Low Hemoglobin A_1c in Nondiabetic Adults: An elevated risk state?

The Atherosclerosis Risk in Communities (ARIC) Study is an ongoing community-based prospective cohort study of 15,792 middle-aged adults from four U.S. communities: Washington County, MD; suburban Minneapolis, MN; Jackson, MS; and Forsyth County, NC. The first study visit occurred between 1987 and 1989 with three follow-up visits that occurred approximately every 3 years (13,14). Visit 2 (1990–1992) was attended by 14,348 participants and is the baseline for the present analysis. Participants were included in our analyses irrespective of the previous occurrence of nonfatal events. We excluded participants who self-identified as other than white or black race (n = 48) or who were missing data on HbA_1c or other covariates of interest (n = 970), leaving a final sample size of 13,288 participants in this analysis. Institutional review boards at each clinical site approved the study protocol, and written informed consent was obtained from all participants.

Frozen whole-blood samples collected at ARIC visit 2 were thawed and assayed for the measurement of HbA_1c using high-performance liquid chromatography (Tosoh A1c 2.2 Plus Glycohemoglobin Analyzer method in 2003 to 2004 and the Tosoh G7 method in 2007 to 2008; Tosoh Corporation) (15). Both instruments were standardized to the Diabetes Control and Complications Trial assay (16).

ARIC Study investigators conduct continuous surveillance for all hospitalizations and deaths among participants via annual phone calls to participants or proxies, and detailed information on deaths is obtained from family members, coroner reports, or health department death certificates. Methods for the ascertainment of death and its causes in ARIC have been published previously (13). We classified deaths according to underlying cause, on the basis of coding from the ICD-9 and -10. We divided causes of death into the following major diagnosis categories defined by the ICD codes: 1) cancers (ICD-10 codes of C00-D48, ICD-9 codes of 140–239); 2) cardiovascular system (ICD-10 codes of I00–I99, ICD-9 codes of 390–459); 3) respiratory system (ICD-10 codes of J00–J99, ICD-9 codes of 460–519); 4) digestive system and liver (ICD-10 codes of K00–K93, ICD-9 codes of 520–579); and 5) genitourinary system and kidney (ICD-10 codes of N00–N99, ICD-9 codes of 580–629). We also identified incident liver disease hospitalizations from hospital discharge records with an ICD-9 code for liver disease using the following ICD-9 codes (listed anywhere in the hospital discharge record): 570.0–573.9.

All covariates were assessed during visit 2 (1990–1992), except education and physical activity, which were assessed during visit 1 (1987–1989). Covariates evaluated as potential confounders included: age (continuous), race/field center (Washington County, MD whites; Minneapolis, MN whites; Forsyth County, NC whites, Forsyth County, NC blacks; and Jackson, MS blacks), sex (male/female), education (less than high school, high school or equivalent, or more than high school), BMI (continuous), waist-to-hip ratio (continuous), cigarette smoking (current, former, or never), alcohol intake (current, former, or never), physical activity [assessed with the use of Baecke index (17)], family history of diabetes, presence of hypertension (defined as systolic blood pressure ≥140 mmHg, diastolic blood pressure ≥90 mmHg, or use of antihypertensive medication during the previous 2 weeks), total cholesterol (continuous), HDL cholesterol (continuous), prevalent coronary heart disease, prevalent stroke, fasting glucose (continuous), hemoglobin concentration (continuous), red-cell mean corpuscular volume (MCV; continuous), total leukocyte count (continuous), and plasma fibrinogen levels (continuous). Further details about data collection methods in ARIC are described elsewhere (13,18).

We divided the baseline study population into groups according to clinical categories of glycated hemoglobin (<5.0, 5.0 to <5.7, 5.7 to <6.5, and ≥6.5% or diagnosed diabetes) (2). With the HbA_1c category of 5.0 to <5.7% as reference, we used logistic regression models to identify predictors of low HbA_1c (<5.0%) at baseline among persons with HbA_1c <5.7% (n = 9,254). To characterize the prospective association of low HbA_1c with all-cause and cause-specific mortality and risk of liver disease hospitalization, we used Cox proportional hazards models to estimate the hazard ratios (HRs) and their corresponding 95% CIs across baseline categories of HbA_1c. We verified that the proportional hazards assumption was met (19). Two models were used: model 1 was adjusted for age, sex, and race/field center, and model 2 was adjusted for the variables in model 1 plus total and HDL cholesterol, BMI, waist-to-hip ratio, hypertension, family history of diabetes, education level, alcohol use, physical activity, smoking status, hemoglobin, MCV, fibrinogen, and leukocyte count. We performed sensitivity analyses with additional adjustment for serum glucose, pulmonary function tests, white-cell differential, platelet count, and mean corpuscular hemoglobin. We formally tested for interaction by hemoglobin level. We also looked separately at the association among persons with and without anemia (defined as hemoglobin <13.5 g/dL for males and <12.0 g/dL for females) and after restricting the population to persons with ≥3 years of follow-up to address the potential for reverse causality. In all analyses, participants were censored if the participant was lost to follow-up, died of other causes, or reached the end of the follow-up period (31 December 2008). We used a restricted cubic spline model to investigate the continuous association between HbA_1c and risk of liver disease hospitalization. All reported P values are two-sided, and P < 0.05 was considered statistically significant. All analyses were performed using Stata Statistical software version 11 (StataCorp, College Station, TX).

Of the 13,288 participants included in the study, 3,078 (23.2%) died by the end of follow-up period (maximum follow-up was 18 years). The demographic, clinical, and lifestyle characteristics of the study population are shown stratified by HbA_1c categories in Table 1. Overall, the mean age was 57 years, 55% were women, and 24% were black. Persons with higher HbA_1c were more likely to be older and current or former smokers.

The odds ratios and 95% CIs for predictors of low HbA_1c are shown in Table 2. Compared with those with HbA_1c values 5.0 to <5.7%, participants with HbA_1c values <5.0% were younger, had a lower BMI, were less likely to smoke, had lower total cholesterol levels, and were much less likely to have hypertension, a family history of diabetes, or a history of cardiovascular disease. Participants with HbA_1c <5.0% were more likely to have either low (<12.9 g/dL) or high hemoglobin (>14.7 g/dL).

Of the total 13,288 participants included in this study, 1,113 (8.4%) participants died of cancer, 1,085 (8.2%) died of cardiovascular disease, 235 (1.8%) died of respiratory system disease, 82 (0.6%) died of digestive system or liver disease, and 61 (0.5%) died of genitourinary system or kidney disease.

The HRs for all-cause and cause-specific mortality by baseline HbA_1c category are shown in Table 3. Comparing models 1 and 2, we observed that multiple adjustments strengthen the association of low HbA_1c with cause-specific and all-cause mortality, which is consistent with negative confounding. Using the group with HbA_1c of 5.0 to <5.7% as the reference group, HbA_1c values <5.0% were associated with a significantly increased risk of all-cause mortality (HR: 1.32, 95% CI: 1.13–1.55) and of cancer death (1.47, 95% CI: 1.16–1.84) in model 2. Low HbA_1c was also associated with an increased risk of cardiovascular system deaths (1.27, 95% CI: 0.93–1.75) and respiratory system deaths (1.42, 95% CI: 0.78–2.56), but these associations were not statistically significant, possibly owing to the smaller number of deaths due to these causes. Low HbA_1c was not significantly associated with risk of deaths related to the digestive system or liver disease (0.96, 95% CI: 0.34–2.71) or deaths due to conditions of the genitourinary system and kidney disease deaths (1.01, 95% CI: 0.23–4.42). Similar results were observed after adjustment for fasting serum glucose, pulmonary function tests, white-cell differential, platelet count, and mean corpuscular hemoglobin and when the analysis was restricted to persons with ≥3 years of follow-up (data not shown). Tests for interaction by hemoglobin level were not significant (all P values for interaction >0.1). However, in sensitivity analyses among persons with anemia (10.2% of the study population), we observed stronger associations of low HbA_1c with all-cause mortality (1.60, 95% CI: 1.07–2.39), cancer mortality (1.94, 95% CI: 1.03–3.65), and cardiovascular death (1.51, 95% CI: 0.74–3.09). Among participants without anemia, the results were somewhat attenuated but remained significant for all-cause mortality (1.22, 95% CI: 1.03–1.46) and cancer death (1.35, 95% CI: 1.05–1.74).

Compared with persons with HbA_1c 5.0 to <5.7%, those persons with high HbA_1c (≥6.5%) also had an increased risk of all-cause mortality and deaths from cancer, cardiovascular disease, diseases of the respiratory system, digestive system and liver disease, diseases of the genitourinary system, and kidney disease (Table 3). Similar results were observed among persons with and without anemia, after adjustment for fasting serum glucose, pulmonary function tests, white-cell differential, platelet count, mean corpuscular hemoglobin, and when the analysis was restricted to persons with ≥3 years of follow-up (data not shown).

There were 353 hospitalizations for liver disease during follow-up. Fig. 1 presents the adjusted HRs and 95% CIs from the restricted cubic spline model for the association of baseline HbA_1c with risk of hospitalization for liver disease. We observed a pronounced J-shaped association; both very low and high HbA_1c values were associated with an increased risk of liver hospitalization in this population.

Compared with persons with HbA_1c in the normal range, we found that low HbA_1c values (<5.0%) were associated with an increased risk of all-cause mortality and death from various causes, including cancer. This finding is consistent with previous studies that have documented a J- or U-shaped association between HbA_1c and all-cause mortality (3,7,8,10,11). Our results are contrary to a recent analysis of the European Prospective Investigation of Cancer-Norfolk Study that reported no increase in risk of mortality at low normal HbA_1c values in a relatively homogenous population of Caucasians (20). The debate regarding the HbA_1c–mortality association also extends to the mortality curves for fasting and 2-h glucose, which are reported to be J-shaped in some studies (21–24). No single cause of death appeared to drive the association between low HbA_1c and risk of death in our study. Prior studies have suggested that liver disease may be an important contributor to low HbA_1c values (10,12). Consistent with this hypothesis, we observed that low HbA_1c values were associated with an increased risk of hospitalization due to liver disease in the ARIC Study population.

The exact mechanisms underlying the association of low HbA_1c values with increased risk of death are not known. We found that low HbA_1c values were significantly associated with higher mean red-cell volume and low hemoglobin. Other studies have demonstrated associations of red-cell indices with long-term mortality in the general population (25,26). The etiology of these relationships is largely unclear but may reflect that some disease processes affect red-cell turnover and thus HbA_1c. Whereas high HbA_1c values are almost entirely determined by circulating glucose levels (27), it is probable that nonglycemic factors such as red-cell turnover may be of disproportionate importance in the low range of HbA_1c. Our observations suggest low HbA_1c values may be a general marker of ill health, analogous to the well-documented U-shaped association between cholesterol and mortality (28–30). There is strong evidence that some disease processes reduce circulation of cholesterol-bearing lipoproteins (28). Low HbA_1c values may characterize a mix of healthy people and a group in which low HbA_1c is a marker or signal of underlying health issues, such as liver disease or early stages of cancer (3,10). This hypothesis is consistent with the negative confounding observed in our study; once traditional risk factors for death were added to our models, the association of low HbA_1c with all-cause mortality and cause-specific mortality became stronger.

Certain limitations should be considered in interpreting the results of this study. We were limited to information available from the death certificate for the classification of the underlying cause of death, which may have resulted in misclassification, particularly for the noncancer and noncardiovascular outcomes (31–33). We also did not have data available on iron indices, reticulocyte count, or red-cell distribution width. Nonetheless, the ARIC population is a large, diverse cohort with virtually complete follow-up for vital status and comprehensive information on important confounders and major risk factors for mortality. To our knowledge, this study is one of the first comprehensive examinations of the association of low HbA_1c with cause-specific mortality in a general population.

In conclusion, our results add to the evidence that low HbA_1c values may be a marker of elevated mortality risk in the general population. With new recommendations for the use of HbA_1c for diagnosis of diabetes, HbA_1c testing will undoubtedly increase. To the extent that HbA_1c is adopted for diabetes screening, clinicians will frequently observe HbA_1c values in the normal range, including low normal values. Additional work should focus on understanding the nonglycemic determinants of HbA_1c and the mechanisms that explain why low HbA_1c values in some individuals may reflect an elevated risk state.

The ARIC Study was carried out as a collaborative study supported by the National Heart, Lung, and Blood Institute contracts (HHSN268201100005C, HHSN268201100006C, HHSN268201100007C, HHSN268201100008C, HHSN268201100009C, HHSN268201100010C, HHSN268201100011C, and HHSN268201100012C). A.L.C.S. was supported by National Institutes of Health/National Institute of Diabetes and Digestive and Kidney Diseases (NIH/NIDDK) training grant T32-DK-062707. This work was supported by NIH/NIDDK grant R21-DK-080294. E.S. was supported by NIH/NIDDK grants K01-DK-076595 and R01-DK-089174.

No potential conflicts of interest relevant to this article were reported.

V.A. analyzed data, wrote the manuscript, reviewed and edited the manuscript, and contributed to discussion. A.L.C.S. analyzed data, reviewed and edited the manuscript, and contributed to discussion. E.S. reviewed and edited the manuscript and contributed to discussion. V.A. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

The authors thank the staff and participants of the ARIC Study for important contributions.