In the present study of the general Korean population with neither a history of cancer nor CVD, non-diabetic adults with very low HbA1c were at significantly higher risk of all-cause and cancer mortality than were individuals with HbA1c of 37–41 mmol/mol (5.5–5.9%).
Higher HbA1c levels in the prediabetic and diabetic ranges were associated with an increased risk of all-cause, cancer, and non-CVD/noncancer mortality in a dose-response manner.
Furthermore, when stratified by the median RBC count or by having liver diseases, there was a strong association between low HbA1c levels and all-cause death among individuals with low RBC counts or liver diseases.
Most previous studies agree that high HbA1c levels increase the risk of death. However, the association between low HbA1c levels and mortality is equivocal. Several studies have reported that only high HbA1c levels increase mortality [10–12, 20]. In the general Japanese population, compared with HbA1c levels <31 mmol/mol (5.0%), high HbA1c levels (>42 mmol/mol; 6.0%) in individuals without treatment for diabetes were significantly associated with an increased risk of all-cause mortality and death from CVD [10]. In Singaporean Chinese adults without diagnosed diabetes, compared with HbA1c levels of 36–38 mmol/mol (5.4–5.6%), only HbA1c levels ≥48 mmol/mol (6.5%) were significantly associated with all-cause (HR, 1.96; 95% CI, 1.56–2.46), CVD (HR, 2.63; 95% CI, 1.77–3.90), and cancer (HR, 1.51; 95% CI, 1.04–2.18) mortality [20]. In Australian adults (aged ≥25 years) without diagnosed diabetes, HbA1c levels exhibited a linear relationship with all-cause and CVD mortality [12]. Our findings of an increased risk of all-cause mortality in the pre-diabetic and diabetic HbA1c ranges are in line with these previous reports. On the other hand, consistent with our findings, some epidemiologic studies [8, 13–16] have observed U- or J-shaped associations between HbA1c levels and the risk of mortality. In the Atherosclerosis Risk in Communities (ARIC) study, non-diabetic participants with HbA1c levels < 31 mmol/mol (5.0%), as well as those with HbA1c levels ≥37 mmol/mol (5.5%), were at a significantly increased risk of death from any cause (HR, 1.48; 95% CI, 1.21-1.81) [8]. In the general German population without known diabetes, restricted cubic spline models showed a U-shaped association between HbA1c levels and all-cause mortality: the lowest risk was at HbA1c levels of 36–38 mmol/mol (5.4–5.6%) and a significantly increased risk at ≤31 mmol/mol (5.0%) and ≥46 mmol/mol (6.4%) [13]. The association between extremely low HbA1c levels and mortality were assessed in several studies. In US adults without diabetes, compared with HbA1c levels of 31–36 mmol/mol (5.0–5.4%), HbA1c levels <20 mmol/mol (4.0%) were associated with an increased risk of all-cause mortality (HR, 2.90; 95% CI, 1.25–6.76) [14]. In the general New Zealand population, individuals with HbA1c levels <20 mmol/mol (4.0%) had a higher risk of mortality (HR, 2.90; 95% CI, 0.91–9.19) than did those with HbA1c levels of 20–30 mmol/mol (4.0–4.9%), with marginal statistical significance [15]. In the current study, we were unable to examine extremely low HbA1c levels because the number of participants with HbA1c levels <20 mmol/mol (4.0%) was very small (n = 2).
Hyperglycaemia has been associated with an increased risk of cardiovascular morbidity [24–26] and mortality [10, 12] in the general population. This may be due to the vascular damage caused by increased oxidative stress and endothelial dysfunction in individuals with impaired fasting glucose or impaired glucose tolerance [27, 28]. Moreover, elevated HbA1c levels influences cancer progression through an increase in the levels of insulin, insulin-like growth factor-1 (IGF-1), and inflammatory cytokines in circulation [29, 30]. However, the potential mechanism underlying the association between low HbA1c levels and increased mortality remains unclear. Low HbA1c levels have been correlated with impaired RBC related indices and increased liver function indices [14, 23]. These factors, in turn, were shown to correlate with inflammatory processes and increased morbidity and mortality [31, 32]. In other words, low HbA1c levels are considered a marker of deteriorated health condition. Thus, the association between low HbA1c levels and higher mortality may be explained by a result of reverse causation due to comorbid conditions [14, 23]. In our study sample, participants with HbA1c levels <31 mmol/mol (5%) had lower RBC, haemoglobin, and haematocrit levels and higher total bilirubin levels and prevalence of liver diseases than did those with HbA1c levels of 37–41 mmol/mol (5.5–5.9%), the reference group. Moreover, low HbA1c levels, <31 mmol/mol (5%), were associated with an increased risk of all-cause mortality only among people with pre-existing liver disease or less than median RBC counts. These results suggest potential roles of RBC and liver function markers in the association between low HbA1c levels and increased mortality. However, after we extensively adjusted for various RBC and liver function markers, the association between low HbA1c levels and all-cause mortality persisted. Further studies on the mechanism leading to low HbA1c levels and the association with mortality are warranted.
To the best of our knowledge, this is the first study to find an increased risk of mortality in the Asian population with low HbA1c levels and to evaluate the association between HbA1c levels and mortality in the general Korean population. The other unique feature of this analysis was the time-dependent modelling of the association between HbA1c levels over time and mortality. Nearly all previous studies on this question utilized time-fixed methods using a single HbA1c measurement at baseline [8, 10–16, 20].
Time-fixed analyses use a single measurement of HbA1c levels; therefore, changes in HbA1c levels during follow up cannot be considered, and there is a possibility of misclassification. In contrast, time-dependent analysis makes use of all the available HbA1c data, and the risk is assessed considering each HbA1c data point during follow-up. This approach is more reflective of clinical practice. Coherent results from both time-fixed and time-dependent analyses strengthen the association of low HbA1c levels with an increased risk of all-cause mortality. An additional strength of this study lies in the long-term population-based cohort design (16-year follow-up) and the use of extensive data on RBC related indices and disease history.
Despite its strengths, the current study had some limitations. First, the participants were aged 40–70 years at the baseline examination and were enrolled from two communities. Therefore, the findings of this study may not be directly generalizable to younger adults or the entire Korean population. However, general characteristics and HbA1c level distribution were similar when we compared our study population with the KNHANES participants, who are a representative sample. Specifically, among the KNHANES participants aged 40–70 in 2019, 44% were men, the mean age was 55.0 years, and the mean HbA1c level was 40.9 mmol/mol (5.9%). Second, among the 9,294 participants who met the inclusion criteria at baseline, 14.5% did not attend any follow-up examination until 2016. However, reassuringly, the major variables, such as HbA1c levels, age, sex, and comorbidities, were similar between participants and nonparticipants in the follow-up examinations (data not shown). Therefore, it is unlikely that the association between HbA1c levels and mortality found in the time-dependent analysis using follow-up data is severely underestimated or overestimated. Third, there was a small number of participants with extremely low HbA1c levels; therefore, we were unable to analyse the relationship between low HbA1c levels and detailed cause-specific mortality. Finally, information on the diagnosis and treatment of diabetes, hypertension, dyslipidaemia, and liver diseases was obtained from questionnaires. Participants omitting to report existing diseases or having undiagnosed health conditions could have limited our ability to classify people with diabetes or other existing diseases. However, we collected data on the disease history using a structured questionnaire with questions about each disease asked separately by trained interviewers. Moreover, missing rates of related variables were very low (0.04%).