The Association of Perioperative Glycemic control with Post-operative wound Infection after Coronary Artery Bypass Grafting in Veterans with Diabetes Mellitus

Introduction: After coronary artery bypass (CABG), diabetes mellitus is associated with increased risk of infection. We model the inter-relationship of pre- and post-operative glycemic control and their composite inuence on post-operative wound infection. Methods: 2,899 Veterans that underwent isolated CABG were stratied (Group I: < 8%, Group II: 8 - 10%, Group III: > 10%) according to preoperative HbA1c levels. We obtained the median blood sugar level (BSL) on post-operative days 0 - 4 (POD) and compared trends in BSL readings according to HbA1c groups. We t a multi-variable random effects mixed model to understand the odds (OR) of developing postoperative mediastinitis. A two-stage joint model was t to evaluate the adjusted hazard (HR) of pre-operative HbA1c, post-operative BSL and clinical factors on 90-day readmission for infection. Results: In groups I, II, and III had 66%, 25% and 9% patients respectively. In 148,810 post-operative BSL readings, median BSL values peaked at POD 2 (145 mg/dl). In group III, 29% BSL reading was above the recommended limit (< 180 mg/dl). Group III (OR 3.5; p < 0.01) and COPD (OR 2.51; p < 0.01) were associated with higher rates of mediastinitis. Groups II (HR 1.4; p = 0.01) and III (HR 1.5; p = 0.04) were associated with increased risk for 90-day readmission for infection. Post-operative blood sugar levels (p = 0.5) were not associated with increased risk for wound infection at 90 days. Conclusion: Among Veterans undergoing coronary artery bypass, a large proportion had HbA1c levels > 8%. In the post-operative period, 22% of the blood sugar readings are higher than 180 mg/dl. Preoperative HbA1c (rather than postoperative glycemic control) is associated with higher rates of deep sternal wound infection. evaluate the association of clinical co-variates with the median BSL obtained from each patients on postoperative days 0 - 4. (b) stage II - The results of the stage I model are included in a multi-variable Cox proportional hazards model fit to evaluate the association of clinical covariates and measured post-operative BSL readings with 30-day readmission for infections after isolated coronary artery bypass grafting among Veterans.


Introduction
Coronary artery bypass grafting (CABG), rather than percutaneous intervention, is the procedure of choice among diabetic patients with multi-vessel disease [1]. Mediastinitis and post-operative wound infection remains a concern among diabetes mellitus (DM) patients after cardiac surgery [2]. The Society of Thoracic Surgeons guidelines recommend that postoperative blood sugar should be maintained < 180 mg% for 72 hours [3].
In 2009, the Joint Commission introduced the Surgical Care Improvement Project to promote uniform evidence-based patient care. An important metric was to ensure a target blood sugar value < 200 mg% on the rst postoperative day at 6 AM [4]. While many researchers have evaluated the independent association of either preoperative or postoperative glycemic control and wound infection, evidence investigating the composite association of pre-and postoperative glycemic control is limited. Thus, we analyzed the Veterans Affairs Database motivated by the following questions (1) What proportion of diabetic patients achieve the target of < 180 mg% after surgery? How does the preoperative glycated hemoglobin (HbA1C) levels, a marker of glycemic control over a period of 90 days, in uence our ability to achieve optimal postoperative glycemic control (de ned as average blood sugar < 180 mg/dl after surgery)? (2) What is the composite impact of preoperative HbA1c and post-operative blood sugar control (de ned as BSL < 180 mg/dl) on infection during the early post-operative period? Methods This is an observational cohort study based on the Veterans Affairs National database. The study was approved by the Institutional Review Board, North-East Ohio VA Medical Center, IRB # 16004-H03. The review committee waived the need for individual patient consent.

Data source
The Veteran Health Affairs Department is the largest integrated health care system in the United States, providing care at 170 hospitals and 1025 out-patient facilities [5]. The VA Informatics and Computing Infrastructure (VINCI) contains treatment information regarding Veterans receiving care at their centers [6].

Study cohort
From the VASQIP database, we initially identi ed Veterans who underwent isolated CABG at 41 different VA medical centers nationwide (January 2007 -December 2014). Included Veterans underwent CABG with or without cardiopulmonary bypass. We further limited our study to only include patients (1) with a diagnosis of diabetes mellitus at admission and (2) where data regarding perioperative glycemic control was documented. A owchart of patient selection is provided in the supplement (S- Figure 1).
Categorization of CABG patients into groups strati ed by preoperative HBA1C levels HbA1C within 90 days prior to surgery was used as a measure of preoperative glycemic control. When multiple laboratory values were available, we selected the most recent value prior to surgical date. The American Diabetes Association recommends a more liberal HbA1c target < 8% among elderly or those with complex atherosclerotic disease [7]. Thus, study Veterans were categorized into three groups: group I -< 8%, group II -8 -10%, and group III -> 10%.
Postoperative protocol used for blood sugar control: The VA Modi ed Portland Insulin Protocol was used for managing postoperative blood sugar level (BSL) after cardiac surgery (Supplement) [8]. Brie y, patients are initiated on intravenous insulin therapy when BSL values > 175 mg%. Initiation of intravenous Insulin can occur in the intraoperative period. In the intensive care unit, BSL measurements are done hourly while on intravenous infusion and intravenous Insulin dose is titrated to maintain BSL values 125 -175 mg%. After adequate oral intake is established, intermittent injectable Insulin is administered according to a sliding scale and BSL measurements are done before every meal and with the morning lab tests. We collected BSL measurements for 5 postoperative days for all patients (postoperative day 0 -day 4). To account for temporal uctuation of readings during a 24-hour period and variation in sample collection timing, we included the daily median BSL value for each patient in our statistical models.
Clinical data: Demographic information (age, sex, race), clinical comorbidities (hypertension, dyslipidemia, CKD) and past medical history (prior myocardial infarction, prior percutaneous intervention, prior heart surgery) were obtained from VINCI. The International Classi cation of Diseases 9 th or 10 th version or the Common Procedures Terminology codes were used to de ne clinical conditions and procedures. Operative details like number of grafts performed and use of arterial conduit was also available. Left ventricular dysfunction was de ned as a left ventricular systolic function < 40% on preoperative echocardiography.
We calculated the estimated glomerular ltration rate (eGFR) and de ned chronic kidney disease (CKD) as eGFR < 60. Patients with a body mass index > 30 kg/m2 were de ned as obese.

De nition of Endpoints
We evaluated patients for two end-points: (1) Mediastinitis / deep sternal wound infection in the postoperative period (2) Readmission within 90 days due to deep sternal wound infection, harvest site infection or sepsis. Patients that died within 90 days of the index procedure were excluded in the analysis of this end-point. Hence, we were certain that all included patients were at risk for occurrence of event.

Statistical Methods
Continuous data was reported as median (interquartile range) while categorical information was presented as counts (percentages). To compare categorical variables between groups I, II and III , the chi 2 test was used. Continuous data was compared using the Kruskal-Wallis test. We used the Nelson-Aalen estimator to obtain non-parametric 90-day estimates for readmission in groups I, II and III; these were compared with the log-rank test. We also estimated and compared 90-day readmission rates according to adequacy of postoperative glycemic control. We t statistical models to present adjusted results for both our study aims (1) To evaluate the association of preoperative glycemic control on DSWI, we chose to t a hierarchical generalized logistic model. As patients were operated at 41 different institutions, an anonymous center code was included as a random effect, while patient-level clinical covariates were tted as xed effects. (2) To understand the independent association of pre-and post-operative glycemic control with 90-day infection rates, we t a two-stage joint model. The joint model framework consists of two components: a longitudinal component tted with a linear mixed model and a time-to-event component which is t using standard survival methods. Our linear mixed model allows us to report the association between pre-operative HbA1c values and postoperative glycemic control adjusted for demographic, clinical and operative covariates. The time-to-event component provides a measure of association between pre-and post-operative BSL values and probability of 90-day readmission. This time-to-event analysis was t as a Cox proportional hazards model. Along with our primary co-variate (HbA1c group), the other variables included in our model were (1) Patient demographics: age, sex, obesity, insulin treated DM, COPD, PAD, prior myocardial infarction, prior percutaneous intervention, chronic kidney disease and (2) Operative details: cardiopulmonary bypass time. Missing values were observed in the following variables: serum creatinine (2.7%) and cardiopulmonary bypass time (0.03%). All categorical variables included in the models were complete. Given the small proportion of missing data, missing elds were imputed with the variable mean prior to tting models. Statistical analysis was performed in R 3.5.3 (The R Foundation for Statistical Computing, Austria). The supplement contains further details regarding the statistical methods used in the study.

Results
Cohort characteristics strati ed by preoperative HbA1c levels ( Table 1): There were 1,915 (66%), 733 (25%) and 251 (9%) patients in groups I (HbA1c < 8%), II (HbA1c 8 -10%) and III (HbA1c > 10%) respectively ( Table 1). The median HbA1c values in groups I, II and III were 6.8 %, 8.7 % and 10.9 % respectively. Compared to group III, Veterans in group I were older (median age 65 vs 61 years). We observed a higher proportion of Blacks in Group III vs Group I (11.6% vs 8.9%; binomial test for proportions p-value = 0.04). In group III, a very high percentage of patients (78.5% vs 43.3% in Group I) were treated with Insulin prior to CABG. Group II patients had a higher rate of left ventricular systolic dysfunction (19% vs 13% in Group I) prior to surgery. Prior myocardial infarction rates were also higher in patients in Group III vs Group I (52% vs 41%). We did not observe any patient receiving bilateral internal thoracic artery grafts. The majority of patients underwent elective surgery.  Table 2. We fit a generalized linear mixed model to evaluate the association of pre-operative clinical co-variates with post-operative mediastinitis observed in the same admission. Given that patients were operated at 40 different Veteran Affairs medical centers, we included the hospital code as the level 1 random effect modifier, while clinical co-variates were fit as fixed effects. We observed that patients with preoperative HbA1c levels > 10% and those with COPD had higher odds of developing post-operative mediastinitis.
Abbreviations: OR -Odds ratio, LCI -lower confidence interval, UCI -upper confidence interval Association of perioperative glycemic control with 90-day Readmission ( Table 3): The 90-day readmission rate was 8.7 (7.6 -9.7) % in the whole group. Readmission rates increased as HbA1c levels increased. We observed readmission rates of 7.5%, 10.7% and 11.7% in Group I, II and III respectively (log-rank test p-value = 0.006) ( Figure 3A). Readmission rates were comparable in patients with BSL values < 180 mg/dl and > 180 mg/dl (log rank test p-value = 0.38) ( Figure 3B).  We performed an observational study to understand trends in post-operative blood sugar levels among Veterans with diabetes mellitus undergoing CABG. Using a joint modelling approach, we analyzed the adjusted effects of preoperative HbA1c levels and postoperative BSL readings on mediastinitis and early infection after CABG.
We observed that: (1) A substantial number of patients undergoing CABG at VA medical centers have elevated (8 -10%) and uncontrolled (> 10 %) HbA1c prior to surgery. (2) Ideal blood sugar control (BSL < 180 mg/dl) is di cult among patients that have poor glycemic control prior to surgery (3) The risk of infection after surgery is increased in patients with elevated HbA1c values. However, adjusting for this fact, high postoperative BSL does not increase this risk. Our graphical abstract provides an overview of our study (Figure 4).
Our results in perspective of the current Guidelines: Initial studies report higher risk of adverse events among patients with postoperative blood sugar > 200 mg/dl. Over time, research has demonstrated that 180 mg/dl may be the optimal BSL cut-off value after cardiac surgery. The recent NICE-SUGAR trial has outlined the detrimental effect of very restrictive BSL targets < 110 mg/dl in critically ill patients [9]. Consensus statements from the American Diabetes Association and American Academy of Clinical Endocrinologists both recommend a BSL target range of 140 -180 mg/dl in critically ill patients [3]. The Society of Thoracic Surgeons also recommend target BSL < 180 mg/dl after cardiac surgery, with tighter control (< 150 mg/dl) in sicker patients [10]. However, we demonstrate that maintaining target BSL levels is very di cult in patients with uncontrolled HbA1c levels before surgery; 1/3 rd of all readings in patients with HbA1c > 10% were above the ideal range. Subramaniam et al [11] report increased glucose level variability among patients with elevated HbA1c (> 6.5 %) in the 24-hour postoperative window. BSL was below 180 mg/dl) in 337 patients after cardiac surgery. They report poor glycemic control in patients with high HbA1c levels; both factors were associated with increased wound infection. Our results support and expand on earlier evidence. We demonstrate that ideal glycemic control is very di cult to attain in patients with elevated HbA1c. As reported by Lehwaldt et al [12], BSL values appear to peak on the second postoperative day; however, we add to current literature by also demonstrating that the decline after the peak depends upon preoperative HbA1c. We have demonstrated that in patients with HbA1c > 10%, postoperative BSL levels actually peak on the 3 rd postoperative day after which it declines. Ample literature already reports the strong association between preoperative HbA1c levels and adverse postoperative outcome. We focus our attention on infectious complications, and demonstrate that both mediastinitis and 90 readmission rates (for infections) are higher in diabetic patients with HbA1c > 8%.
Clinical Implications of our study We believe that our study is among the few that attempt to evaluate the composite effect of both pre and postoperative control of diabetes mellitus on wound infection after CABG. Our main message in this analysis is to demonstrate the importance of preoperative rather than postoperative blood sugar control in patients with postoperative wound infection.

Strengths & Limitations:
We present speci c strengths and limitations related to our study. We have performed a retrospective analysis; hence, any inferences are association rather than causation. We attempt to adjust for many clinical covariates that were available in our national database. However, our database does not contain information like details regarding inotropic use or duration of Insulin infusion therapy. These and other unmeasured factors do in uence postoperative blood sugar levels. Patients in our study were operated over a 10-year period at 40 different VA Medical centers; hence, we agree that practice variation is an important factor. We hence adopted a mixed methods approach to obtain results averaged over institutions. We believe that our study is among the rst (and possibly largest) that investigates the composite in uence of pre and postoperative blood glucose control on infection after CABG.

Conclusion
Among Veterans undergoing coronary artery bypass, a large proportion had HbA1c levels > 8%.
In the postoperative period, 22% of the blood sugar readings are higher than 180 mg/dl.  Figure 1 We recorded 148,810 BSL readings from 2899 Veterans with diabetes mellitus that underwent coronary artery bypass grafting between 2007 -2014. We observed that the median value for BSL gradually increased to peak at post-operative day 2 and then declined. However, 22% readings were above the acceptable limit of 180 mg/dl. The dotted line corresponds to 180 mg/dl.