Visit-to-Visit Variability ofHemoglobin A1c in PeopleWithoutDiabetes and Risk of MajorAdverse Cardiovascular Eventsand All-Cause MortalityDiabetes Care 2019;42:134–141 | https://doi.org/10.2337/dc18-1396

OBJECTIVE

We aimed to study whether visit-to-visit variability of glycated hemoglobin A1c

(HbA1c) is associated with incident major adverse cardiovascular events (MACE),all-cause mortality, and type 2 diabetes in people without diabetes.

RESEARCH DESIGN AND METHODS

We included primary care patients with no history of diabetes or cardiovasculardisease and with three annual HbA1c measurements within normal range (<6.5%[48mmol/mol]). For each individual, wemeasured theHbA1c variability as the SDofthe residuals obtained from a linear regression on the three HbA1c measurements.From the linear regression, we also obtained the estimated index HbA1c (intercept)and the trend over time (slope). Follow-up began at the date of the third mea-surement. Associations between HbA1c variability and outcome were analyzedusing Cox regression, adjusted for traditional risk factors, intercept, and trend andreported as hazard ratio per SD increase in variability (HRSD).

RESULTS

In total, 6,756 individuals were included. During a median follow-up time of 6.3years, 996 developed MACE, 856 died, and 1,267 developed type 2 diabetes. Wefound a significant association between increasing HbA1c variability and incidentMACE (HRSD 1.09 [95% CI 1.03–1.15]) and all-cause mortality (HRSD 1.13 [95% CI1.07–1.20]), whereas there were no associations with type 2 diabetes (HRSD 1.00[95% CI 0.95–1.05]). We calculated 5-year absolute risks of MACE and all-causemortality and found clinically relevant differences across several age, sex, comor-bidity, and HbA1c variability–defined subgroups.

CONCLUSIONS

In a primary care population freeof diabetes and cardiovascular disease, highHbA1c

variability was associated with increased risks of MACE and all-cause mortality.

Type 2 diabetes is an important risk factor for major adverse cardiovascular events(MACE) and premature death from cardiovascular causes (1,2). Glycated hemoglobinA1c (HbA1c) is a readily available and highly reproducible biomarker that reflects the3-month average plasma glucose concentration; it can bemeasured in the nonfastingstate, and a value $6.5% (48 mmol/mol) is considered diagnostic for diabetes (3).

1Laboratory of Molecular Cardiology, Depart-ment of Cardiology, Heart Centre, CopenhagenUniversity Hospital, Rigshospitalet, Copenhagen,Denmark2Laboratory of Molecular Cardiology, Depart-ment of Biomedical Sciences, University of Co-penhagen, Copenhagen, Denmark3Laboratory of Experimental Cardiology, Depart-ment of Biomedical Sciences, University of Co-penhagen, Copenhagen, Denmark4Department of Clinical Biochemistry, Copenha-gen University Hospital, Hvidovre, Denmark5Department of Health Science and Technology,Aalborg University, Aalborg, Denmark6Section of Biostatistics, University of Copenha-gen, Copenhagen, Denmark7Department of Cardiology, Heart Centre, Co-penhagen University Hospital, Rigshospitalet,Copenhagen, Denmark8Department of Clinical Medicine, Faculty ofHealth and Medical Sciences, University of Co-penhagen, Copenhagen, Denmark9Division of Cardiovascular Medicine, Depart-ment of Internal Medicine, University of Mich-igan, Ann Arbor, MI

Corresponding author: Jonas B. Nielsen, jonas.bille.nielsen@gmail.com

Received 29 June 2018 and accepted 29 Septem-ber 2018

This article contains Supplementary Data onlineat http://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc18-1396/-/DC1.

J.G. andM.W.S. contributed equally to this work.

© 2018 by the American Diabetes Association.Readers may use this article as long as the workis properly cited, the use is educational and notfor profit, and the work is not altered. More infor-mation is available at http://www.diabetesjournals.org/content/license.

Jonas Ghouse,1,2 Morten W. Skov,1,2

Jørgen K. Kanters,3 Bent Lind,4

Jonas L. Isaksen,3,5 Paul Blanche,6

Stig Haunsø,1,2 Lars Køber,7,8

Jesper H. Svendsen,1,2,7,8

Morten S. Olesen,1,2 Anders G. Holst,1

Thomas A. Gerds,6 and Jonas B. Nielsen1,9

134 Diabetes Care Volume 42, January 2019

CARDIOVASCULA

RANDMETABOLICRISK

In addition to high HbA1c (4–8), recentobservational studies have also shownthat high long-term glycemic variability,measured as intraindividual variabilityin HbA1c over time, is associated withmicro- and macrovascular complications,independently of mean HbA1c levels,both in patientswith type 1 and thosewithtype 2 diabetes (9–13). The clinicalutility of variation in HbA1c as a markerfor micro- or macrovascular complicationshas yet to be established, but the findingsare supported by data from in vitro andanimal studies indicating that a highplasma glucose variation has a greater ef-fect on endothelial function and oxidativestress generation compared with a sus-tained high glucose level (14–16). Whetherglycemic variability in people without di-abetes is an independent risk factor forfuture cardiovascular events and mortal-ity or even type 2 diabetes is currentlyunknown.Our objective was to evaluate the

relationship between visit-to-visit vari-ability in HbA1c and the risk of MACE,all-cause mortality, and incident type 2diabetes in adults without diabetes.

RESEARCH DESIGN AND METHODS

Study PopulationMost general practitioners practicingin Copenhagen, Denmark, refer theirpatients to one core facility (CopenhagenGeneral Practitioner’s Laboratory [CGPL])for a large variety of clinical tests, includingblood sampling for biochemical analyses.For the current study, we identified allindividuals who had at least one HbA1cmeasurement taken between 1 January2003 and 31 December 2013. Individualswere only included if they, in addition tothe first measurement, had two consec-utive measurements, one after 1 year(63 months) and another after 2 years(63 months) (Supplementary Fig. 1). Thedate of the first measurement corre-sponds to the index date. The baselinedate for follow-up was set at the thirdmeasurement of HbA1c. Patients wereexcluded if they 1) were ,18 or .100years of age at index date, 2) had anyHbA1c measurement$6.5% (48mmol/mol)prior to baseline, or if prior to base-line they were diagnosed with 3) anytype of diabetes (or were prescribedantidiabetes medication), 4) ischemicheart disease, 5) ischemic stroke ortransient ischemic attack, or 6) peripheral

vascular disease. A detailed description ofthe HbA1c assays and quality assessmenthas been published previously (17).

Data SourcesAll residents in Denmark are assigned aunique and permanent civil registrationnumber that enables linkage of admin-istrative registries on an individual level.The Danish National Patient Registryholds nationwide information on all hos-pital, outpatient clinic, and emergencyroom admissions since 1978. At dis-charge, each hospitalization or visit atan outpatient clinic or emergency roomis registered with one primary diagnosisand, if applicable, one ormore secondarydiagnosesaccording to the ICD. The ICD-8was used from 1978 to 1994 and the ICD-10 was used from 1995 onward. TheDanish Register of Medicinal ProductStatistics holds nationwide informationon all dispensed prescriptions from Dan-ish pharmacies since 1995 based on theAnatomical Therapeutic Chemical Clas-sification System (ATC) (18). Sex, date ofbirth and death, and date of emigrationand immigration can be obtained fromthe Danish Central Personal Registry.Annual data on personal income wereprovided by Statistics Denmark (19). Thecause of death, classified by ICD codes,can be obtained from the Danish Registerof Causes of Death (20). In accordancewith Danish law, no approval from anethics committee is needed in a registry-based study without any active partici-pation from study subjects. The use ofdeidentified registry data was approvedby the Danish Data Protection Agency(record number 2007-58-0015).

Baseline Variables and Follow-upWe defined hypertension from dis-charge diagnosis or if a subject priorto baseline was treated simultaneouslywith at least two types of antihyperten-sive drugs (Supplementary Tables 1 and2). The Charlson Comorbidity Index (CCI)was used to account for the generalmorbidity burden (21). The CCI takesmultiple comorbidities into account, in-cluding chronic obstructive pulmonarydisease, various cancer diseases, liverdiseases, renal disease, and cardiovas-cular diseases (Supplementary Table 3).Based on the total score, we categorizedpatients into levels of no comorbidity(score 0) and comorbidity (score $1).The score has been reported to have

high discrimination in predicting mor-tality (21,22). We separately ascertainedcontacts for obesity- and smoking-relateddisorders (23) and used these as cova-riates in a sensitivity analysis as describedbelow. Smoking was defined from dis-charge diagnosis for smoking or chronicobstructive pulmonary disease, or phar-macological or nonpharmacological treat-ment for smoking cessation. We definedsocioeconomic status as the individualmean annual gross income during a5-year period prior to baseline (24). In-come was categorized into tertiles. Theend points of interest were incidentMACE, death from all causes, and type2 diabetes. MACE represented a compos-ite end point of myocardial infarction,unstable angina pectoris, ischemic stroke,peripheral vascular disease, or cardiovas-cular death, whichever came first. Thediagnoses of myocardial infarction andischemic stroke have previously beenvalidated with high predicted values of93.6% and 79%, respectively (25,26). In-formation on cause of death was ex-tracted from the Danish Register ofCauses of Death (20). The type 2 diabetesend point was defined from dischargediagnoses of type 2 diabetes and/ortreatment with oral antidiabetes medica-tions, whichever came first. This algo-rithm has previously been shown tohave a sensitivity of 96% and a positivepredictive value of 89% for identifyingpeople with type 2 diabetes (27). Thealgorithms and registry diagnoses usedfor covariates and outcomes are listedin Supplementary Tables 1 and 2.

Statistical AnalysesThe primary exposure variable of interestwas visit-to-visit HbA1c variability definedas the intraindividual variability in HbA1clevels. The variability was defined as theSDof the residuals obtained froma linearregression analysis of the three HbA1cmeasurements of each individual, similarto what has been done previously (28).From the linear regression analysis, weused the intercept, the slope, and the SDof the residuals to characterize the pa-tients according to index level, trend, andvariability, respectively. Index HbA1c wascategorized into levels of low HbA1c(,5.3% [34 mmol/mol]), moderate HbA1c(5.3–5.7% [34–39 mmol/mol]), and highHbA1c (.5.7% [39 mmol/mol]). We cat-egorized the slope into decreasing (lessthan 20.11%/year), stable (20.11 to

care.diabetesjournals.org Ghouse and Associates 135

0.11%/year), and increasing (.0.11%/year) groups (Supplementary Fig. 1).These cutoffs were chosen based onmanual inspection of histograms show-ing the population distributions of indexHbA1c and slope, respectively, to obtaingroups with roughly equal sizes. We sub-sequently tested other cutoffs in a sen-sitivity analysis (see below).Time zero for all time-to-event anal-

yses was set at the date of the thirdHbA1c measurement (baseline). Individualfollow-up ended in case of the event ofinterest, death, emigration, or at 31 De-cember 2015, whichever occurred first.Multiple cause-specific Cox regressionwas used to evaluate the associationbetween HbA1c variability and the hazardrates of MACE, all-cause mortality, andtype 2 diabetes. Cause-specific hazardratios (HRs) were reported for a 1-SDincrease in variability (SupplementaryFig. 2). To test the linearity assumption,we used restricted cubic splines plotsto assess the functional relationshipbetween HbA1c variability and the endpoints (Supplementary Fig. 3). All Coxregression analyses included main andinteraction effects of the index HbA1clevel (low, moderate, and high) and thetrend (decreasing, stable, and increasing)and were further adjusted for main ef-fects of age-groups (,50, 50–60, 60–70,70–80, and .80 years), sex, baselinehypertension status (yes or no), andCCI (0 and $1).The 5-year risk of all-cause mortality

was predicted based on the Cox re-gression model for the hazard rate ofall-cause mortality. By combining aCox regression model for the compet-ing risk of noncardiovascular mortalitywith the Cox regression model for thehazard rate of MACE, we predicted the5-year absolute risks of MACE (29). Ab-solute 5-year risks were reported with95% confidence limits for selected riskprofiles defined by age-group, sex, hy-pertension, CCI, index HbA1c group (low,moderate, and high), and trend (de-creasing, stable, and increasing). Threerisk profiles were defined based on thecombination of risk factor values asso-ciated with the lowest, median, andhighest absolute 5-year risks. We alsoconstructed a risk chart that displays5-year absolute risks of MACE and all-cause mortality for a moderate indexHbA1c, stable trend, and all combina-tions of the other risk factors.

The potential effect modifications ofassociation between variability and theoutcome hazard rates by age, sex, CCI,hypertension, index HbA1c, and trendwere evaluated with likelihood ratio testscomparing the main model to a modelwith the interaction term. A two-sidedP , 0.05 was considered statisticallysignificant.

Sensitivity AnalysesTo test the robustness of our analyses, weconducted multiple sensitivity analyses.First, patients who increased in CCI fromfirst HbA1c measurement to last HbA1cmeasurement were excluded. This wasdone to assess whether HbA1c variabilityis merely a reflection of increased mor-bidity. Second, to test whether HbA1cvariability is a measure of generalizedfrailty, secondary to, for example, cancer,we excluded individuals with CCI $1 atbaseline. This is also of importance, since,for example, severe renal diseases andcompromised bone marrow function areassociated with altered red blood cellproduction and life span, affecting thevalidity of HbA1c as amarker of long-termblood glucose level. Third, to assesswhether the effect of HbA1c variabilityon the hazard rate of MACE and all-causemortality could be confounded by weightand smoking patterns, we adjusted forseparately ascertained registry diagno-ses of obesity and smoking-related dis-orders. Fourth, to assess whether amoreliberal time-related inclusion criterionmight influence the results, we includedindividuals with three measurementsspaced annually 64 months. Fifth, toassess whether the results are depen-dent on the number of measurementsused in the analyses, we conducted aseparate analysis including individualswith four measurements spaced annually63 months. Sixth, to assess whether theresults could be reproduced using an-other marker for glucose levels, we usedrandom blood glucose variability asexposure. From the CGPL database,we identified all individuals who hadthree or more measurements of randomblood glucose, taken in the nonfastingstate and annually spaced (63 months).Individuals with any measurement$11.1 mmol/L ($200 mg/dL) were ex-cluded. All other exclusion criteria, var-iability measurements, intercept, andtrend were obtained using the samemethods as described for the main

analyses. Seventh, we also used a dif-ferent index for HbA1c variability, re-cently published by Forbes et al. (30).Here, HbA1c variability was defined asthe number of times successive mea-surements differed by $0.5% (5.5mmol/mol), divided by the number ofcomparisons and then multiplied by100. Eighth, we used different cutoffsfor categorization of the slope estimateto test whether the position of the cut-off could influence the results.

RESULTS

Baseline Demographics and ClinicalCharacteristicsThe greater region of Copenhagen has acurrent population of 1.3 million citizens.At CGPL, 390,747 adults underwent HbA1ctesting. Of the individuals referred forHbA1c testing, 118,476 individuals hadat least three measurements of HbA1ctaken between 2003 and 2013. A totalof 6,756 individuals were eligible for in-clusion. Supplementary Fig. 4 summarizesthe number of included and excludedstudy subjects. Baseline clinical charac-teristics are summarized in Table 1.

HbA1c Variability and Risk of IncidentMACEThe median follow-up period was 6.3years (interquartile range 4.3–9.0). In to-tal, 996 individuals experienced a MACEduring the follow-up period. We foundan association between higher HbA1c var-iability and incident MACE (Table 2) (HR1.09 [95% CI 1.03–1.15]; P = 0.002, per1-SD [0.09%] increase in HbA1c variability).Associations between interaction effectof index HbA1c and trend categories andhazard of MACE (P value for effect mod-ification = 0.014) are summarized in Table2. We calculated absolute 5-year risks ofMACE for selected combinations of age-,sex-, and disease-specific risk factors(Fig. 1) along with extreme risk profiles(Supplementary Fig. 5) and found that therisk of MACE increased with increasinglevels of HbA1c variability.

HbA1c Variability and Risk of All-CauseMortalityA total of 856 individuals died duringfollow-up. We found an association be-tween higher HbA1c variability and thehazard of all-cause mortality (HR 1.13[95% CI 1.07–1.20]; P , 0.001, per 1-SDincrease in HbA1c variability). We foundthat the effect of trend on the hazardof all-cause mortality also depended on

136 HbA1c, Cardiovascular Events, and Death Diabetes Care Volume 42, January 2019

index HbA1c (P value for effect modifi-cation = 0.013). For example, the HR for adecreasing trend compared with a stabletrend was 1.92 (95% CI 1.14–3.21; P =0.013) (Table 2) in a low index HbA1ccategory, 1.54 (95% CI 1.12–2.13; P =0.009) in a moderate index HbA1c cate-gory, and not significant (1.03 [95% CI0.81–1.29]; P = 0.825) in a high indexHbA1c category. We calculated abso-lute 5-year risks of all-cause mortalityfor selected combinations of age-, sex-,and disease-specific risk factors (Sup-plementary Fig. 6) along with extremerisk profiles (Supplementary Fig. 5) andfound that HbA1c variability generallyconferred a lower absolute risk of all-cause mortality compared with therisk of MACE, at least in people ,70years of age (Fig. 1 and SupplementaryFig. 6).

HbA1c Variability and Risk of IncidentType 2 DiabetesDuring follow-up, 1,267 individuals de-veloped type 2 diabetes. We observedno statistically significant association be-tween HbA1c variability and type 2 di-abetes (HR 1.00 [95% CI 0.95–1.05]; P =

0.935, per 1-SD increase in variability)(Table 2). We found significant interac-tion effect between index HbA1c andtrends and type 2 diabetes (P valuefor effect modification ,0.001), whereincreasing trend compared with a stabletrend conferred an increased hazard oftype 2 diabetes, whereas a decreasingtrend was associated with a decreasedhazard of type 2 diabetes, for each cat-egory of index HbA1c (Table 2).

Interaction AnalysesWe found no statistically significant in-teractions between HbA1c variability andage-groups, sex, hypertension, CCI, indexHbA1c, and trend for outcome MACE, all-cause mortality, and type 2 diabetes.

Sensitivity AnalysesWe conducted eight sensitivity analysesto test the robustness of our results.As shown in Fig. 2 and SupplementaryTables 4 and 5, the associations betweenHbA1c variability and the three outcomesdid not materially change when theanalyses were restricted to different sub-sets, random blood glucose variabil-ity, use of a different variability index

or cutoffs for the slope estimate, oradjusted for various potential con-founders or common causes of HbA1cvariability.

CONCLUSIONS

We are, to the best of our knowledge, thefirst to describe an association betweenglycemic variability and the risk of in-cident MACE and all-cause mortality in apopulation without diabetes.

We found that HbA1c variability, mea-sured as the SD of the residuals, issignificantly associated with increasedrisks of incident MACE and all-causemortality, independent of traditional car-diovascular risk factors, index HbA1c, andtrend. The same effects were observedfor random blood glucose variability. Wedid not observe an association betweenHbA1c variability and incident type 2diabetes, but as expected, combinationsof index HbA1c and HbA1c trend weresignificantly associated with incidenttype 2 diabetes. Our findings suggestthat glycemic variability may containvaluable prognostic information forthe outcomes of MACE and mortality,

Table 1—Baseline characteristics of the study cohort

Trend

Total Decreasing Stable IncreasingCharacteristics (n = 6,756) (n = 1,462) (n = 3,483) (n = 1,811)

Measurement variablesHbA1c variability, % 0.08 (0.04, 0.15) 0.10 (0.04, 0.17) 0.07 (0.03, 0.13) 0.09 (0.04, 0.16)Index HbA1c category, %Low 5.1 (5.0, 5.2) 5.2 (5.1, 5.3) 5.1 (5.0, 5.2) 5.1 (4.9, 5.2)Moderate 5.5 (5.4, 5.6) 5.5 (5.4, 5.6) 5.5 (5.4, 5.6) 5.5 (5.4, 5.6)High 5.9 (5.8, 6.1) 6.0 (5.8, 6.2) 5.9 (5.8, 6.1) 5.8 (5.7, 5.9)

DemographicsAge, years 64.9 (56.3, 72.6) 64.7 (56.4, 72.6) 65.1 (56.4, 72.2) 64.7 (56.2, 72.2)Male 2,864 (42.4) 636 (43.5) 1,455 (41.8) 773 (42.7)

Average annual incomeFirst tertile 2,252 (33.3) 444 (30.4) 1,173 (33.7) 635 (35.1)Second tertile 2,252 (33.3) 514 (35.2) 1,147 (32.9) 591 (32.6)Third tertile 2,252 (33.3) 504 (34.5) 1,163 (33.4) 585 (32.3)

MedicationAcetylsalicylic acid 1,677 (24.8) 356 (25.0) 820 (23.5) 501 (27.7)Statin 2,818 (41.7) 607 (41.5) 1,395 (40.1) 816 (45.1)

Medical historyHypertension 3,216 (47.6) 668 (45.7) 1,676 (48.1) 872 (48.2)Obesity 241 (3.6) 50 (3.4) 114 (3.3) 77 (4.3)Smoking 417 (6.2) 88 (6.0) 234 (6.7) 95 (5.2)CCI0 point 5,117 (75.7) 1,088 (74.4) 2,635 (75.7) 1,394 (77.0)$1 point 1,639 (24.3) 374 (25.6) 848 (24.3) 417 (23.0)

Data are median (interquartile range) or n (%). Variability is defined as the SD of the residuals from the linear regression. Index HbA1c corresponds tothe intercept and is categorized as low (,5.3% [34 mmol/mol]), moderate (5.3–5.7% [34–39 mmol/mol]), and high (.5.7% [39 mmol/mol]) indexHbA1c. Trend corresponds to the slope estimate and is categorized as decreasing (less than 20.11%/year), stable (20.11 to 0.11%/year), andincreasing trend (.0.11%/year). Average annual income corresponds to mean annual gross income during a 5-year period prior to baseline.

care.diabetesjournals.org Ghouse and Associates 137

even in individuals with HbA1c levelswithin normal range.Several pathophysiological mechanisms

may be involved in the observed associ-ation between visit-to-visit glycemic var-iability andMACE and all-cause mortality.Mechanistic animal and human studieshave mainly focused on short-term vari-ability in blood glucose, and studies onlong-term variability, as reflected in HbA1clevels, have not been conducted. How-ever, intermittent high blood glucoseexposure, rather than chronic hyperglyce-mia, has been shown to have deleterious

effects on endothelial function, mediatedthrough oxidative stress (14,31). In vitrostudies have shown that intermittenthigh glucose levels stimulate overproduc-tion of reactive oxygen species, whichleads to increased cellular apoptosis inhuman umbilical vein endothelial cellscompared with a constant high-glucoseenvironment (14,15). Furthermore,in vivo studies on diabetic rats haveshown that rats with induced “glycemicswings” have higher levels of oxidativestress markers and endothelial dysfunc-tion, compared with rats with sustained

hyperglycemia (16). Fluctuations in bloodglucose concentrations have also beenassociated with an increase in circulatinginflammatory cytokines and acceleratedmacrophage adhesion to endothelialcells, stimulating the progression andformation of fibrotic atherosclerotic le-sions (32,33).

The association between HbA1c vari-ability and cardiovascular events that wereport could reflect that HbA1c is a proxyfor other systemic conditions that in-crease cardiovascular risk. Such systemicconditions could in theory lead to gen-eralized frailty, in which higher variabilityin numerous biological parameters con-fer risk through parallel pathologicalpathways. We have tried to accountfor this through various sensitivity anal-yses. First, we performed a sensitivityanalysis including only individuals with-out significant diseases. Hence, subjectswith severe systemic conditions such ascancer and kidney and liver disorderswere excluded at baseline. Second, weperformed a sensitivity analysis whereindividuals who increased in CCI duringthe inclusion period were excluded. Assuch, we accounted for individuals withan a priori increased risk of cardiovas-cular events due to a more rapid declinein overall health status. Although ourstudy design does not allow for casualinference, both of these sensitivity anal-yses gave results that were very similarto the main analyses (Fig. 2), supportingthat the associations we report are notdriven by yet undiagnosed severe sys-temic disorders.

We believe the present results mayhighlight an important issue. Patientswith large fluctuations in HbA1c levels,although with absolute measurementlevels within normal range, may befalsely reassured of their low cardiovas-cular risk, based on their HbA1c residingwithin standard guideline limits. How-ever, our results warrant replication inlarger studies, and future long-termstudies are needed to examine whetherstrategies to reduce variability in HbA1ccan effectively reduce risk of cardiovas-cular disease in subjects with HbA1cwithin the normal range.

In this study, we used three consec-utive annual measurements of HbA1c inthe normoglycemic range, allowing for amore detailed estimation of the effect oflongitudinal trajectories within HbA1ccategories. Although not statistically

Table 2—Results of event-specific Cox regression

Outcome Exposure HR 95% CI P value

MACEHbA1c variability (%) 1.09 1.03–1.15 0.002Low index HbA1cDecreasing vs. stable trend 1.78 1.08–2.94 0.025Increasing vs. stable trend 1.21 0.91–1.61 0.188

Moderate index HbA1cDecreasing vs. stable trend 1.18 0.88–1.58 0.274Increasing vs. stable trend 1.21 0.98–1.51 0.083

High index HbA1cDecreasing vs. stable trend 0.94 0.76–1.17 0.583

Increasing vs. stable trend 0.68 0.48–0.95 0.025

All-cause mortalityHbA1c variability (%) 1.13 1.07–1.20 ,0.001Low index HbA1cDecreasing vs. stable trend 1.92 1.14–3.21 0.013Increasing vs. stable trend 0.85 0.63–1.14 0.269

Moderate index HbA1cDecreasing vs. stable trend 1.54 1.12–2.13 0.009Increasing vs. stable trend 1.18 0.92–1.51 0.205

High index HbA1cDecreasing vs. stable trend 1.03 0.81–1.29 0.825

Increasing vs. stable trend 1.18 0.87–1.60 0.292

Type 2 diabetesHbA1c variability (%) 1.00 0.95–1.05 0.935Low index HbA1cDecreasing vs. stable trend 0.44 0.11–1.86 0.266Increasing vs. stable trend 2.74 1.78–4.21 ,0.001

Moderate index HbA1cDecreasing vs. stable trend 0.57 0.36–0.91 0.018Increasing vs. stable trend 3.40 2.72–4.26 ,0.001

High index HbA1cDecreasing vs. stable trend 0.70 0.59–0.82 ,0.001

Increasing vs. stable trend 1.78 1.49–2.13 ,0.001

Shown are effects of HbA1c variability, interaction effects of index HbA1c, and trend categories.HbA1c variability is reported in HR per SD increase in variability. Index HbA1c correspondsto the intercept and is categorized as low (,5.3% [34 mmol/mol]), moderate (5.3–5.7%[34–39 mmol/mol]), and high (.5.7% [39 mmol/mol]) index HbA1c. Trend correspondsto the slope estimate and is categorized as decreasing (less than 20.11%/year), stable(20.11 to 0.11%/year), and increasing trend (.0.11%/year). Interaction effects are assessedwithin each category of index HbA1c, comparing increasing/decreasing trend categorywith stable trend (reference) for each outcome. The Cox model was additionally adjustedfor age-groups, sex, hypertension, and CCI.

138 HbA1c, Cardiovascular Events, and Death Diabetes Care Volume 42, January 2019

significant for higher HbA1c levels, wefound indications of a J-shaped rela-tionship between trend in HbA1c and all-cause mortality, which is consistent withprevious studies (34–36). We also foundan increased risk of MACE in individualswith low HbA1c levels and a decreasingtrend. Suggestive risk increase in thelower glycemic region has also beenreported in previous studies of adultswithout diabetes (37–39). Our resultssuggest that clinicians should be atten-tive to individuals not only with increas-ing levels but also decreasing levels ofHbA1c over time, in particular to thosewith initially low HbA1c levels.

Strengths of this study include a largesample size, data on multiple visit-to-visit measurements of HbA1c, and up to13 years of follow-up. Despite thesestrengths, this study has importantlimitations. First, the successive HbA1cvalues were collected from availablelaboratory records as a part of the pa-tient’s routine clinical follow-up, and theindication for HbA1c referral is unknown.In Denmark, general practitioners referpatients for blood sampling based on avariety of clinical indications or as part ofan annual health check. Rather than care-fully selecting individual blood tests,patients may be referred for a collection

of blood samples in which HbA1c is in-cluded. As such, many primary carepatients will have HbA1c measured onindications other than screening fordiabetes. Second, we were unable tocontrol for alternating diets or life-style interventions causing fluctuatingweight, which could have influencedthe variability observed in the study.However, we adjusted for the individualHbA1c trend over time to account foran overall decline or improvement inhealth or lifestyle interventions. Third,we did not have direct access to clinicalvariables important for MACE and death,such as BMI and smoking. However, we

Figure 1—Five-year absolute risk prediction chart for MACE, based on individuals with a stable trend and moderate index HbA1c, for differentcombinations of risk factors and with respect to different levels of HbA1c variability (%). The color scheme refers to the absolute 5-year risk (%)for MACE.

care.diabetesjournals.org Ghouse and Associates 139

were able to account for this by usingregistry diagnoses for obesity and smok-ing, although we acknowledge thatthese conditions are underreported inthe registries. Last, there is a potentialfor selection bias due to exclusion ofindividuals without annually spaced mea-surements. This limitation is inherent tothe study design, which diminishes theeffect of time between successive valueson the variability index. We found thatincluded individuals were significantlyolder, had higher levels of HbA1c, andhad more concomitant medicationscompared with excluded individuals(Supplementary Table 6). However,we found similar degrees of comor-bidity between the included groupand the excluded group (with three ormoreHbA1c outside design period). Sincewe used three (and four) annual HbA1cmeasurements, our results can only be

extrapolated to other populations with-out diabetes with the same number ofmeasurements.

In conclusion, we found an associa-tion between high HbA1c variability andincreased risks of incident MACE anddeath from all causes. Both findingswere independent of traditional cardio-vascular risk factors, index HbA1c levels,and trends in HbA1c.

Funding. J.G. was supported by the ResearchFoundation at Copenhagen University Hospital,Rigshospitalet, KarenMarie Jespersen ogDattersLegat, and John og Birthe Meyer Foundation.J.B.N. was supported by grants from the DanishHeart Foundation (16-R107-A6779) and theLundbeck Foundation (R220-2016-1434), Fondentil Lægevidenskabens Fremme (the A.P. Møllerand Chastine Mc-Kinney Møller Foundationfor General Purposes), and FondsbørsvekselererHenry Hansen og Hustru Karla Hansen FødtVestergaards Legat.

Duality of Interest. L.K. has received a speakerhonorarium from Novartis. J.H.S. has receivedresearch grants from Medtronic, Biotronik, andGilead; personal fees as a speaker forMedtronic,Biotronik, and Boehringer Ingelheim; and per-sonal fees as a member of an advisory committeefor Medtronic. A.G.H. is an employee of NovoNordisk A/S, Denmark. No other potential con-flicts of interest relevant to this article werereported.Author Contributions. J.G., M.W.S., and J.B.N.made primary contributions to study conception,data analysis, interpretation of results, and writ-ing of themanuscript. J.K.K. and J.L.I. contributedto the interpretation of results and helped writethe first draft of the manuscript. P.B. and T.A.G.contributed to the study conception, design,and statistical analyses; wrote the statisticalanalysis plan; and contributed to the interpre-tation of the results. B.L. collected the data andcontributed to the interpretation of results.S.H., L.K., J.H.S., M.S.O., and A.G.H. contributedto the interpretation of results. All authors re-viewed and revised the manuscript critically forimportant intellectual content and approved thefinal version. J.B.N. is the guarantor of this work

Figure 2—Sensitivity analyses with the main model included for comparison. S1 includes only individuals who did not increase in CCI score between thefirst and third HbA1c measurement. S2 includes only individuals with a CCI of 0 at baseline. S3 is the main model additionally adjusted for registrydiagnoses of obesity and smoking-related disorders or contacts. S4 includes individuals with three annual measurements (64 months). S5 includesindividuals with four annual measurements (63 months). S6 includes random blood glucose variability. All models were adjusted as described in Table 2.

140 HbA1c, Cardiovascular Events, and Death Diabetes Care Volume 42, January 2019

and, as such, had full access to all the data in thestudy and takes responsibility for the integrity ofthe data and the accuracy of the data analysis.

References1. Danaei G, Finucane MM, Lu Y, et al.; GlobalBurden of Metabolic Risk Factors of ChronicDiseases Collaborating Group (Blood Glucose).National, regional, and global trends in fastingplasma glucose and diabetes prevalence since1980: systematic analysis of health examinationsurveys and epidemiological studies with 370country-years and 2.7 million participants. Lan-cet 2011;378:31–402. Gregg EW, Cheng YJ, Saydah S, et al. Trends indeath rates among U.S. adults with and withoutdiabetes between 1997 and 2006: findings fromthe National Health Interview Survey. DiabetesCare 2012;35:1252–12573. American Diabetes Association. Diagnosisand classification of diabetes mellitus. DiabetesCare 2011;34(Suppl. 1):S62–S694. NathanDM,GenuthS, Lachin J, etal.;DiabetesControl and Complications Trial Research Group.The effect of intensive treatment of diabetes onthe development and progression of long-termcomplications in insulin-dependent diabetesmellitus. N Engl J Med 1993;329:977–986.5. Abraira C, Colwell J, Nuttall F, et al.; VeteransAffairs Cooperative Study on Glycemic Controland Complications in Type II Diabetes. Cardio-vascular events and correlates in the VeteransAffairs Diabetes Feasibility Trial. Arch InternMed 1997;157:181–1886. UK Prospective Diabetes Study (UKPDS)Group. Intensive blood-glucose control withsulphonylureas or insulin compared with con-ventional treatment and risk of complications inpatients with type 2 diabetes (UKPDS 33). Lancet1998;352:837–8537. Turnbull FM, Abraira C, Anderson RJ, et al.;Control Group. Intensive glucose control andmacrovascular outcomes in type 2 diabetes.Diabetologia 2009;52:2288–22988. Currie CJ, Peters JR, Tynan A, et al. Survivalas a function of HbA1c in people with type 2diabetes: a retrospective cohort study. Lancet2010;375:481–4899. Kilpatrick ES, Rigby AS, Atkin SL. A1C variabilityand the risk of microvascular complications intype 1 diabetes: data from the Diabetes Controland Complications Trial. Diabetes Care 2008;31:2198–220210. Waden J, Forsblom C, Thorn LM, Gordin D,Saraheimo M, Groop PH; Finnish Diabetic Ne-phropathy Study Group. A1C variability predictsincident cardiovascular events, microalbuminu-ria, and overt diabetic nephropathy in patientswith type 1 diabetes. Diabetes 2009;58:2649–265511. Takao T, Matsuyama Y, Yanagisawa H,Kikuchi M, Kawazu S. Association betweenHbA1c variability and mortality in patientswith type 2 diabetes. J Diabetes Complications2014;28:494–49912. Hermann JM, Hammes H-P, Rami-Merhar B,et al.; DPV Initiative the German BMBF Com-petence Network Diabetes Mellitus. HbA1c

variability as an independent risk factor fordiabetic retinopathy in type 1 diabetes: a Ger-man/Austrian multicenter analysis on 35,891patients. PLoS One 2014;9:e9113713. Lin C-C, Chen C-C, Chen F-N, et al. Risks ofdiabetic nephropathy with variation in hemo-globin A1c and fasting plasma glucose. Am J Med2013;126:1017.e1–1017.e1014. Quagliaro L, Piconi L, Assaloni R,Martinelli L,Motz E, Ceriello A. Intermittent high glucoseenhances apoptosis related to oxidative stress inhuman umbilical vein endothelial cells: the roleof protein kinase C and NAD(P)H-oxidase acti-vation. Diabetes 2003;52:2795–280415. Risso A, Mercuri F, Quagliaro L, Damante G,Ceriello A. Intermittent high glucose enhancesapoptosis in human umbilical vein endothelialcells in culture. Am J Physiol Endocrinol Metab2001;281:E924–E93016. Horvath EM, Benko R, Kiss L, et al. Rapid‘glycaemic swings’ induce nitrosative stress, ac-tivate poly(ADP-ribose) polymerase and impairendothelial function in a rat model of diabetesmellitus. Diabetologia 2009;52:952–96117. Borg R, Persson F, Siersma V, Lind B, de FineOlivarius N, Andersen CL. Interpretation of HbA1cin primary care and potential influence of anae-mia and chronic kidney disease: an analysis fromthe Copenhagen Primary Care Laboratory (CopLab)Database. Diabet Med 2018;35:1700–170618. Kildemoes HW, Sørensen HT, Hallas J. TheDanish National Prescription Registry. Scand JPublic Health 2011;39(Suppl.):38–4119. Baadsgaard M, Quitzau J. Danish registerson personal income and transfer payments.Scand J Public Health 2011;39(Suppl.):103–10520. Helweg-Larsen K. The Danish register ofcauses of death. Scand J Public Health 2011;39(Suppl.):26–2921. Charlson ME, Pompei P, Ales KL, MacKenzieCR. A new method of classifying prognostic co-morbidity in longitudinal studies: developmentand validation. J Chronic Dis 1987;40:373–38322. Quan H, Li B, Couris CM, et al. Updating andvalidating the Charlson comorbidity index andscore for risk adjustment in hospital dischargeabstracts using data from 6 countries. Am JEpidemiol 2011;173:676–68223. KhalidU,EgebergA,AhlehoffO,SmedegaardL, Gislason GH, Hansen PR. Nationwide study onthe risk of abdominal aortic aneurysms in pa-tients with psoriasis. Arterioscler Thromb VascBiol 2016;36:1043–104824. Schjerning Olsen A-M, Fosbøl EL, LindhardsenJ, et al. Duration of treatment with nonsteroidalanti-inflammatory drugs and impact on risk ofdeath and recurrent myocardial infarction inpatients with prior myocardial infarction: a na-tionwide cohort study. Circulation 2011;123:2226–223525. Madsen M, Davidsen M, Rasmussen S,Abildstrom SZ, Osler M. The validity of thediagnosis of acute myocardial infarction in rou-tine statistics: a comparison of mortality andhospital discharge data with the DanishMONICAregistry. J Clin Epidemiol 2003;56:124–13026. Olesen JB, Lip GYH, Hansen ML, et al. Val-idation of risk stratification schemes for

predicting stroke and thromboembolism in pa-tients with atrial fibrillation: nationwide cohortstudy. BMJ 2011;342:d12427. Kristensen JK, Drivsholm TB, Carstensen B,Steding-Jensen M, Green A. Validation of meth-ods to identify known diabetes on the basis ofhealth registers. Ugeskr Laeger 2007;169:1687–1692 [in Danish]28. Skriver MV, Sandbæk A, Kristensen JK,Støvring H. Relationship of HbA1c variability,absolute changes in HbA1c, and all-cause mor-tality in type 2 diabetes: a Danish population-based prospective observational study. BMJOpen Diabetes Res Care 2015;3:e00006029. Rosthøj S, Andersen PK, Abildstrom SZ. SASmacros for estimation of the cumulative inci-dence functions basedon aCox regressionmodelfor competing risks survival data. Comput Meth-ods Programs Biomed 2004;74:69–7530. Forbes A, Murrells T, Mulnier H, Sinclair AJ.Mean HbA1c, HbA1c variability, and mortality inpeople with diabetes aged 70 years and older:a retrospective cohort study. Lancet DiabetesEndocrinol 2018;6:476–48631. Ceriello A, Esposito K, Piconi L, et al. Oscil-lating glucose is more deleterious to endothelialfunction and oxidative stress than mean glucosein normal and type 2 diabetic patients. Diabetes2008;57:1349–135432. Tanaka A, Azuma K, Toyofuku Y, et al. Insulinand nateglinide reduce monocyte adhesion toendothelial cells in Goto-Kakizaki rats exhibitingrepetitive blood glucose fluctuation. BiochemBiophys Res Commun 2006;350:195–20133. Esposito K, Nappo F, Marfella R, et al. In-flammatory cytokine concentrations are acutelyincreased by hyperglycemia in humans: role ofoxidative stress. Circulation 2002;106:2067–207234. Khaw K-T, Wareham N, Bingham S, Luben R,Welch A, Day N. Association of hemoglobin A1cwith cardiovascular disease and mortality inadults: the European Prospective Investigationinto Cancer in Norfolk. Ann Intern Med 2004;141:413–42035. SaydahS, TaoM, ImperatoreG,GreggE.GHblevel and subsequent mortality among adults inthe U.S. Diabetes Care 2009;32:1440–144636. Selvin E, Steffes MW, Zhu H, et al. Glycatedhemoglobin, diabetes, and cardiovascular riskin nondiabetic adults. N Engl J Med 2010;362:800–81137. Selvin E, Coresh J, Golden SH, Brancati FL,Folsom AR, Steffes MW. Glycemic control andcoronary heart disease risk in persons with andwithout diabetes: the Atherosclerosis Risk inCommunities Study. Arch Intern Med 2005;165:1910–191638. Brewer N, Wright CS, Travier N, et al. A NewZealand linkage study examining the associa-tions between A1C concentration and mortality.Diabetes Care 2008;31:1144–114939. Bancks MP, Odegaard AO, Pankow JS, et al.Glycated hemoglobin and all-cause and cause-specific mortality in Singaporean Chinese with-out diagnosed diabetes: the Singapore ChineseHealth Study. Diabetes Care 2014;37:3180–3187

care.diabetesjournals.org Ghouse and Associates 141