The Presence of Diabetes and Higher HbA_1c Are Independently Associated With Adverse Outcomes After Surgery

The prevalence of diabetes is rising, which in 2015 was estimated worldwide at 415 million (8.8%) (1). In Australia, the prevalence of diabetes was 1.2 million (5.1%) in 2014 and 2015 (2). One-third of inpatients age ≥54 years at a tertiary referral center were reported to have diabetes (3).

Surgical procedures place physical stress on patients and can lead to morbidity and mortality (4). Patients with diabetes and chronic hyperglycemia, measured by HbA_1c, may be at particular risk for perioperative morbidity from diabetes-related complications (5,6). Although diabetes has been associated with increased mortality and morbidity in the setting of cardiac surgery (7,8), this association in the noncardiac surgery setting is variable (9–14). Inconsistency remains in both the reporting and the definitions of surgical complications and perioperative morbidity among studies, making it difficult to draw conclusions on the association between diabetes and postoperative complications. Furthermore, previous studies have used medical records alone to identify diabetes in surgical patients (7,12,14), which may fail to identify the up to 18% who may have diabetes (15).

Studies investigating the association of HbA_1c with surgical outcomes have shown conflicting results (16), with some demonstrating associations with higher mortality (14,17), infection (17,18), myocardial infarction (17), renal failure (17), cerebrovascular accident (17), major complications (19), and hospital length of stay (LOS) (20) and others failing to observe such associations (14,21). Of note, many studies had limited HbA_1c results available for analysis, introducing selection bias (14,18–20), whereas others had limited sample sizes (19).

Most studies assessing the association of hyperglycemia and postoperative outcomes have focused on perioperative hyperglycemia by using blood glucose readings around the time of surgery (21,22). However, perioperative hyperglycemia is not an accurate indicator of diabetes status because it is affected by perioperative fasting and stress hyperglycemia from surgical trauma (23). HbA_1c has been proposed as a reliable indicator of glycemic status in the inpatient setting because it is unaffected by fasting status and less affected by stress hyperglycemia (24) and has been endorsed as an appropriate method of diagnosing diabetes (6,25).

Accordingly, in this prospective study, we used HbA_1c to determine both presence of diabetes and severity of chronic preadmission glycemic status. We tested the hypothesis that diabetes, defined as a categorical variable, and HbA_1c, defined as a continuous variable, carry an independent association with adverse outcomes after surgery.

We performed a prospective, observational study in surgical inpatients admitted to Austin Health, a tertiary teaching hospital affiliated with the University of Melbourne in Melbourne, Victoria, Australia. During the period of 6 May 2013 to 23 January 2016, as part of a process, labeled the Diabetes Discovery Initiative, all patients age ≥54 years without an HbA_1c reading within 3 months of admission received an automatic HbA_1c measurement on hospital admission (3). This age cutoff was chosen on the basis of a previous study that used HbA_1c measurements to demonstrate a higher prevalence of undiagnosed diabetes in inpatients aged >54 years (26). HbA_1c was measured by immunoassay on a COBAS INTEGRA 800 (Roche Diagnostics, Indianapolis, IN). This study was approved by the Austin Health Research Ethics Committee (LNR/15/Austin/41), which waived the need for informed consent for a planned practice change agreed to by hospital senior medical staff members as part of the Diabetes Discovery Initiative.

As part of the Diabetes Discovery Initiative, patients with HbA_1c ≥8.3% (67 mmol/mol) were seen by an endocrinology advanced trainee who generated a personalized plan for glycemic control. Patients undergoing high-risk surgery, including cardiac, orthopedic, and general surgery, with HbA_1c between 7.5% (58 mmol/mol) and 8.2% (66 mmol/mol) and patients with newly diagnosed diabetes were seen by the internal medicine advanced trainee. Patients were managed according to the hospital’s perioperative guidelines for patients with diabetes, which are based on Australian Diabetes Society guidelines (27).

Inclusion criteria were age ≥54 years, surgery with at least one overnight hospital stay, HbA_1c measurement performed <3 months before or 7 days after surgery, and an available serum creatinine level. The cutoff date for HbA_1c measurement was set at 7 days after surgery to capture patients with delays in HbA_1c measurement while minimizing interference in HbA_1c levels from surgery. We excluded patients undergoing minor interventional or noninterventional procedures, such as gastroscopy, colonoscopy, bronchoscopy, lumbar puncture, electroconvulsive therapy, direct cardioversion, paravertebral block, video fluoroscopy swallowing or urodynamic studies, and MRI. Patients also were excluded if they received a blood transfusion ≤90 days before HbA_1c measurement because blood transfusions affect the reliability of HbA_1c results (6). Where patients had more than one surgical procedure in their hospital episode, analyses were performed on the basis of the first procedure.

Detailed demographic and clinical data were entered into a Cerner electronic database (Cerner Corporation, Kansas City, MO). Baseline data collected were age, sex, serum creatinine level, HbA_1c result, diabetes status, separation unit, type of surgery, length of surgery, elective or emergency status of surgery, and comorbid conditions (Supplementary Table 1).

Patients were followed for 6 months. The primary study outcome was mortality within 6 months. Secondary outcomes were presence of a major complication, admission to an intensive care unit (ICU), requirement for mechanical ventilation, hospital LOS, readmission within 6 months, and cost of hospital episode. We included the STROBE (Strengthening the Reporting of Observational Studies in Epidemiology) statement checklist for observational studies to report findings (28).

Patients were diagnosed with diabetes if they had an HbA_1c ≥6.5% (48 mmol/mol) or preexisting diagnosis of diabetes on medical records regardless of HbA_1c level. Patients were diagnosed with prediabetes if they had an HbA_1c ≥5.7% (39 mmol/mol) and <6.5% (48 mmol/mol). Patients with an HbA_1c <5.7% (39 mmol/mol) were considered to have normoglycemia. Patients with prediabetes and normoglycemia were classified as not having diabetes. These definitions are in accordance with the International Expert Committee and American Diabetes Association (6,25).

LOS was determined by the period from admission to discharge, including days in the hospital-in-the-home unit. Readmission was defined as unplanned readmission to this hospital within the 6-month follow-up period. Mechanical ventilation referred to the requirement of such postoperatively. Mortality was established if patient death had occurred during admission or had been reported to the hospital during the 6-month study period.

Hospital-acquired complications were graded on the basis of the Clavien-Dindo classification guide for surgical complications (29). The Clavien-Dindo classification is a validated approach to surgical outcome assessment that assigns severity grades to surgical complications. Surgical complications were assessed from diagnoses of in-hospital complications from medical records, with major complication defined as a hospital-acquired complication of Clavien-Dindo grade ≥4 and included life-threatening complications and in-hospital death (Supplementary Table 2). Our institution comprises specialist wards equipped to manage severe conditions that might otherwise be managed in the ICU. Thus, we also included life-threatening complications involving documented organ failure unable to be managed by medications or surgical interventions alone, irrespective of ICU admission. In case of disagreement on grading by two assessors, the case was discussed with reference to the classification guide (29).

We a priori identified confounders as comorbid conditions, age, renal function, and length of surgery. Comorbid conditions reported as ICD-10, Australian Modification, codes were used to calculate each patient’s Charlson comorbidity index score. This validated method weights the impact of chronic disease by assigning scores to chronic conditions on the basis of severity and effect on mortality (30) (Supplementary Table 1). Diabetes and age were excluded because they were assessed as separate characteristics. Estimated glomerular filtration rate (eGFR) was calculated by the Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) equation by using extracted data on age, sex, and first creatinine value in the hospital episode (31).

Length of surgery in minutes was used as a marker for severity of surgery. Another potential marker of severity is emergency or nonemergency status of surgery, with emergency surgery contributing to poorer surgical outcomes (32). In addition, patients were categorized into 15 surgical units according to the separation unit from which they were discharged (Table 1). Patients with a separation unit outside these 15 units were assigned by two independent assessors to a unit on the basis of their surgical procedure. These units were used as random effects in the statistical models. Subgroup analyses were performed for surgical units.

Statistical analysis was performed with Stata 14.2 software (StataCorp, College Station, TX). Baseline characteristics were reported as medians and interquartile ranges (IQRs) (continuous characteristics) or counts and percentages (categorical characteristics) and compared between patients included in and excluded from the study and by diabetes category using Kruskal-Wallis and χ²/Fisher exact tests, respectively. Multivariable analyses were conducted using random-effects negative binomial regression for LOS outcome (presented as count of days) and random-effects logistic regression for binary outcomes, with surgical units as random effects. Two analyses were performed: 1) with diabetes status classified as diabetes or no diabetes (including prediabetes and normoglycemia) and 2) with HbA_1c as a continuous marker. A priori chosen adjustment covariates were age, eGFR, Charlson comorbidity index (excluding diabetes and age), and length of surgery in minutes. Standard analyses of collinearity and model fit were performed. Because emergency surgeries had statistically significantly shorter length than nonemergency surgeries, to avoid excessive collinearity, emergency status of surgery was not included as a covariate in addition to length of surgery but instead was used in robustness analysis. Exploratory subgroup analysis by surgical unit was performed and presented as forest plots. The cost of a patient episode was described using median and IQRs and analyzed by Wilcoxon rank sum test. A two-sided P < 0.05 was considered statistically significant.

We also used classification and regression tree (CART) advanced analysis (Salford Predictive Modeler 7 software; Salford Systems, San Diego, CA) to investigate complex interactions among patient characteristics, HbA_1c, length of surgery, and outcome of 6-month mortality and to design a predictive 6-month mortality risk model. CART is a binary partitioning statistical method that starts with the total sample and, in a stepwise manner, splits the sample into subsamples that are homogenous with respect to a defined outcome (33). The input variable that achieves the most effective split is dichotomized by automated analysis at an optimal threshold, maximizing the homogeneity within and separation between resulting subgroups. To establish predictive validity of the CART model, its performance is assessed by area under the receiver operating characteristic (ROC) curve with a 10-fold internal cross-validation, where data are randomly divided into 10 groups with 9 used to build the model (training) and 1 used to validate (testing). From this, CART generates a classification tree and numerical rank for each input used to build the tree by relative importance. Our CART analysis variables were age, eGFR, length of surgery, presence of diabetes, and Charlson comorbidity index (excluding diabetes and age) as inputs.

During the study period, 12,128 patients age ≥54 years underwent surgery at Austin Health. The following patients were excluded: those with no HbA_1c result within the specified time frame (n = 1,648), minor interventional or noninterventional procedure (n = 1,947), blood transfusion ≤90 days before HbA_1c measurement (n = 286), and no available creatinine value (n = 682) (Fig. 1). The final study population included 7,565 patients of whom 2,283 (30%) had diabetes, 2,825 (37%) had prediabetes, and 2,457 (32%) had normoglycemia. Of patients with diabetes, 236 (10%) were previously undiagnosed. Patients included in the study had a higher median eGFR, lower Charlson comorbidity index, longer median length of surgery, and smaller proportion with diabetes than those excluded (Supplementary Table 3).

The clinical characteristics of the study population are described in Table 1. Median (IQR) age of patients with normoglycemia, prediabetes, and diabetes was 68 (61, 77), 71 (63, 80), and 71 (64, 78) years, respectively (P < 0.001). The majority of patients in all groups were male. Median (IQR) HbA_1c in patients with normoglycemia, 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]), and 6.9% (6.3, 7.8) (52 mmol/mol [45, 62]), respectively (P < 0.001). Of the surgical procedures studied, 3,404 (45%) were elective and 4,161 (55%) were emergency. The proportion of patients with diabetes who received diabetes medication is presented in Supplementary Fig. 1. The association of diabetes and HbA_1c with surgical outcomes is summarized in Fig. 2.

The incidence of 6-month mortality was 6% (95% CI 5.5–6.8%) in patients without diabetes and 9% (95% CI 7.4–9.7%) in patients with diabetes. On multivariable analysis, presence of diabetes was associated with increased mortality 6 months after surgery (adjusted odds ratio [aOR] 1.29 [95% CI 1.05–1.58]; P = 0.014) (Fig. 2). No statistically significant association between HbA_1c on a continuous scale and 6-month mortality was identified.

A major complication, defined as Clavien-Dindo grade ≥4, was present in 14% (95% CI 13–15%) of patients without diabetes and 21% (95% CI 20–23) of patients with diabetes. On multivariable analysis, presence of diabetes as a categorical variable was associated with greater risk of major complications (aOR 1.32 [95% CI 1.14–1.52]; P < 0.001) (Fig. 2). When assessed as a continuous variable, each 1% increase in HbA_1c was associated with greater risk of major complications (aOR 1.07 [95% CI 1.01–1.14]; P = 0.030) (Fig. 2).

Eighteen percent (95% CI 17–19) of patients without diabetes and 27% (95% CI 26–29) with diabetes were admitted to the ICU. On multivariable analysis, presence of diabetes as a categorical variable was associated with an increased likelihood of ICU admission (aOR 1.50 [95% CI 1.28–1.75]; P < 0.001) (Fig. 2). When assessed as a continuous variable, each 1% increase in HbA_1c was associated with an increased likelihood of ICU admission (aOR 1.14 [95% CI 1.07–1.21]; P < 0.001) (Fig. 2).

Mechanical ventilation was applied to 10% (95% CI 9–11) of patients without diabetes and 16% (95% CI 15–18) with diabetes. On multivariable analysis, presence of diabetes as a categorical variable was associated with an increased likelihood of receiving mechanical ventilation (aOR 1.67 [95% CI 1.32–2.10]; P < 0.001) (Fig. 2). No statistically significant association between HbA_1c on a continuous scale and mechanical ventilation was identified.

The median (IQR) hospital LOS was 6 (3, 11) days in patients without diabetes and 7 (4, 14) days in those with diabetes. On multivariable analysis, presence of diabetes as a categorical variable was associated with an increased hospital LOS (adjusted incidence rate ratio [aIRR] 1.08 [95% CI 1.04–1.12]; P < 0.001) (Fig. 2). When assessed as a continuous variable, each 1% increase in HbA_1c was associated with an increased hospital LOS (aIRR 1.05 [95% CI 1.03–1.06]; P < 0.001) (Fig. 2).

The incidence of 6-month readmission was 16% (95% CI 15–17) in patients without diabetes and 17% (95% CI 15–19) in those with diabetes. On multivariable analysis, no statistically significant association between diabetes as a categorical variable or HbA_1c as a continuous variable with 6-month readmission was identified (Fig. 2).

The median episode cost of the study population was $18,189, with patients without diabetes having a median (IQR) cost of $17,439 (11,438, 27,564) compared with $20,440 (12,186, 33,261) for patients with diabetes (P < 0.001).

Because of excessive collinearity with length of surgery, emergency status of surgery was not included as a covariate in the original model. When adjusting for emergency status instead of or in addition to length of surgery, the results remained similar except for the relationship between HbA_1c and major complications, which became less significant but trended toward statistical significance (Supplementary Tables 4 and 5).

On CART analysis, using the variables age, eGFR, presence of diabetes, and Charlson comorbidity index (excluding diabetes and age), we found a training ROC of 0.79 and testing ROC of 0.76. Charlson comorbidity index (excluding diabetes and age) was the variable of highest relative contribution to the model (100%), followed by age (33.62%), eGFR (22.69%), presence of diabetes (10.96%), and length of surgery (7.43%). The classification tree is described in Fig. 3. Patients with a Charlson comorbidity index (excluding diabetes and age) >2.5 fared the worst, with a 21.6% risk for 6-month mortality. For patients with a Charlson comorbidity index (excluding diabetes and age) ≤2.5, age ≤79.5 years, Charlson comorbidity index (excluding diabetes and age) >0.5, length of surgery ≤159.5 min, and eGFR ≤89.16 mL/min/1.73 m², diabetes conferred 9.0% of the risk of 6-month mortality versus 2.7% for no diabetes.

No significant association between prediabetes, defined categorically, and adverse outcomes was observed (Supplementary Table 6). No significant differences in outcomes were observed between patients with previously undiagnosed diabetes and those with previously known diabetes (Supplementary Table 7). Fewer significant associations were found between higher HbA_1c and adverse surgical outcomes within patients with diabetes (Supplementary Table 8). Subgroup analyses of associations of diabetes and HbA_1c with patient outcomes within each surgical unit are displayed in Supplementary Figs. 2–13.

We conducted a prospective, observational study in 7,565 patients aged ≥54 years undergoing surgery in a tertiary teaching hospital to investigate the independent association of diabetes defined categorically or by using HbA_1c as a continuous variable with outcomes after surgery. We observed a 30% prevalence of diabetes and found that diabetes and higher HbA_1c were independently associated with adverse postoperative outcomes, including 6-month mortality, major complications, ICU admission, mechanical ventilation, and hospital LOS. We also found a higher median hospital cost for patients with diabetes. CART analysis confirmed a higher risk of 6-month mortality with diabetes in conjunction with other risk factors. In contrast, we did not find a significant association between prediabetes and adverse postoperative outcomes.

The prevalence of diabetes in this cohort is higher than previously reported and may be related to age (5). We also used routine HbA_1c measurements to diagnose diabetes, whereas previous studies only used medical records (7,12,14).

The association between diabetes and LOS has been previously demonstrated in surgical patients (7,11) and was confirmed in this study. The association between diabetes and mortality has been demonstrated in the cardiac (7,8) and noncardiac surgery setting (11,13,14), although some studies have refuted this relationship (9,10). Of these studies, all identified diabetes from medical records without taking into account HbA_1c levels, one had a limited sample size (9), and several did not adjust for other confounding factors (13,14). The current study demonstrates that 6-month mortality is higher in patients with diabetes undergoing surgery.

ICU admission and mechanical ventilation are markers of postoperative morbidity. One retrospective study found that patients with diabetes had a prolonged ICU stay after cardiac surgery (34). Mechanical ventilation has not been previously studied in this context. ICU admission and mechanical ventilation were significantly increased with diabetes in the current study.

Previous studies have investigated a range of postoperative complications and their relationship with diabetes (7,12,14). The complications studied and their definitions have been variable, making it difficult to draw meaningful conclusions from these studies. In addition, severity of complications has rarely been accounted for. This study used the Clavien-Dindo classification, an established method of grading surgical complications (29) and demonstrates a higher rate of major complications among patients with diabetes.

To our knowledge, this prospective study is the largest to assess the impact of HbA_1c across a wide range of surgical specialties. We did not find an association with higher HbA_1c and mortality consistent with results from most studies (14,16). We also found higher HbA_1c to be associated with greater LOS, consistent with other studies (20).

The literature remains variable on the association of HbA_1c with other postoperative complications (14,17,18,20,21). In the current study, we demonstrated that higher HbA_1c was associated with a higher rate of major complications.

Our study implies that, in patients age ≥54 years, diagnosis of diabetes identifies those at higher risk of morbidity and mortality after surgery and implies that poor glycemic control before surgery, indicated by an elevated HbA_1c, remains an important risk factor for adverse outcomes after surgery. Logically, therefore, patients with diabetes and especially those with high HbA_1c should be triaged to pathways of care dedicated to higher-risk populations. Risks of excessively tight blood glucose control perioperatively should be considered, but controversy exists regarding glycemic targets in the intensive care setting, with conflicting results from various studies, including one from our institution (23,35,36). The current findings provide robust data for future interventional trials to examine the role of intensive pre- and postoperative glycemia management of patients with diabetes. Finally, we found that prediabetes as defined by HbA_1c levels is not a risk factor for adverse outcomes after surgery.

Strengths of the study include its prospective nature, large sample size, and availability of HbA_1c measurements for analysis, which is in contrast to other studies that only included patients with available HbA_1c readings retrospectively (14,18–20), introducing selection bias. To our knowledge, this study is the first to evaluate the association of diabetes and hyperglycemia, using routine HbA_1c measurements, with surgical outcomes in a large patient cohort with accurate diabetes diagnoses. Previous studies assessing the relationship of HbA_1c with outcomes after surgery have used variable HbA_1c cutoff values (14,16,18–20). The current study evaluated HbA_1c as a continuous variable. Furthermore, we investigated the independent association of diabetes and HbA_1c with various outcomes by adjusting for patient-related factors, including age, renal function, and Charlson comorbidity index, and for surgical factors, including type and length of surgery. The broad range of surgeries included makes informing preoperative guidelines more practical without the complexity of accounting for specific units.

The addition of CART analysis provided further insight into the relationship among patient characteristics, diabetes, and mortality at 6 months. It showed the circumstances where association between diabetes and 6-month mortality was maximal. As a predictive model, it can be used as a clinical decision-making tool for identifying surgical patients at risk for mortality at 6 months according to their Charlson comorbidity index, age, eGFR, length of surgery, and diabetes status.

Limitations of this study include its observational nature and, as such, the possibility of residual confounding effects of comorbidities unable to be fully adjusted for, such as preoperative anemia (37), preoperative hypoalbuminemia (38), blood pressure (39), and nutritional status (40). Moreover, the stronger associations of diabetes compared with HbA_1c with adverse outcomes (Fig. 2) may be explained by more comorbidities in this group, irrespective of preoperative blood glucose control. Despite adjusting for comorbidities comprehensively by using the Charlson comorbidity index, age, and renal function, the possibility of residual confounding cannot be excluded.

The degree to which the relationship between higher HbA_1c and adverse patient outcomes was due to perioperative hyperglycemia is difficult to ascertain. Of note, 99% of inpatients had their HbA_1c measured preoperatively or within 3 days of admission, making their HbA_1c values unlikely to be affected by inpatient blood glucose levels. Higher HbA_1c levels have been positively correlated with perioperative hyperglycemia (19), and perioperative hyperglycemia has been associated with poorer patient outcomes (21–23). Regardless, the finding that higher HbA_1c is associated with adverse patient outcomes postoperatively raises the need to explore preoperative optimization of glycemic control as a means of diminishing risk.

Although HbA_1c is reliable for detecting glycemic control during periods of surgical stress, its validity is affected by hemoglobinopathies, blood loss, iron deficiency anemia, shortened red blood cell life span, race variation, and recent blood transfusion (6,25). Thus, we excluded patients who received blood transfusion up to 90 days before HbA_1c measurement.

Compared with patients excluded from this study, patients who met the inclusion criteria had a higher median eGFR, lower Charlson comorbidity index, and longer median length of surgery, and a smaller proportion had diabetes. Patients excluded for receiving a blood transfusion likely were sicker. Furthermore, patients excluded for having minor procedures despite having at least one overnight stay in the hospital could have had a longer hospital stay than intended as a result of underlying illness, which we acknowledge could have introduced selection bias. This study also likely underestimated the incidence of mortality because mortality was only established if patient death occurred in the hospital or was reported to the hospital within the study period.

Diabetes was prevalent in 30% of inpatients age ≥54 years undergoing surgery in a tertiary teaching hospital. Presence of diabetes was independently associated with a higher risk of 6-month mortality, major complications, ICU admission, mechanical ventilation, and increased hospital LOS. Higher HbA_1c was independently associated with a higher risk of major complications, ICU admission, and increased hospital LOS. The relationship between poor glycemic control and poor surgical outcomes suggests that this is an area for future intervention.

Funding. E.I.E. was supported by a National Stroke Foundation Small Project grant, Sylvia and Charles Viertel Charitable Foundation Viertel Clinical Investigatorship, Sir Edward Weary Dunlop Medical Research Foundation grant, Royal Australasian College of Physicians fellowship, and Diabetes Australia Research Program grant.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. P.H.Y. was involved in the project conception, literature review and synthesis, data acquisition, detailed data analysis, statistical analysis, critical discussion, and drafting of the manuscript. L.W. was involved in the critical discussion, data collection, and detailed data analysis. N.T. was involved in the critical discussion and detailed data analysis. L.C. was involved in the statistical analysis and critical discussion. R.J.R., R.M., Q.T.L., and A.N.M. were involved in the data acquisition and analysis. R.B., J.M., D.S., and D.J. were involved in the critical discussion. J.D.B. contributed to the experimental design and assisted with the decision support programs in the electronic health records projects. G.K.H. was involved in the inception of decision support programs in the electronic health records projects. J.F.L. was involved in the detailed data analysis. A.N.M. was involved in the data acquisition. J.D.Z. was involved in the supervision of the project. E.I.E. was involved in the project conception, experimental design, data acquisition, data analysis, detailed data analysis, statistical analysis, critical discussion, and supervision of the project. All authors contributed to the revision of the manuscript and reviewed the final version of the manuscript. E.I.E. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.