Appendix A). Figure 3 displays the adjORs for mortality according to ∆HbA 1C in the rapid-and slow-changing groups (adjusted to baseline characteristics, HbA 1C levels and time lag from hospital discharge). It is evident that when HbA 1C increases rapidly, it carries a greater risk of mortality than a slow increase. However, the rate of decreasing HbA 1C has the same risk of mortality irrespective of the time of change.

Discussion
The current study evaluated the prognostic significance of the levels of and changes in HbA1C in patients with DM following non-fatal AMI. The main findings include: (1) the adjusted risk of mortality increases at HbA1C values ≤ 5.5% and ≥7%; (2) ∆HbA1C is a significant independent prognostic marker for the risk of mortality following AMI with a U-shaped association (stronger association with mortality for a decrease in HbA1C); (3) rapid versus slower increase in HbA1C following AMI is associated with a greater risk of mortality. However, the time interval of decrease in HbA1C does not seem to be a significant determinant of mortality risk.
The finding that HbA1C levels are associated with long-term mortality of AMI patients is overall consistent with previous studies of AMI patients [19,24,25]. Other studies often used admission HbA1C values while we evaluated all values (not necessarily upon admission). The association between low and high HbA1C and ∆HbA1C with longterm mortality following AMI is consistent with a recent study by Lee et al. [22] that evaluated the association of long-term mean HbA1C with long-term survival following AMI. They reported that patients with a mean HbA1C value ≥ 6.0% to <7.5% had a better prognosis than patients with HbA1C values of <6.0% or >7.5%. However, in our study, we did not use mean values and quantified the change and the trajectory of the change more elaborately, which is more informative.
Our findings are also consistent, in some parts, with those of Olsson et al. [5] who evaluated HbA1C as an AMI predictor in a large population of patients with type-2 DM and compared three distinct methods using HbA1C measurements taken from diabetes diagnosis and beyond (baseline-, latest-and mean-HbA1C). The authors found that the baseline HbA1C variable was a much weaker predictor of AMI than the updated latest and  year. The numbers represent the p-values for disparity (p for interaction) between the rapid-and slow-changing HbA 1C (separately for each ∆HbA 1C category). Adjusted for: repeated measures, HbA1C levels, time from hospital discharge, age, sex, supraventricular arrhythmias, congestive heart failure, chronic ischemic heart disease, history of myocardial infarction, history of percutaneous coronary intervention, history of a coronary artery bypass graft, atrioventricular block, renal failure, diabetes mellitus, hypertension, peripheral vascular disease, chronic obstructive pulmonary disease, neurological disorders, malignancy, anemia, gastrointestinal bleeding, alcohol/drug addiction, type of acute myocardial infarction, severe left ventricular dysfunction, left ventricular hypertrophy and three-vessel/left-main coronary arteries disease. Abbreviations: AdjOR-adjusted odds ratio, HbA 1C -hemoglobin A 1C , HbA 1C change (∆HbA 1C )-the difference between two HbA 1C tests (later-earlier).

Discussion
The current study evaluated the prognostic significance of the levels of and changes in HbA 1C in patients with DM following non-fatal AMI. The main findings include: (1) the adjusted risk of mortality increases at HbA 1C values ≤ 5.5% and ≥7%; (2) ∆HbA 1C is a significant independent prognostic marker for the risk of mortality following AMI with a U-shaped association (stronger association with mortality for a decrease in HbA 1C ); (3) rapid versus slower increase in HbA 1C following AMI is associated with a greater risk of mortality. However, the time interval of decrease in HbA 1C does not seem to be a significant determinant of mortality risk.
The finding that HbA 1C levels are associated with long-term mortality of AMI patients is overall consistent with previous studies of AMI patients [19,24,25]. Other studies often used admission HbA 1C values while we evaluated all values (not necessarily upon admission). The association between low and high HbA 1C and ∆HbA 1C with long-term mortality following AMI is consistent with a recent study by Lee et al. [22] that evaluated the association of long-term mean HbA 1C with long-term survival following AMI. They reported that patients with a mean HbA 1C value ≥ 6.0% to <7.5% had a better prognosis than patients with HbA 1C values of <6.0% or >7.5%. However, in our study, we did not use mean values and quantified the change and the trajectory of the change more elaborately, which is more informative.
Our findings are also consistent, in some parts, with those of Olsson et al. [5] who evaluated HbA 1C as an AMI predictor in a large population of patients with type-2 DM and compared three distinct methods using HbA 1C measurements taken from diabetes diagnosis and beyond (baseline-, latest-and mean-HbA 1C ). The authors found that the baseline HbA 1C variable was a much weaker predictor of AMI than the updated latest and updated mean HbA 1C variables, which in our study, are equivalent to the ∆HbA 1C . In addition, they reported a J-shaped association between the updated latest HbA 1C variable and AMI with increased risk associated with HbA 1C < 6.0% and >7.0% (reference 6.0-7.0%). Furthermore, a recent secondary analysis from the Action to Control Cardiovascular Risk in Diabetes (ACCORD) trial [26] showed that substantial changes in HbA 1C were significantly associated with a higher risk of heart failure: hazard ratio (HR) 1.32 (95% CI: 1.08-1.75) for ≥10% decrease in HbA 1C and 1.55 (95% CI: 1.19-2.04) for ≥10% increase in HbA 1C (reference < 10% change in HbA 1C ). Among ambulatory DM patients, high HbA 1C variability was associated with increased risk of all-cause and cardiovascular mortality and DM complications [21].
Interestingly, Carson et al. [27] reported a U-shaped association with HbA 1C levels, more robust in the lower categories (like our study), associated with increased all-cause mortality among US adults without diabetes. The reasons for the increased risk of mortality associated with low levels of HbA 1C are unknown. However, it has been suggested that patients in the lowest category (HbA 1C < 6.0%) may represent a frail and vulnerable group. Lower glycemic levels can be attributed not only to intensive glycemic control. Other factors such as old age and comorbidities may cause low HbA 1C [22,[27][28][29][30]. In addition, increased oxidative stress and endothelial dysfunction, which are markers of worse outcomes, were reported to be associated with significant glycemic changes [31,32]. Furthermore, a confounding-rather than causal-relationship may explain the observed association between HbA 1C changes and reduced long-term survival. These include poor compliance with medications, lifestyle recommendations, comorbidity, lack of support and infections [15,[33][34][35][36].
Nevertheless, Lee et al. [21] recently found an association between hypoglycemic frequency, HbA 1C variability and mortality. They concluded that these associations suggest that intermittent hypoglycemia may result in poorer outcomes in diabetic patients. Interestingly, we found that rapid versus slow increase in HbA 1C following AMI is associated with greater risk of mortality. Nevertheless, when a decrease in HbA 1C occurred, its rate had a minimal impact. This could be explained by an association between the rate of changes in HbA 1C and the magnitude of the deleterious consequences of DM (and its various complications) on patients following AMI. Furthermore, it is possible that a rapid increase in HbA 1C results from a systemic illness (e.g., infection or inflammation) or a stressor that activates counterregulatory hormones and additional mechanisms that support insulin intolerance and hyperglycemia, which are significant and long enough to cause a rapid increase in HbA 1C [37]. Additionally, various medications may signal more complicated cardiac disease (e.g., hydrochlorothiazides, beta-blockers) or systemic illness (e.g., corticosteroids) and may cause an increase in glucose levels, explaining this association [38]. Finally, a decrease of HbA 1C may occur due to other causes such as hypoglycemia (often associated with older age and other comorbidities), frailty and comorbidities (e.g., chronic inflammation, liver function derangement), rather than related to DM itself, thus potentially explaining the relatively weak association between the rate of decrease in HbA 1C and mortality [21,22,28]. Moreover, changes in HbA 1C levels could result from false decreases/increases, occurring due to various clinical situations and diseases such as iron deficiency anemia, deficiency of vitamin B12 or folic acid, severe hypertriglyceridemia, severe hyperbilirubinemia, chronic salicylate ingestion, chronic opioid ingestion, lead poisoning, acute or chronic blood loss, splenomegaly, pregnancy, vitamin E ingestion, red blood cell transfusion, ribavirin and interferon-alpha use, hemoglobin variants, vitamin C ingestion, uremia, hemodialysis and erythropoietin injections [39]. Although such clinical situations and diseases could, on the one hand, bias our findings, they could on the other hand (mostly the diseases) explain the observed increased mortality associated with such fluctuations.

Limitations
First, this is a non-randomized retrospective observational study; hence, it has the limitations of such a design and is subject to bias; mainly, it cannot show causality. Second, data regarding therapeutic interventions in general, and particularly some medications for DM treatment, were not available and could be a source of bias, particularly since it is currently known that the way in which glucose levels are reduced has a significant effect on the ultimate outcome of diabetic patients [40]. Third, the HbA 1C tests were taken at the discretion of the treating physicians rather than at prespecified intervals. Fourth, we used all-cause mortality rather than cause-specific mortality (e.g., cardiovascular). Fifth, data regarding multiple factors that could bias HbA 1C levels were not included.

Conclusions
The study shows for the first time that among DM patients with non-fatal AMI, longterm changes in HbA 1C values and the time interval for these changes (when increasing) are significant independent prognostic markers for long-term all-cause mortality with a U-shaped association. Thus, ∆HbA 1C , in addition to absolute HbA 1C values, should be monitored in post-AMI patients. Furthermore, additional studies to further evaluate the mechanisms for these changes and potential targets for interventions are warranted.
Author Contributions: Y.P., H.G. and A.S. contributed to the conception and design of the work. Y.P. contributed to the acquisition and analysis of data for the work. A.S. and H.G. contributed to data interpretation. A.S. drafted the manuscript. Y.P. and H.G. critically revised the manuscript. All gave final approval and agreed to be accountable for all aspects of the work, ensuring its integrity and accuracy. All authors have read and agreed to the published version of the manuscript.

Limitations
First, this is a non-randomized retrospective observational study; hence, it has the limitations of such a design and is subject to bias; mainly, it cannot show causality. Second, data regarding therapeutic interventions in general, and particularly some medications for DM treatment, were not available and could be a source of bias, particularly since it is currently known that the way in which glucose levels are reduced has a significant effect on the ultimate outcome of diabetic patients [40]. Third, the HbA1C tests were taken at the discretion of the treating physicians rather than at prespecified intervals. Fourth, we used all-cause mortality rather than cause-specific mortality (e.g., cardiovascular). Fifth, data regarding multiple factors that could bias HbA1C levels were not included.

Conclusions
The study shows for the first time that among DM patients with non-fatal AMI, longterm changes in HbA1C values and the time interval for these changes (when increasing) are significant independent prognostic markers for long-term all-cause mortality with a U-shaped association. Thus, ∆HbA1C, in addition to absolute HbA1C values, should be monitored in post-AMI patients. Furthermore, additional studies to further evaluate the mechanisms for these changes and potential targets for interventions are warranted.
Author Contributions: Y.P., H.G. and A.S. contributed to the conception and design of the work. Y.P. contributed to the acquisition and analysis of data for the work. A.S. and H.G. contributed to data interpretation. A.S. drafted the manuscript. Y.P. and H.G. critically revised the manuscript. All gave final approval and agreed to be accountable for all aspects of the work, ensuring its integrity and accuracy. All authors have read and agreed to the published version of the manuscript.