Relationship of HbA1c and All-cause Mortality in T2DM
The administrative unit Aarhus County, Denmark, existing from 1970 until 2007, had a population of approximately 650 000 persons, corresponding to 12% of the Danish population. The present study is based on this background population. Individuals with type 2 diabetes were identified from public data files in Aarhus County, Denmark, with a dedicated and validated algorithm. This algorithm is described in detail elsewhere, but briefly it is designed to identify people with type 2 diabetes in Aarhus County, Denmark, has a sensitivity of 96% and a positive predictive value of 89%, and includes information on age, sex, laboratory results, and redeemed prescriptions. This algorithm is unique in Denmark due to the population-based setting with inclusion of laboratory results, unlike the population-based Danish National Diabetes Registry, which was established in 2006, where laboratory results are lacking. In this study, we identified individuals who were registered with type 2 diabetes either by December 31, 2001, 2002, 2003, 2004, 2005, or 2006, and who subsequently had at least three HbA1c measurements as follows: one index measure in the same year as they were registered as having type 2 diabetes, one closing measure 22–26 months after the index measure, and at least one measurement in-between. HbA1c measurements were analyzed at four different laboratories with standardized analyses and a coefficient of variation of 2.3%. Each general practitioner (GP; and clinical department) has an agreement with one specific laboratory (out of the four) as to why measurements for a single person most likely have been analyzed at the same laboratory. A total of 25 490 individuals were identified with type 2 diabetes. Of those, 22 466 had an HbA1c measurement in the same year as they were registered in the database, 11 487 had HbA1c measurements 22–26 months after the first measurement, and 11 205 had at least one HbA1c measurement in-between. These 11 205 individuals form the present study population.
Disease diagnoses included in Charlson's comorbidity index were obtained from record linkage with the Danish National Patient Register, which covers all hospitalizations and outpatient visits in Denmark. Information on mortality and individual migration history was obtained by record linkage with the nationwide Danish Civil Registration System, where all persons living in Denmark are registered for administrative purposes, using the national Civil Registration Number as the unique person identifier across registries. For all events, dates were available.
The first registered HbA1c measurement was used as index value, and the measurement within 22–26 months after the first measurement was used as closing HbA1c. If one individual had more than one measurement 22–26 months after the first measurement, the first registered measurement in the timespan was used. HbA1c variability was defined as the mean absolute residual of HbA1c measurements to the line connecting index HbA1c with closing HbA1c. Figure 1 illustrates how index HbA1c, closing HbA1c, and residual and follow-up time were defined for a single person. Absolute change in HbA1c was defined as the difference between closing HbA1c and index HbA1c. Prior cardiovascular disease and number of non-cardiovascular diseases were defined on the basis of diseases included in Charlson's comorbidity index: myocardial infarction, congestive heart failure, peripheral vascular disease, and cerebrovascular disease represented prior cardiovascular disease. The number of prior non-cardiovascular diseases was the number of different diseases each individual was hospitalized with from the list: dementia, chronic pulmonary disease, connective tissue disease, ulcer disease, mild liver disease, hemiplegia, moderate to severe renal disease, diabetes with end-organ damage, any tumor, leukemia, lymphoma, moderate or severe liver disease, metastatic solid tumor, and AIDS.
(Enlarge Image)
Figure 1.
An example of how index HbA1c (glycated hemoglobin), closing HbA1c, and residual and follow-up time were defined. The blue dots represent all HbA1c measurements for a single person.
As studies have shown that both low and high HbA1c values might be associated with higher mortality, a specific absolute change in HbA1c might have a different impact depending on the index value. Therefore, we stratified analyses according to index HbA1c in five groups. The five groups were chosen to reflect values used in daily clinical practice and were: ≤6% (42 mmol/mol), 6.1–6.5% (43–48 mmol/mol), 6.6–7.4% (49–57 mmol/mol), 7.5–8% (58–64 mmol/mol), and ≥8.1% (65 mmol/mol).
In order to make efficient use of within-category information by allowing for different shapes of the underlying risk function within each category, we used restricted cubic splines, with four knots and an a priori reference point. For HbA1c variability, the reference point was set to 0.5 HbA1c percentage points, and to null for the absolute change in HbA1c. Unadjusted survival was estimated by the Kaplan-Meier method, and Cox proportional hazard models were used to adjust for covariates. The assumption of proportional hazards was examined by testing the equality of HRs within <1 year after start of follow-up and ≥1 year after. Adjustment was made for age, sex (male, female), preventive pharmacological treatment of cardiovascular diseases (yes, no), prior cardiovascular disease (yes, no), number of non-cardiovascular diseases, and index HbA1c. Analyses addressing either HbA1c variability or absolute change in HbA1c were adjusted for the other.
Secondary analyses exploring if the association patterns were different for different strata of sex (female, male), treatment modality (no glucose-lowering treatment, oral glucose-lowering treatment, insulin), and self-reported diabetes duration (≤1, ≥6 years) were performed, and adjustment for self-reported diabetes duration was made to the primary analyses. Likewise, analyses replacing HbA1c variability defined as the absolute mean residual, with SD and with the coefficient of variation (COV), were performed.
The time since closing HbA1c measurement was used as timescale. Each person was followed until death, emigration (censoring), or August 31, 2010, whichever came first. Restricted cubic splines with knots determined by Harrell's default percentiles (5th, 35th, 65th, and 95th) were calculated using Stata V.12.1 (StataCorpLP, College Station, Texas, USA). Estimates are presented with 95% CIs.
Methods
Study Design
The administrative unit Aarhus County, Denmark, existing from 1970 until 2007, had a population of approximately 650 000 persons, corresponding to 12% of the Danish population. The present study is based on this background population. Individuals with type 2 diabetes were identified from public data files in Aarhus County, Denmark, with a dedicated and validated algorithm. This algorithm is described in detail elsewhere, but briefly it is designed to identify people with type 2 diabetes in Aarhus County, Denmark, has a sensitivity of 96% and a positive predictive value of 89%, and includes information on age, sex, laboratory results, and redeemed prescriptions. This algorithm is unique in Denmark due to the population-based setting with inclusion of laboratory results, unlike the population-based Danish National Diabetes Registry, which was established in 2006, where laboratory results are lacking. In this study, we identified individuals who were registered with type 2 diabetes either by December 31, 2001, 2002, 2003, 2004, 2005, or 2006, and who subsequently had at least three HbA1c measurements as follows: one index measure in the same year as they were registered as having type 2 diabetes, one closing measure 22–26 months after the index measure, and at least one measurement in-between. HbA1c measurements were analyzed at four different laboratories with standardized analyses and a coefficient of variation of 2.3%. Each general practitioner (GP; and clinical department) has an agreement with one specific laboratory (out of the four) as to why measurements for a single person most likely have been analyzed at the same laboratory. A total of 25 490 individuals were identified with type 2 diabetes. Of those, 22 466 had an HbA1c measurement in the same year as they were registered in the database, 11 487 had HbA1c measurements 22–26 months after the first measurement, and 11 205 had at least one HbA1c measurement in-between. These 11 205 individuals form the present study population.
Disease diagnoses included in Charlson's comorbidity index were obtained from record linkage with the Danish National Patient Register, which covers all hospitalizations and outpatient visits in Denmark. Information on mortality and individual migration history was obtained by record linkage with the nationwide Danish Civil Registration System, where all persons living in Denmark are registered for administrative purposes, using the national Civil Registration Number as the unique person identifier across registries. For all events, dates were available.
Statistical Analysis
The first registered HbA1c measurement was used as index value, and the measurement within 22–26 months after the first measurement was used as closing HbA1c. If one individual had more than one measurement 22–26 months after the first measurement, the first registered measurement in the timespan was used. HbA1c variability was defined as the mean absolute residual of HbA1c measurements to the line connecting index HbA1c with closing HbA1c. Figure 1 illustrates how index HbA1c, closing HbA1c, and residual and follow-up time were defined for a single person. Absolute change in HbA1c was defined as the difference between closing HbA1c and index HbA1c. Prior cardiovascular disease and number of non-cardiovascular diseases were defined on the basis of diseases included in Charlson's comorbidity index: myocardial infarction, congestive heart failure, peripheral vascular disease, and cerebrovascular disease represented prior cardiovascular disease. The number of prior non-cardiovascular diseases was the number of different diseases each individual was hospitalized with from the list: dementia, chronic pulmonary disease, connective tissue disease, ulcer disease, mild liver disease, hemiplegia, moderate to severe renal disease, diabetes with end-organ damage, any tumor, leukemia, lymphoma, moderate or severe liver disease, metastatic solid tumor, and AIDS.
(Enlarge Image)
Figure 1.
An example of how index HbA1c (glycated hemoglobin), closing HbA1c, and residual and follow-up time were defined. The blue dots represent all HbA1c measurements for a single person.
As studies have shown that both low and high HbA1c values might be associated with higher mortality, a specific absolute change in HbA1c might have a different impact depending on the index value. Therefore, we stratified analyses according to index HbA1c in five groups. The five groups were chosen to reflect values used in daily clinical practice and were: ≤6% (42 mmol/mol), 6.1–6.5% (43–48 mmol/mol), 6.6–7.4% (49–57 mmol/mol), 7.5–8% (58–64 mmol/mol), and ≥8.1% (65 mmol/mol).
In order to make efficient use of within-category information by allowing for different shapes of the underlying risk function within each category, we used restricted cubic splines, with four knots and an a priori reference point. For HbA1c variability, the reference point was set to 0.5 HbA1c percentage points, and to null for the absolute change in HbA1c. Unadjusted survival was estimated by the Kaplan-Meier method, and Cox proportional hazard models were used to adjust for covariates. The assumption of proportional hazards was examined by testing the equality of HRs within <1 year after start of follow-up and ≥1 year after. Adjustment was made for age, sex (male, female), preventive pharmacological treatment of cardiovascular diseases (yes, no), prior cardiovascular disease (yes, no), number of non-cardiovascular diseases, and index HbA1c. Analyses addressing either HbA1c variability or absolute change in HbA1c were adjusted for the other.
Secondary analyses exploring if the association patterns were different for different strata of sex (female, male), treatment modality (no glucose-lowering treatment, oral glucose-lowering treatment, insulin), and self-reported diabetes duration (≤1, ≥6 years) were performed, and adjustment for self-reported diabetes duration was made to the primary analyses. Likewise, analyses replacing HbA1c variability defined as the absolute mean residual, with SD and with the coefficient of variation (COV), were performed.
The time since closing HbA1c measurement was used as timescale. Each person was followed until death, emigration (censoring), or August 31, 2010, whichever came first. Restricted cubic splines with knots determined by Harrell's default percentiles (5th, 35th, 65th, and 95th) were calculated using Stata V.12.1 (StataCorpLP, College Station, Texas, USA). Estimates are presented with 95% CIs.
SHARE
