The impact of HbA1c variability and of an absolute change in HbA1c during 22–26 months differed with the index HbA1c level. In this population-based observational study among Danish individuals with type 2 diabetes, we found that HbA1c variability was associated with mortality irrespective of the magnitude of absolute change in HbA1c and index HbA1c for those with an index HbA1c≤8%. In general, a higher HbA1c variability or a decrease in HbA1c level was associated with increased mortality for individuals with an index HbA1c≤8%. For individuals with an index HbA1c>8% we found no association between HbA1c variability and mortality, but a linear relationship was identified between absolute change in HbA1c and mortality, with the lowest mortality for greatest decline in HbA1c.
To the best of our knowledge, evidence with respect to associations between HbA1c variability and mortality is limited to one study among 881 participants revealing SD and CV of HbA1c measurements, assessed every 2–4 months, to be independently associated with all-cause mortality, and with a tendency of stronger associations in participants with a mean HbA1c below 7.3% (56 mmol/mol) (the median HbA1c value in the study). This contradicts our finding among individuals with index HbA1c>8% (64 mmol/mol) where no association between HbA1c variability and mortality could be identified. However, they analyzed independency of mean HbA1c while ignoring any trend in HbA1c, whereas we analyzed independency of index HbA1c and absolute change in HbA1c. A Korean study investigated associations between HbA1c variability and intima-media thickness. The predominant cause of death in type 2 diabetes is cardiovascular disease, and carotid intima-media thickness is a well-established index of early atherosclerosis, which is widely used as a surrogate marker of cardiovascular disease. Among individuals with an index HbA1c≤8% (64 mmol/mol), we found an association between HbA1c variability and mortality quite independent of index HbA1c and absolute change in HbA1c; this is in contrast to the Korean study, where no independent association between HbA1c variability and intima-media thickness was found. However, they analyzed independency of index HbA1c and mean HbA1c while ignoring any trend in HbA1c, whereas we analyzed independency of index HbA1c and absolute change in HbA1c. Likewise, they defined HbA1c variability in terms of SD to the mean HbA1c, whereas we defined HbA1c variability as absolute value residuals to the line connecting index HbA1c and closing HbA1c. When defining HbA1c variability as the SD to the mean HbA1c, a steady linear change in HbA1c will entail a high SD value, which we would suggest should not be considered as an indication of a high fluctuation in HbA1c. However, secondary analyses in our population replacing HbA1c variability with SD did not change the results of why a different definition of HbA1c variability is unlikely to be the cause of the dissimilar findings. For individuals with an index HbA1c>8% (64 mmol/mol), our results agreed with the Korean study, as we found no association between HbA1c variability and mortality. Recent studies of associations between HbA1c variability and development of microalbuminuria, progression of nephropathy, and incidence of cardiovascular disease demonstrated HbA1c variability to be an independent risk factor. However, the definition of HbA1c variability was similar to that of the Korean study, and not to the one used in our study. Their analyses also differed as they did not include index HbA1c and used the classic Cox proportional regression analysis in contrast to our modeling with restricted cubic splines. Restricted cubic splines modeling allows piecewise cubic polynomial shapes between knots, unlike the classic Cox proportional hazard regression in which linear relationships are assumed. Whether similar statistical modeling would have produced similar results is unknown. However, in our observational study, it is intriguing why the association between high variability in HbA1c and mortality only holds for individuals with an index HbA1c below a certain level. Perhaps a persistent high glycemic load outweighs the harm of high variability. Future research should address this and clarify the pathway.
Meta-analyses of randomized controlled trials have demonstrated no association between mortality and intensive glycemic control as measured by the HbA1c level, but as far as we know, studies investigating associations between individual absolute change in HbA1c and mortality are limited to a post hoc analysis of data from the ACCORD trial. The ACCORD study found no association between mortality and change in HbA1c in the standard treatment group, but demonstrated an increased mortality in the intensively treated group for participants whose HbA1c declined little or not at all. This contradicts our findings of an association between a decrease in HbA1c and increased mortality among people with an index HbA1c ≤8% (64 mmol/mol) and a linear relationship between absolute change in HbA1c and mortality for people with index HbA1c>8% (64 mmol/mol). However, only people with type 2 diabetes and either a prior cardiovascular event or other evidence of high cardiovascular risk were enrolled in the ACCORD study, and their analysis did not include information on HbA1c variability. For individuals with an index HbA1c≤8% (64 mmol/mol), we found HbA1c variability to have an association with mortality, independent of the index level, and absolute change in HbA1c, which may indicate that it is important to include this information in analyses of the impact of HbA1c on mortality.
Yet our results of an association between a decrease in HbA1c and increased mortality among people with an index HbA1c ≤8% (64 mmol/mol) is in line with a UK study demonstrating an association between low HbA1c and increased mortality; results which were replicated in a study from our database. The finding of no association between an increase in HbA1c and mortality among people with an index HbA1c ≤8% (64 mmol/mol) is, however, not in agreement with previous studies, which identified HbA1c levels above 7% (53 mmol/mol) and 8% (64 mmol/mol), respectively, to be associated with increased mortality. Therefore, we expected to find increased mortality among those who had a high increase in HbA1c. Among people with an index HbA1c>8% (64 mmol/mol), we found a linear relationship between absolute change in HbA1c and mortality, which is in line with the previously mentioned studies.
We used restricted cubic splines modeling with adjustment for HbA1c variability whereas the previously mentioned studies only addressed index HbA1c or mean HbA1c and mortality, ignoring trend and variability in HbA1c.
Significant strengths of this study are its large population-based design with full follow-up of all individuals and the possibility to adjust for comorbidities. Further, the statistical modeling with restricted cubic splines allows a flexible and robust modeling of the relationship between mortality and HbA1c variability or absolute change in HbA1c. Moreover, index HbA1c, HbA1c variability, and absolute change in HbA1c were included in the same model.
This study is an observational study, and we do not know the reasons for having HbA1c measured or the reasons behind the changes in HbA1c, which are the primary limitations. Most likely, the previously measured HbA1c level and the general condition of the person will have influenced when a new measurement was made. Those who did not have a measurement 22–26 months after the index HbA1c (n=10 979) and those who did not have at least one measurement in-between (n=282) have not been included. As in Denmark all healthcare visits and laboratory measurements addressing a person's type 2 diabetes are free of charge and individuals with type 2 diabetes are encouraged to have their HbA1c measured every 3 months and are offered a special annual diabetes control at the GP every 12 months, we believe the findings in our study to be representative of people attending their GP at least for their annual diabetes control. Whether individuals with type 2 diabetes, who do not attend their GP (or other healthcare professionals), or who choose not to have their HbA1c measured, experience a different impact on mortality of HbA1c variability or absolute changes in HbA1c is unknown. For some individuals, the reason behind having no HbA1c measurement is death. This might have induced immortal time bias to the results, and it is important to keep in mind that associations might look different for the unhealthiest individuals who die within 2 years after the first HbA1c measurement or who are too ill to undertake a new HbA1c measurement. HbA1c measurements were analyzed at four different laboratories with standardized analyses and a coefficient of variation of 2.3%. Any measurement error might lead to a false classification into the five HbA1c index groups. To investigate the robustness of our results, we have added random measurement noise to the index HbA1c accordingly to a coefficient of 2.3% and completed the analysis as before. For each analysis, we have cycled this procedure 100 times and estimated the average relative variance increase (RVI). It indicates the increase in variance of the estimates because of the added noise: the closer the number is to 0, the less effect the added noises have on the variance of the estimates. RVI due to the added noise was small: 0.077 and 0.073 for absolute change and variability analysis, respectively. We lacked information of diabetes duration in some of the included individuals and for those where we had this information, it was primarily self-reported, and therefore it might not be valid. The overall conclusion did not change significantly when including the available information on diabetes duration, so we consider it unlikely that omission of duration has introduced substantial bias. However, it is unknown whether duration would have affected the results if the "true" durations had been available for analysis.
We have no information regarding HbA1c after the closing HbA1c measurement, nor on treatment modality. Consequently the results of the analyses are solely based on changes in HbA1c, HbA1c variability, and treatment modality before start of follow-up, and do not take any associations between changes during follow-up and mortality into account. Further, we did not have access to information on smoking habits. The positive predictive value of the algorithm used to identify people with type 2 diabetes is estimated at 89%, and thus some of the included individuals in this study might not have type 2 diabetes. However, owing to the extra selection criteria of at least three HbA1c measurements during 2 years, this problem is likely negligible.
In summary, the associations of HbA1c variability and absolute changes in HbA1c with mortality differed across index HbA1c levels among individuals with type 2 diabetes. Increased HbA1c variability was associated with increased mortality for individuals with index HbA1c≤8% (64 mmol/mol), and no association could be identified for individuals with index HbA1c>8% (64 mmol/mol). For individuals with index HbA1c≤8% (64 mmol/mol), a decline in HbA1c was associated with increased mortality, whereas for individuals with index HbA1c>8% (64 mmol/mol) a decline in HbA1c was associated with decreased mortality. Thus, we have identified HbA1c variability and absolute changes in HbA1c to be associated with mortality independently of the index HbA1c level. However, results need to be confirmed in other study populations. Hopefully, in the long term, this can lead to a targeted prevention of complications by more individualized treatment, presumably by aiming at reducing HbA1c variability for individuals with HbA1c≤8% (64 mmol/mol).