The Presence of Diabetes andHigher HbA1c Are IndependentlyAssociated With AdverseOutcomes After SurgeryDiabetes Care 2018;41:1172–1179 | https://doi.org/10.2337/dc17-2304

OBJECTIVE

Limited studies have examined the association between diabetes and HbA1c withpostoperative outcomes. We investigated the association of diabetes, definedcategorically, and the association of HbA1c as a continuous measure, with post-operative outcomes.

RESEARCH DESIGN AND METHODS

In thisprospective,observational study,wemeasured theHbA1cof surgical inpatientsage ‡54 years at a tertiary hospital between May 2013 and January 2016. Patientswere diagnosedwith diabetes if they had preexisting diabetes or anHbA1c‡6.5% (48mmol/mol) or with prediabetes if they had an HbA1c between 5.7 and 6.4% (39 and48 mmol/mol). Patients with an HbA1c <5.7% (39 mmol/mol) were categorized ashaving normoglycemia. Baseline demographic and clinical data were obtained fromhospital records, and patients were followed for 6 months. Random-effects logisticand negative binomial regression models were used for analysis, treating surgicalunits as random effects. We undertook classification and regression tree (CART)analysis to design a 6-month mortality risk model.

RESULTS

Of 7,565 inpatients, 30% had diabetes, and 37% had prediabetes. After adjusting forage, Charlson comorbidity index (excluding diabetes and age), estimated glomerularfiltration rate, and lengthof surgery, diabeteswas associatedwith increased6-monthmortality (adjusted odds ratio [aOR] 1.29 [95% CI 1.05–1.58]; P = 0.014), majorcomplications (1.32 [1.14–1.52]; P < 0.001), intensive care unit (ICU) admission (1.50[1.28–1.75]; P < 0.001), mechanical ventilation (1.67 [1.32–2.10]; P < 0.001), andhospital length of stay (LOS) (adjusted incidence rate ratio [aIRR] 1.08 [95% CI 1.04–1.12]; P < 0.001). Each percentage increase in HbA1c was associated with increasedmajor complications (aOR1.07 [1.01–1.14];P=0.030), ICUadmission (aOR1.14 [1.07–1.21]; P < 0.001), and hospital LOS (aIRR 1.05 [1.03–1.06]; P < 0.001). CART analysisconfirmed a higher risk of 6-monthmortality with diabetes in conjunctionwith otherrisk factors.

CONCLUSIONS

Almost one-third of surgical inpatients age ‡54 years had diabetes. Diabetes andhigher HbA1c were independently associated with a higher risk of adverse outcomesafter surgery.

1Department of Medicine, Austin Health, TheUniversity of Melbourne, Heidelberg, Victoria,Australia2Department of Anaesthesia, Austin Health,Heidelberg, Victoria, Australia3Department of Endocrinology, Austin Health,Heidelberg, Victoria, Australia4The Florey Institute of Neuroscience & MentalHealth, Melbourne, Victoria, Australia5Department of Administrative Informatics, AustinHealth, Heidelberg, Victoria, Australia6Clinical Costing Finance, Austin Health, Heidel-berg, Victoria, Australia7Department of Intensive Care, Austin Health,Heidelberg, Victoria, Australia8Centre for Integrated Critical Care, The Univer-sity of Melbourne, Parkville, Victoria, Australia9Australian and New Zealand Intensive CareSociety Research Centre, School of Public Healthand Preventive Medicine, Monash University,Melbourne, Victoria, Australia10Department of Pathology, Austin Health, Hei-delberg, Victoria, Australia11Clinical Informatics Unit, Austin Health, Heidel-berg, Victoria, Australia12Section of Anesthesia and Intensive CareMedicine, Department of Physiology and Phar-macology, Karolinska Institutet, Stockholm, Swe-den13Quality and Patient Safety Unit, Austin Health,Heidelberg, Victoria, Australia14Department of General Medicine, AustinHealth, Heidelberg, Victoria, Australia

Corresponding author: Elif I. Ekinci, elif.ekinci@unimelb.edu.au.

Received 3 November 2017 and accepted 26February 2018.

This article contains Supplementary Data onlineat http://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc17-2304/-/DC1.

This article is featured in a podcast available athttp://www.diabetesjournals.org/content/diabetes-core-update-podcasts.

© 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.

Priscilla H. Yong,1 Laurence Weinberg,2

Niloufar Torkamani,1,3 Leonid Churilov,4

Raymond J. Robbins,5 Ronald Ma,6

Rinaldo Bellomo,7,8,9 Que T. Lam,10

James D. Burns,11 Graeme K. Hart,7

Jeremy F. Lew,1 Johan Martensson,7,12

David Story,2,8 Andrew N. Motley,13

Douglas Johnson,1,14 JeffreyD. Zajac,1,3 and

Elif I. Ekinci1,3

1172 Diabetes Care Volume 41, June 2018

EPIDEM

IOLO

GY/HEA

LTHSERVICES

RESEA

RCH

The prevalence of diabetes is rising,whichin 2015 was estimated worldwide at415 million (8.8%) (1). In Australia, theprevalence of diabetes was 1.2 million(5.1%) in 2014 and 2015 (2). One-third ofinpatients age $54 years at a tertiaryreferral center were reported to havediabetes (3).Surgical procedures place physical

stress on patients and can lead to mor-bidity and mortality (4). Patients withdiabetes and chronic hyperglycemia, mea-sured by HbA1c, may be at particular riskfor perioperativemorbidity from diabetes-related complications (5,6). Althoughdiabetes has been associated with in-creased mortality and morbidity in thesetting of cardiac surgery (7,8), this asso-ciation in the noncardiac surgery setting isvariable (9–14). Inconsistency remains inboth the reporting and the definitions ofsurgical complications and periopera-tive morbidity among studies, making itdifficult to draw conclusions on the as-sociation between diabetes and post-operative complications. Furthermore,previous studies have used medical re-cords alone to identify diabetes in sur-gical patients (7,12,14), whichmay fail toidentify the up to 18% who may have di-abetes (15).Studies investigating the association of

HbA1cwithsurgicaloutcomeshaveshownconflicting results (16), with some dem-onstrating associations with highermortality (14,17), infection (17,18), myo-cardial infarction (17), renal failure (17),cerebrovascularaccident (17),major com-plications (19), and hospital length of stay(LOS) (20) and others failing to observesuch associations (14,21). Of note, manystudieshad limitedHbA1c results availablefor analysis, introducing selection bias(14,18–20), whereas others had limitedsample sizes (19).Most studies assessing the association

of hyperglycemia and postoperative out-comes have focused on perioperativehyperglycemia by using blood glucosereadings around the time of surgery(21,22). However, perioperative hyper-glycemia is not an accurate indicator ofdiabetes status because it is affected byperioperative fasting and stress hypergly-cemia from surgical trauma (23). HbA1chas been proposed as a reliable indicatorof glycemic status in the inpatient settingbecause it is unaffected by fasting statusand less affected by stress hyperglyce-mia (24) and has been endorsed as an

appropriate method of diagnosing diabe-tes (6,25).

Accordingly, in this prospective study,we used HbA1c to determine both pres-ence of diabetes and severity of chronicpreadmission glycemic status. We testedthe hypothesis that diabetes, defined as acategorical variable, and HbA1c, definedas a continuous variable, carry an inde-pendent association with adverse out-comes after surgery.

RESEARCH DESIGN AND METHODS

Study DesignWe performed a prospective, observa-tional study in surgical inpatients admit-ted to Austin Health, a tertiary teachinghospital affiliated with the University ofMelbourne in Melbourne, Victoria, Aus-tralia. During the period of 6 May 2013to 23 January 2016, as part of a process,labeled the Diabetes Discovery Initiative,all patients age $54 years without anHbA1c reading within 3 months of admis-sion received an automatic HbA1c mea-surement on hospital admission (3). Thisage cutoff was chosen on the basis of aprevious study that usedHbA1cmeasure-ments to demonstrate a higher preva-lenceofundiagnoseddiabetes in inpatientsaged.54years (26).HbA1cwasmeasuredby immunoassay on a COBAS INTEGRA800 (Roche Diagnostics, Indianapolis, IN).This study was approved by the AustinHealth Research Ethics Committee (LNR/15/Austin/41), whichwaived the need forinformed consent for a planned practicechange agreed to by hospital senior med-ical staff members as part of the DiabetesDiscovery Initiative.

As part of the Diabetes Discovery Ini-tiative, patients with HbA1c $8.3% (67mmol/mol) were seen by an endocrinol-ogy advanced trainee who generated apersonalized plan for glycemic control.Patients undergoing high-risk surgery,including cardiac, orthopedic, and gen-eral surgery, with HbA1c between 7.5%(58 mmol/mol) and 8.2% (66 mmol/mol)and patients with newly diagnosed di-abeteswere seenbythe internalmedicineadvanced trainee.Patientsweremanagedaccording to the hospital’s perioperativeguidelinesforpatientswithdiabetes,whichare based on Australian Diabetes Societyguidelines (27).

Inclusion criteria were age $54 years,surgery with at least one overnight hos-pital stay,HbA1cmeasurementperformed,3months before or 7 days after surgery,

and an available serum creatinine level.The cutoff date for HbA1c measurementwas set at 7 days after surgery to capturepatients with delays in HbA1c measure-ment while minimizing interference inHbA1c levels from surgery. We excludedpatients undergoing minor interventionalor noninterventional procedures, such asgastroscopy, colonoscopy, bronchoscopy,lumbar puncture, electroconvulsive ther-apy, direct cardioversion, paravertebralblock, video fluoroscopy swallowing orurodynamic studies, and MRI. Patientsalsowereexcludedif theyreceivedabloodtransfusion#90 days before HbA1c mea-surement because blood transfusionsaffect the reliability of HbA1c results (6).Where patients had more than one sur-gical procedure in their hospital episode,analyses were performed on the basis ofthe first procedure.

Detailed demographic and clinical datawere entered into a Cerner electronicdatabase (Cerner Corporation, KansasCity, MO). Baseline data collected wereage, sex, serum creatinine level, HbA1cresult, diabetes status, separation unit,type of surgery, length of surgery, elec-tive or emergency status of surgery,and comorbid conditions (SupplementaryTable 1).

Patients were followed for 6 months.The primary study outcomewasmortalitywithin 6 months. Secondary outcomeswere presence of a major complication,admission to an intensive care unit(ICU), requirement for mechanical venti-lation, hospital LOS, readmission within6 months, and cost of hospital episode.We included the STROBE (Strengtheningthe Reporting of Observational Studiesin Epidemiology) statement checklist forobservational studies to report findings(28).

DefinitionsPatients were diagnosed with diabetes ifthey had an HbA1c$6.5% (48mmol/mol)or preexisting diagnosis of diabetes onmedical records regardless of HbA1c level.PatientswerediagnosedwithprediabetesiftheyhadanHbA1c$5.7%(39mmol/mol)and,6.5% (48 mmol/mol). Patients withan HbA1c ,5.7% (39 mmol/mol) wereconsidered to have normoglycemia. Pa-tients with prediabetes and normoglyce-miawereclassifiedasnothavingdiabetes.These definitions are in accordance withthe International Expert Committee andAmerican Diabetes Association (6,25).

care.diabetesjournals.org Yong and Associates 1173

LOS was determined by the periodfrom admission to discharge, includingdays in the hospital-in-the-home unit.Readmission was defined as unplannedreadmission to this hospital within the6-month follow-up period. Mechanicalventilation referred to the requirementof such postoperatively. Mortality wasestablished if patientdeathhadoccurredduring admission or had been reportedto the hospital during the 6-month studyperiod.Hospital-acquired complications were

graded on the basis of the Clavien-Dindoclassification guide for surgical complica-tions (29). TheClavien-Dindoclassificationisavalidatedapproachtosurgicaloutcomeassessment that assigns severity grades tosurgical complications. Surgical complica-tions were assessed from diagnoses ofin-hospital complications frommedical re-cords, withmajor complication defined asahospital-acquiredcomplicationofClavien-Dindograde$4andincludedlife-threateningcomplications and in-hospital death (Sup-plementary Table 2). Our institution com-prises specialist wards equipped tomanage severe conditions thatmight oth-erwise be managed in the ICU. Thus, wealso included life-threatening complica-tions involving documented organ failureunable to be managed by medications orsurgical interventions alone, irrespectiveof ICU admission. In case of disagreementon grading by two assessors, the case wasdiscussed with reference to the classifica-tion guide (29).We a priori identified confounders as

comorbid conditions, age, renal function,and length of surgery. Comorbid condi-tions reported as ICD-10, AustralianMod-ification, codes were used to calculateeachpatient’sCharlsoncomorbidity indexscore. This validated method weights theimpact of chronic disease by assigningscores to chronic conditions on the basisof severity and effect on mortality (30)(SupplementaryTable1).Diabetesandagewereexcludedbecausetheywereassessedas separate characteristics. Estimated glo-merularfiltrationrate(eGFR)wascalculatedby the Chronic Kidney Disease Epidemi-ology Collaboration (CKD-EPI) equationby using extracted data on age, sex, andfirst creatinine value in the hospital epi-sode (31).Length of surgery in minutes was used

asamarker for severityof surgery.Anotherpotential marker of severity is emergencyor nonemergency status of surgery, with

emergency surgery contributing to poorersurgical outcomes (32). In addition, pa-tients were categorized into 15 surgicalunits according to the separation unitfrom which they were discharged (Table1). Patients with a separation unit outsidethese 15 units were assigned by twoindependent assessors to a unit on thebasis of their surgical procedure. Theseunits were used as random effects in thestatisticalmodels.Subgroupanalyseswereperformed for surgical units.

Statistical AnalysisStatistical analysis was performed withStata 14.2 software (StataCorp, CollegeStation, TX). Baseline characteristics werereported as medians and interquartileranges (IQRs) (continuous characteristics)or counts and percentages (categoricalcharacteristics) and compared betweenpatients included in and excluded fromthe study and by diabetes category usingKruskal-Wallis and x2/Fisher exact tests,respectively. Multivariable analyses wereconducted using random-effects nega-tive binomial regression for LOS outcome

(presented as count of days) and random-effects logistic regression for binary out-comes,withsurgicalunitsasrandomeffects.Two analyses were performed: 1) with di-abetes status classified as diabetes or nodiabetes (including prediabetes and nor-moglycemia)and2)withHbA1casa contin-uous marker. A priori chosen adjustmentcovariates were age, eGFR, Charlson co-morbidity index (excluding diabetes andage),andlengthofsurgeryinminutes.Stan-dard analyses of collinearity andmodel fitwere performed. Because emergency sur-geries had statistically significantly shorterlength than nonemergency surgeries, toavoid excessive collinearity, emergencystatus of surgery was not included as acovariate in addition to length of surgerybut instead was used in robustness anal-ysis. Exploratory subgroup analysis by sur-gical unit was performed and presented asforest plots. The cost of a patient episodewas described using median and IQRsand analyzed byWilcoxon rank sum test. Atwo-sided P, 0.05was considered statis-tically significant.

Table 1—Baseline patient characteristics

Baseline characteristic Normoglycemia Prediabetes Diabetes P value*

Patients 32 (2,457) 37 (2,825) 30 (2,283)

Male sex 57 (1,395) 54 (1,534) 62 (1,424) ,0.001

Age (years) 68 (61, 77) 71 (63, 80) 71 (64, 78) ,0.001

HbA1c ,0.001% 5.4 (5.3, 5.5) 5.9 (5.8, 6.1) 6.9 (6.3, 7.8)mmol/mol 36 (34, 37) 41 (40, 43) 52 (45, 62)

eGFR (mL/min/1.73 m2)† 79 (58, 92) 76 (58, 90) 67 (44, 86) ,0.001

CCI (excluding diabetes and age)‡ 0 (0, 2) 0 (0, 2) 1 (0, 2) ,0.001

Length of surgery (min) 150 (94, 230) 155 (100, 234) 149 (86, 250) 0.146

Emergency surgery 56 (1,366) 54 (1,532) 55 (1,263) 0.571

Surgical unit ,0.001Breast surgery 0.5 (11) 0.5 (14) 0.6 (13)Cardiac surgery 5.7 (139) 6.8 (192) 10.6 (243)Colorectal surgery 7.0 (172) 6.7 (189) 5.0 (113)ENT, HN, ophthalmology 2.2 (55) 2.1 (60) 2.4 (54)Hepatobiliary 9.8 (241) 8.8 (249) 9.8 (223)Liver transplant 1.7 (41) 0.4 (11) 1.7 (38)Maxillary facial 1.2 (29) 1.0 (27) 0.7 (15)Neurosurgery 10.3 (252) 10.5 (297) 7.7 (175)Orthogeriatrics 15.1 (371) 11.1 (313) 7.5 (171)Orthopedics 9.6 (236) 16.2 (457) 13.5 (307)Plastic surgery 6.8 (167) 6.7 (188) 5.1 (117)Thoracic surgery 4.8 (117) 5.7 (162) 3.9 (90)Upper gastrointestinal surgery 9.9 (243) 7.8 (220) 7.2 (164)Urology 9.8 (241) 10.0 (281) 10.1 (231)Vascular surgery 5.8 (142) 5.8 (165) 14.4 (329)

Data are % (n) or median (IQR). CCI, Charlson comorbidity index; ENT, ear nose throat; HN, headand neck. *P valueswere determinedby Fisher exact test for categorical variables and Kruskal-Wallistest forcontinuousvariables.†Derivedusing theCKD-EPIequation.‡Avalidatedmethodofweightingchronic medical conditions (the scores for diabetes and age were excluded because they wereanalyzed as a separate variable).

1174 Diabetes, HbA1c, and Surgical Outcomes Diabetes Care Volume 41, June 2018

We also used classification and regres-sion tree (CART)advancedanalysis (SalfordPredictive Modeler 7 software; SalfordSystems, San Diego, CA) to investigatecomplex interactions among patient char-acteristics, HbA1c, length of surgery, andoutcome of 6-month mortality and to de-sign a predictive 6-month mortality riskmodel. CART is a binary partitioning sta-tistical method that starts with the totalsample and, in a stepwise manner, splitsthe sample into subsamples that are ho-mogenous with respect to a defined out-come(33). The inputvariable thatachievesthemost effective split is dichotomized byautomated analysis at an optimal thresh-old,maximizingthehomogeneitywithinandseparation between resulting subgroups.Toestablishpredictivevalidityof theCARTmodel, itsperformance isassessedbyareaunder the receiver operating characteris-tic (ROC) curve with a 10-fold internalcross-validation,wheredataare randomlydivided into10groupswith9usedtobuildthemodel (training) and1usedtovalidate(testing).Fromthis,CARTgeneratesaclas-sification treeandnumerical rank for eachinput used to build the tree by relativeimportance. Our CART analysis variableswere age, eGFR, length of surgery, pres-ence of diabetes, and Charlson comorbid-ity index (excluding diabetes and age) asinputs.

RESULTS

During the study period, 12,128 patientsage$54 years underwent surgery at AustinHealth. The following patients were ex-cluded: those with no HbA1c result withinthespecifiedtimeframe(n=1,648),minorinterventional or noninterventional pro-cedure (n = 1,947), blood transfusion#90daysbeforeHbA1cmeasurement(n=286),andno available creatinine value (n = 682)(Fig. 1). Thefinal studypopulation included7,565 patients of whom 2,283 (30%) haddiabetes, 2,825 (37%) had prediabetes,and 2,457 (32%) had normoglycemia. Ofpatients with diabetes, 236 (10%) werepreviouslyundiagnosed.Patients includedin the study had a higher median eGFR,lower Charlson comorbidity index, longermedian length of surgery, and smaller pro-portionwith diabetes than those excluded(Supplementary Table 3).The clinical characteristics of the study

populationaredescribedinTable1.Median(IQR) ageofpatientswithnormoglycemia,prediabetes, and diabetes was 68 (61,77), 71 (63, 80), and 71 (64, 78) years,

respectively (P , 0.001). The majority ofpatients in all groups were male. Median(IQR) HbA1c in patients with normoglyce-mia, prediabetes, and diabetes was 5.4%(5.3, 5.5) (36 mmol/mol [34, 37]), 5.9%(5.8, 6.1) (41 mmol/mol [40, 43]), and6.9% (6.3, 7.8) (52 mmol/mol [45, 62]),respectively (P , 0.001). Of the surgicalprocedures studied, 3,404 (45%) wereelectiveand4,161 (55%)wereemergency.The proportion of patients with diabeteswho received diabetes medication is pre-sented in Supplementary Fig. 1. The asso-ciationofdiabetes andHbA1cwithsurgicaloutcomes is summarized in Fig. 2.

Primary Outcome: Six-Month MortalityThe incidence of 6-month mortality was6% (95% CI 5.5–6.8%) in patients withoutdiabetes and 9% (95% CI 7.4–9.7%) inpatients with diabetes. On multivariableanalysis, presence of diabetes was asso-ciated with increased mortality 6 monthsafter surgery (adjusted odds ratio [aOR]1.29 [95%CI1.05–1.58];P=0.014) (Fig. 2).No statistically significant association be-tween HbA1c on a continuous scale and6-month mortality was identified.

Secondary Outcomes

Major Complications

Amajor complication, defined as Clavien-Dindo grade$4,was present in 14% (95%

CI 13–15%) of patients without diabetesand 21% (95% CI 20–23) of patients withdiabetes. On multivariable analysis, pres-ence of diabetes as a categorical variablewas associated with greater risk of majorcomplications (aOR 1.32 [95% CI 1.14–1.52]; P, 0.001) (Fig. 2). When assessedas a continuous variable, each1% increasein HbA1c was associated with greater riskof major complications (aOR 1.07 [95% CI1.01–1.14]; P = 0.030) (Fig. 2).

ICU Admission

Eighteen percent (95% CI 17–19) ofpatients without diabetes and 27% (95%CI26–29)withdiabeteswereadmittedtothe ICU. On multivariable analysis, pres-ence of diabetes as a categorical variablewas associated with an increased likeli-hood of ICU admission (aOR 1.50 [95% CI1.28–1.75]; P , 0.001) (Fig. 2). Whenassessed as a continuous variable, each1% increase in HbA1c was associated withan increased likelihood of ICU admission(aOR 1.14 [95% CI 1.07–1.21]; P, 0.001)(Fig. 2).

Mechanical Ventilation

Mechanical ventilation was applied to10% (95% CI 9–11) of patients withoutdiabetes and 16% (95% CI 15–18) withdiabetes. On multivariable analysis, pres-ence of diabetes as a categorical variable

Figure 1—Flowchart of patient population according to inclusion and exclusion criteria.

care.diabetesjournals.org Yong and Associates 1175

was associated with an increased likeli-hood of receiving mechanical ventilation(aOR 1.67 [95% CI 1.32–2.10]; P, 0.001)(Fig. 2). No statistically significant associ-ationbetweenHbA1conacontinuousscaleandmechanical ventilationwas identified.

LOS

The median (IQR) hospital LOS was 6 (3,11) days in patients without diabetes and7 (4, 14) days in those with diabetes. Onmultivariable analysis, presence of diabe-tesasacategoricalvariablewasassociatedwith an increased hospital LOS (adjustedincidence rate ratio [aIRR] 1.08 [95%CI 1.04–1.12]; P , 0.001) (Fig. 2). Whenassessed as a continuous variable, each1% increase in HbA1c was associated withan increased hospital LOS (aIRR 1.05 [95%CI 1.03–1.06]; P , 0.001) (Fig. 2).

Six-Month Readmission

The incidence of 6-month readmissionwas 16% (95% CI 15–17) in patientswithout diabetes and 17% (95% CI 15–19) in those with diabetes. On multivari-able analysis, no statistically significantassociation between diabetes as a cat-egorical variable or HbA1c as a continuousvariable with 6-month readmission wasidentified (Fig. 2).

Episode Cost

The median episode cost of the studypopulation was $18,189, with patients

without diabetes having a median (IQR)cost of $17,439 (11,438, 27,564) com-pared with $20,440 (12,186, 33,261) forpatients with diabetes (P , 0.001).

Robustness AnalysisBecause of excessive collinearity withlength of surgery, emergency status ofsurgery was not included as a covariate inthe original model. When adjusting foremergency status insteadof or in additionto length of surgery, the results remainedsimilar except for the relationship be-tween HbA1c and major complications,which became less significant but trendedtoward statistical significance (Supplemen-tary Tables 4 and 5).

Six-Month Mortality Risk PredictionUsing CART AnalysisOn CART analysis, using the variables age,eGFR, presence of diabetes, and Charlsoncomorbidity index (excluding diabetes andage), we found a training ROC of 0.79 andtesting ROC of 0.76. Charlson comorbidityindex(excludingdiabetesandage)wasthevariable of highest relative contributionto the model (100%), followed by age(33.62%), eGFR (22.69%), presence ofdiabetes (10.96%), and length of surgery(7.43%). The classification tree is de-scribed in Fig. 3. Patients with a Charlsoncomorbidity index(excludingdiabetesandage) .2.5 fared the worst, with a 21.6%

risk for 6-month mortality. For patientswith a Charlson comorbidity index (exclud-ing diabetes and age) #2.5, age #79.5years, Charlson comorbidity index (ex-cluding diabetes and age).0.5, length ofsurgery #159.5 min, and eGFR #89.16mL/min/1.73m2,diabetesconferred9.0%of the risk of 6-month mortality versus2.7% for no diabetes.

No significant association between pre-diabetes, defined categorically, and adverseoutcomes was observed (SupplementaryTable 6). No significant differences in out-comes were observed between patientswith previously undiagnosed diabetesand thosewithpreviously knowndiabetes(Supplementary Table 7). Fewer signifi-cant associations were found betweenhigher HbA1c and adverse surgical out-comes within patients with diabetes(Supplementary Table 8). Subgroup anal-yses of associations of diabetes andHbA1cwith patient outcomes within each surgi-cal unit are displayed in SupplementaryFigs. 2–13.

CONCLUSIONS

Key FindingsWe conducted a prospective, observa-tional study in 7,565 patients aged $54years undergoing surgery in a tertiaryteaching hospital to investigate the in-dependent association of diabetes defined

Figure 2—Association of diabetes and HbA1c with postoperative outcomes. Adjusted for age (years), Charlson comorbidity index (excluding diabetes andage), eGFR by CKD-EPI equation (mL/min/1.73 m2), and length of operation (min), with surgical unit treated as a random effect. aIRR is applicable tocontinuous variables; aOR is applicable to categorical variables. 6m, 6 months; Cx, complication; mech., mechanical.

1176 Diabetes, HbA1c, and Surgical Outcomes Diabetes Care Volume 41, June 2018

categorically or by using HbA1c as a con-tinuous variable with outcomes after sur-gery. We observed a 30% prevalence ofdiabetes and found that diabetes andhigher HbA1c were independently asso-ciated with adverse postoperative out-comes, including 6-month mortality, majorcomplications, ICU admission,mechanicalventilation, and hospital LOS. We alsofound a higher median hospital cost forpatientswithdiabetes. CART analysis con-firmed a higher risk of 6-month mortal-ity with diabetes in conjunction withother risk factors. In contrast, we did notfind a significant association between

prediabetes and adverse postoperativeoutcomes.

Relationship to Previous StudiesTheprevalenceofdiabetes in this cohort ishigher than previously reported and maybe related toage (5).Wealsoused routineHbA1c measurements to diagnose diabe-tes, whereas previous studies only usedmedical records (7,12,14).

The association between diabetes andLOS has been previously demonstrated insurgical patients (7,11)andwasconfirmedin this study. The association betweendiabetes and mortality has been demon-strated in the cardiac (7,8) andnoncardiac

surgery setting (11,13,14), although somestudieshaverefutedthisrelationship(9,10).Of these studies, all identified diabetesfrom medical records without taking intoaccount HbA1c levels, one had a limitedsample size (9), and several did not adjustfor other confounding factors (13,14). Thecurrent studydemonstrates that 6-monthmortality is higher in patients with di-abetes undergoing surgery.

ICU admission and mechanical ventila-tion are markers of postoperative mor-bidity. One retrospective study found thatpatients with diabetes had a prolongedICU stay after cardiac surgery (34). Me-chanical ventilationhasnotbeenpreviouslystudied in this context. ICU admission andmechanical ventilation were significantlyincreased with diabetes in the currentstudy.

Previous studies have investigated arange of postoperative complications andtheir relationship with diabetes (7,12,14).The complications studied and theirdefinitions have been variable,making itdifficult to drawmeaningful conclusionsfrom these studies. In addition, severityof complications has rarely been ac-counted for. This study used theClavien-Dindoclassification,anestablishedmethodof grading surgical complications (29)and demonstrates a higher rate of majorcomplications among patients with di-abetes.

To our knowledge, this prospectivestudy is the largest to assess the impactof HbA1c across a wide range of surgicalspecialties.We did not find an associationwith higher HbA1c and mortality consis-tentwith results frommoststudies (14,16).We also found higher HbA1c to be asso-ciated with greater LOS, consistent withother studies (20).

The literature remains variable on theassociation of HbA1c with other postop-erative complications (14,17,18,20,21). Inthe current study, we demonstrated thathigherHbA1cwas associatedwith a higherrate of major complications.

Study ImplicationsOur study implies that, in patients age$54years,diagnosisofdiabetes identifies thoseat higher risk of morbidity and mortalityafter surgery and implies that poor glyce-mic control before surgery, indicated by anelevatedHbA1c, remains an important riskfactor for adverseoutcomesafter surgery.Logically, therefore, patients with diabe-tes and especially those with high HbA1c

Figure 3—CART analysis showing interactions among patient characteristics, presence of diabetes,and risk of mortality at 6months (6mmortality). eGFR by CKD-EPI equation values aremilliliters perminute per 1.73 m2; age values are years. *n = 7,564 compared with the study sample size of 7,565because in one individual, 6-month mortality status was unclear. CCI, Charlson comorbidity index(excluding diabetes and age).

care.diabetesjournals.org Yong and Associates 1177

should be triaged to pathways of carededicatedtohigher-riskpopulations.Risksof excessively tight blood glucose controlperioperatively should be considered, butcontroversy exists regarding glycemictargets in the intensive care setting, withconflicting results from various studies,includingonefromourinstitution(23,35,36).The current findings provide robust datafor future interventional trials to examinethe role of intensive pre- and postopera-tive glycemia management of patientswith diabetes. Finally, we found that pre-diabetes as defined by HbA1c levels is nota risk factor for adverse outcomes aftersurgery.

Strengths and LimitationsStrengths of the study include its pro-spective nature, large sample size, andavailability of HbA1c measurements foranalysis, which is in contrast to otherstudies that only included patients withavailable HbA1c readings retrospectively(14,18–20), introducing selection bias. Toour knowledge, this study is the first toevaluate the association of diabetes andhyperglycemia, using routineHbA1cmeas-urements, with surgical outcomes in alarge patient cohort with accurate diabe-tes diagnoses. Previous studies assessingthe relationship of HbA1c with outcomesafter surgery have used variable HbA1ccutoff values (14,16,18–20). The currentstudy evaluated HbA1c as a continuousvariable. Furthermore, we investigatedthe independent association of diabetesand HbA1c with various outcomes by ad-justingforpatient-relatedfactors, includingage, renal function, and Charlson comor-bidity index, and for surgical factors, in-cluding type and length of surgery. Thebroad range of surgeries included makesinforming preoperative guidelines morepractical without the complexity of ac-counting for specific units.The addition of CART analysis provided

furtherinsightintotherelationshipamongpatient characteristics,diabetes, andmor-tality at 6 months. It showed the circum-stanceswhere association between diabetesand 6-month mortality was maximal. As apredictivemodel, itcanbeusedasaclinicaldecision-making tool for identifying sur-gical patients at risk formortality at 6monthsaccordingtotheirCharlsoncomorbidityin-dex, age, eGFR, length of surgery, and di-abetes status.Limitations of this study include its

observational nature and, as such, the

possibility of residual confounding ef-fects of comorbidities unable to be fullyadjusted for, such as preoperative ane-mia (37), preoperative hypoalbuminemia(38), blood pressure (39), and nutritionalstatus (40). Moreover, the stronger as-sociations of diabetes compared withHbA1cwith adverseoutcomes (Fig. 2)maybe explained by more comorbidities inthis group, irrespective of preoperativeblood glucose control. Despite adjust-ing for comorbidities comprehensivelyby using the Charlson comorbidity in-dex, age, and renal function, the pos-sibility of residual confounding cannotbe excluded.

The degree to which the relationshipbetween higher HbA1c and adverse pa-tient outcomes was due to perioperativehyperglycemia is difficult to ascertain. Ofnote, 99% of inpatients had their HbA1cmeasured preoperatively or within 3 daysof admission, making their HbA1c valuesunlikely to be affected by inpatient bloodglucose levels. Higher HbA1c levels havebeen positively correlated with perioper-ative hyperglycemia (19), and perioper-ative hyperglycemia has been associatedwith poorer patient outcomes (21–23).Regardless, the finding that higher HbA1cisassociatedwithadversepatientoutcomespostoperatively raises the need to explorepreoperative optimization of glycemic con-trol as a means of diminishing risk.

Although HbA1c is reliable for detectingglycemic control duringperiodsof surgicalstress, its validity is affected by hemoglo-binopathies, blood loss, iron deficiencyanemia, shortened redbloodcell life span,race variation, and recent blood trans-fusion (6,25). Thus, we excluded patientswho received blood transfusion up to90 days before HbA1c measurement.

Comparedwith patients excluded fromthis study, patientswhomet the inclusioncriteria had a higher median eGFR, lowerCharlson comorbidity index, and longermedian length of surgery, and a smallerproportionhaddiabetes.Patientsexcludedfor receivingabloodtransfusion likelyweresicker. Furthermore, patients excluded forhaving minor procedures despite havingat least one overnight stay in the hospitalcould have had a longer hospital stay thanintended as a result of underlying illness,which we acknowledge could have in-troduced selection bias. This study alsolikely underestimated the incidence ofmortality because mortality was only es-tablished if patient death occurred in the

hospital or was reported to the hospitalwithin the study period.

ConclusionDiabetes was prevalent in 30% of inpa-tients age$54years undergoing surgeryin a tertiary teaching hospital. Presenceofdiabeteswas independentlyassociatedwith a higher risk of 6-month mortality,major complications, ICU admission, me-chanical ventilation, and increasedhospitalLOS. Higher HbA1c was independentlyassociated with a higher risk of majorcomplications, ICU admission, and in-creased hospital LOS. The relationshipbetweenpoor glycemic control andpoorsurgical outcomes suggests that this isan area for future intervention.

Funding. E.I.E. was supported by a NationalStroke Foundation Small Project grant, Sylviaand Charles Viertel Charitable Foundation ViertelClinical Investigatorship, Sir Edward WearyDunlopMedical ResearchFoundation grant, RoyalAustralasian College of Physicians fellowship, andDiabetes Australia Research Program grant.Duality of Interest. No potential conflicts ofinterest relevant to this article were reported.AuthorContributions.P.H.Y.was involved in theproject conception, literature review and synthe-sis, data acquisition, detailed data analysis, statis-tical analysis, critical discussion, and drafting ofthe manuscript. L.W. was involved in the criticaldiscussion, data collection, and detaileddata anal-ysis.N.T.was involved in the critical discussion anddetailed data analysis. L.C. was involved in thestatistical analysis and critical discussion. R.J.R.,R.M., Q.T.L., and A.N.M.were involved in the dataacquisition and analysis. R.B., J.M., D.S., andD.J. were involved in the critical discussion. J.D.B.contributed to the experimental design andassisted with the decision support programsin the electronic health records projects. G.K.H.was involved in the inception of decision sup-port programs in the electronic health recordsprojects. J.F.L. was involved in the detaileddata analysis. A.N.M. was involved in the dataacquisition. J.D.Z. was involved in the supervi-sion of the project. E.I.E. was involved in theproject conception, experimental design, dataacquisition, data analysis, detailed data anal-ysis, statistical analysis, critical discussion,and supervision of the project. All authorscontributed to the revision of themanuscript andreviewed thefinal versionof themanuscript. E.I.E.is the guarantor of this work and, as such, hadfull access to all the data in the study and takesresponsibility for the integrity of the data and theaccuracy of the data analysis.

References1. Ogurtsova K, da Rocha Fernandes JD, Huang Y,et al. IDF Diabetes Atlas: global estimates for theprevalence of diabetes for 2015 and 2040. Di-abetes Res Clin Pract 2017;128:40–502. Australian Bureau of Statistics (ABS). NationalHealth Survey: First Results, 2014-15 [Internet],

1178 Diabetes, HbA1c, and Surgical Outcomes Diabetes Care Volume 41, June 2018

2015. Available from http://www.abs.gov.au/ausstats/abs@.nsf/mf/4364.0.55.001. Accessed26 September 20173. Nanayakkara N, Nguyen H, Churilov L, et al.Inpatient HbA1c testing: a prospective observa-tional study. BMJ Open Diabetes Res Care 2015;3:e0001134. Gawande AA, Thomas EJ, Zinner MJ, BrennanTA. The incidence and nature of surgical adverseevents inColoradoandUtahin1992.Surgery1999;126:66–755. ValentineNA,Alhawassi TM,RobertsGW,VoraPP, Stranks SN, Doogue MP. Detecting undiag-nosed diabetes using glycated haemoglobin: anautomated screening test in hospitalised patients[published correction appears inMed J Aust 2014;200:87]. Med J Aust 2011;194:160–1646. American Diabetes Association. Diagnosis andclassification of diabetes mellitus. Diabetes Care2014;37(Suppl. 1):S81–S907. Carson JL, Scholz PM, Chen AY, Peterson ED,Gold J, Schneider SH. Diabetes mellitus increasesshort-term mortality and morbidity in patientsundergoing coronary artery bypass graft surgery.J Am Coll Cardiol 2002;40:418–4238. Thourani VH, Weintraub WS, Stein B, et al.Influence of diabetes mellitus on early and lateoutcome after coronary artery bypass grafting.Ann Thorac Surg 1999;67:1045–10529. Little SA, JarnaginWR, DeMatteo RP, BlumgartLH, Fong Y. Diabetes is associated with increasedperioperative mortality but equivalent long-termoutcome after hepatic resection for colorectalcancer. J Gastrointest Surg 2002;6:88–9410. Hughes K, Jackson JD, Prendergast TI, et al.Diabetes mellitus is not associated with majormorbidity following open abdominal aortic an-eurysm repair. J Surg Res 2013;184:751–75411. Yeh CC, Liao CC, Chang YC, et al. Adverseoutcomesafternoncardiacsurgeryinpatientswithdiabetes: a nationwide population-based retro-spective cohort study. Diabetes Care 2013;36:3216–322112. AtaA,ValerianBT,LeeEC,BestleSL,ElmendorfSL, Stain SC. The effect of diabetes mellitus onsurgical site infections after colorectal and non-colorectal general surgical operations. Am Surg2010;76:697–70213. KrolikowskaM, KatajaM, Poyhia R,DrzewoskiJ, Hynynen M. Mortality in diabetic patientsundergoing non-cardiac surgery: a 7-year follow-up study. Acta Anaesthesiol Scand 2009;53:749–75814. AcottAA,TheusSA,KimLT.Long-termglucosecontrol and risk of perioperative complications.Am J Surg 2009;198:596–59915. Wexler DJ, Nathan DM, Grant RW, Regan S,Van Leuvan AL, Cagliero E. Prevalence of elevated

hemoglobin A1c among patients admitted to thehospital without a diagnosis of diabetes. J ClinEndocrinol Metab 2008;93:4238–424416. RollinsKE,VaradhanKK,DhatariyaK,LoboDN.Systematic review of the impact of HbA1c onoutcomes following surgery in patients with dia-betes mellitus. Clin Nutr 2016;35:308–31617. Halkos ME, Puskas JD, Lattouf OM, et al.Elevated preoperative hemoglobin A1c level ispredictive of adverse events after coronary arterybypass surgery. J Thorac Cardiovasc Surg 2008;136:631–64018. Dronge AS, Perkal MF, Kancir S, Concato J,Aslan M, Rosenthal RA. Long-term glycemiccontrol and postoperative infectious complica-tions.ArchSurg2006;141:375–380;discussion38019. Goodenough CJ, Liang MK, Nguyen MT, et al.Preoperative glycosylated hemoglobin and post-operative glucose together predict major compli-cations after abdominal surgery. J Am Coll Surg2015;221:854–861.e120. Underwood P, Askari R, Hurwitz S, ChamarthiB, Garg R. Preoperative A1C and clinical outcomesin patients with diabetes undergoing major non-cardiac surgical procedures. Diabetes Care 2014;37:611–61621. King JT Jr., Goulet JL, PerkalMF, Rosenthal RA.Glycemic control and infections in patients withdiabetesundergoingnoncardiacsurgery.AnnSurg2011;253:158–16522. FrischA,ChandraP,SmileyD, et al. Prevalenceand clinical outcome of hyperglycemia in theperioperative period in noncardiac surgery. Di-abetes Care 2010;33:1783–178823. vandenBergheG,WoutersP,WeekersF,etal.Intensive insulin therapy in critically ill patients.N Engl J Med 2001;345:1359–136724. Greci LS, KailasamM,Malkani S,et al.UtilityofHbA(1c) levels for diabetes case finding in hospi-talizedpatientswithhyperglycemia.DiabetesCare2003;26:1064–106825. International Expert Committee. Interna-tional Expert Committee report on the role ofthe A1C assay in the diagnosis of diabetes. Dia-betes Care 2009;32:1327–133426. Simpson AJ, Krowka R, Kerrigan JL, et al.Opportunistic pathology-based screening for di-abetes. BMJ Open 2013;3:e00341127. Moghissi ES, Korytkowski MT, DiNardo M,et al.; American Association of Clinical Endocrinol-ogists; American Diabetes Association. AmericanAssociation of Clinical Endocrinologists and Amer-icanDiabetesAssociationconsensus statementoninpatientglycemiccontrol.DiabetesCare2009;32:1119–113128. von Elm E, Altman DG, Egger M, Pocock SJ,GøtzschePC,VandenbrouckeJP;STROBEInitiative.The Strengthening the Reporting of Observational

Studies in Epidemiology (STROBE) statement:guidelines for reporting observational studies.Int J Surg 2014;12:1495–149929. Dindo D, Demartines N, Clavien PA. Classifi-cation of surgical complications: a new proposalwith evaluation in a cohort of 6336 patientsand results of a survey. Ann Surg 2004;240:205–21330. Charlson ME, Pompei P, Ales KL, MacKenzieCR. A new method of classifying prognosticcomorbidity in longitudinal studies: developmentand validation. J Chronic Dis 1987;40:373–38331. Levey AS, Stevens LA, Schmid CH, et al.;CKD-EPI (Chronic Kidney Disease Epidemiol-ogy Collaboration). A new equation to estimateglomerular filtration rate. Ann Intern Med 2009;150:604–61232. Havens JM, Peetz AB, DoWS, et al. The excessmorbidity and mortality of emergency generalsurgery. J Trauma Acute Care Surg 2015;78:306–31133. JamesG,WittenD,HastieT, TibshiraniR.Tree-based methods. In An Introduction to StatisticalLearning with Applications in R. 1st ed. New York,Springer Science + Business Media, 2013, p. 30334. Morricone L, Ranucci M, Denti S, et al. Di-abetes and complications after cardiac surgery:comparison with a non-diabetic population. ActaDiabetol 1999;36:77–8435. Di Muzio F, Presello B, Glassford NJ, et al.Liberal versus conventional glucose targets incritically ill diabetic patients: an exploratory safetycohort assessment. Crit Care Med 2016;44:1683–169136. Buchleitner AM, Mart ınez-Alonso M,Hernandez M, Sola I, Mauricio D. Perioperativeglycaemic control for diabeticpatientsundergoingsurgery. Cochrane Database Syst Rev 2012;9:CD00731537. Carson JL, Duff A, Poses RM, et al. Effect ofanaemia and cardiovascular disease on surgicalmortality and morbidity. Lancet 1996;348:1055–106038. Gibbs J, Cull W, HendersonW, Daley J, Hur K,Khuri SF. Preoperative serum albumin level as apredictor of operative mortality and morbidity:results from the National VA Surgical Risk Study.Arch Surg 1999;134:36–4239. Patel A, MacMahon S, Chalmers J, et al.;ADVANCE Collaborative Group. Effects of a fixedcombination of perindopril and indapamide onmacrovascular and microvascular outcomesin patients with type 2 diabetes mellitus (theADVANCE trial): a randomised controlled trial.Lancet 2007;370:829–84040. Dindo D, Muller MK, Weber M, Clavien PA.Obesity in general elective surgery. Lancet 2003;361:2032–2035

care.diabetesjournals.org Yong and Associates 1179