Association Between Preoperative Hemoglobin A_1c Levels, Postoperative Hyperglycemia, and Readmissions Following Gastrointestinal Surgery

Study Population

The study population includes a subset of patients who were experiencing inpatient digestive surgeries (principal Current Procedural Terminology codes 40000-49999) at a Veterans Healthcare Administration hospital between October 1, 2007, and September 31, 2014, and who were assessed by VASQIP. Initially, 350 593 unique hospital stays involving a VASQIP-assessed surgery were identified by VASQIP. Eligible surgeries were limited to digestive surgeries only by their principal Current Procedural Terminology code (40000-49999). Patients were excluded from the analysis if they had experienced a documented preoperative infection or infectious complication during the 96 hours immediately following their operation. Patients were also excluded from the analysis if they were missing a postoperative glucose level assessment or preoperative HbA_1c assessment. The final analytic sample included 23 094 unique hospital stays involving at least 1 VASQIP-assessed surgical procedure (Figure 1). The unit of analysis for this study was the index surgical procedure. If a patient had undergone more than 1 digestive surgery during a hospitalization, only the procedural characteristics of the first surgery were included in the analysis.

Data Sources

The covariates of interest included in this analysis were collected from the VASQIP, the VA Decision Support System (DSS), and the VA Corporate Data Warehouse Inpatient data domains. Preoperative, operative, and postoperative characteristics were obtained from VASQIP. The VA Surgical Quality Improvement Program collects these characteristics via a trained nurse abstractor, thus ensuring the reliability of the variables collected.¹⁸ Inpatient administrations of insulin in the postoperative period were identified within the DSS pharmacy data. Insulin use in the postoperative period was defined as any administration of insulin during the 48 hours following an operation.

Variables

The main exposure of interest was the closest preoperative HbA_1c in the 90 days before a hospital admission as assessed by DSS (DSS laboratory result number,  17) and the highest postoperative glucose level during the 48 hours following an operation (DSS laboratory result number of 10 or 57). Hemoglobin A_1c was collected as a continuous variable and categorized as normal (<5.7%), prediabetic (5.7%-6.4%), or diabetic (>6.4%). These categories correspond to a 90-day preoperative mean glucose level of less than 117, 117 to 137, and more than 137 mg/dL, respectively. The peak glucose level was also collected as a continuous variable and examined as a 4-category variable for ease of interpretation (<120, 120-179, 180-250, and >250) after carefully examining trends between glucose as a continuous variable and each of the outcomes examined. In addition to the glucose result, the frequency of glucose level checks was also calculated as the total number of all completed glucose results during the 48 hours following an operation. We limited the postoperative setting to 48 hours to minimize the effect of postoperative events on our outcomes.

The covariates of age, sex, body mass index (BMI) (calculated as weight in kilograms divided by height in meters squared), American Society of Anesthesiologists (ASA) classification, preoperative HbA_1c, preoperative glucose level, peak postoperative glucose level, emergency case status, and postoperative length of stay were selected a priori because they are significant predictors for and markers of readmission.¹⁹ All variables, with the exception of race/ethnicity and ASA classification, were complete. While ASA classification was only missing for 4 patients (0.02%), race/ethnicity was missing for 3016 patients (14.8%). Race/ethnicity was not a statistically significant predictor of any of the outcomes and thus an unlikely confounder. Notably, when race/ethnicity was tested we found no evidence of collinearity between preoperative diabetes, postoperative blood glucose, and postoperative insulin use. Furthermore, there was no change in the interpretation of results when the variables were removed from the models.

The main outcome of interest was postdischarge readmission or the development of an infection-related postoperative complication as identified by VASQIP. While postoperative infections (a wound infection, pneumonia, urinary tract infecgtion, or sepsis) are not mutually exclusive, they were counted only once in the overall postoperative complications variable. We excluded all patients with a known infection preoperatively and for 96 hours postoperatively to minimize the likelihood that hyperglycemia during the first 48 hours reflected an ongoing infection. Inpatient readmissions in the 30 days following a hospital discharge were identified using the Corporate Data Warehouse Inpatient data domain.

Statistical Analysis

Univariate and bivariate statistics were used to examine the population and compare patients in terms of diabetes status. Univariate and bivariate differences in postoperative outcomes were examined by HbA_1c categories, glucose categories, and insulin use. In unadjusted analyses, χ² tests or Kruskal-Wallis tests were used to test for differences across the groups. Smoothed plots were developed to examine changes in event rates across preoperative HbA_1c values. Logistic models adjusting for clinically relevant preoperative and operative characteristics were used to calculate the odds of each outcome across categories of preoperative HbA_1c and postoperative glucose. Preoperative HbA_1c was added to the adjusted peak 48-hour glucose levels model to understand the contribution of peak 48-hour glucose levels independently of preoperative HbA_1c. Peak 48-hour glucose levels were not included in the preoperative HbA_1c model. All analyses were completed using SAS, version 9.4 (SAS Institute). Smoothed plots were developed using the R package GGPLOT2.²⁰ Interaction terms were tested in the final model and included in the final model if statistically significant at P < .05.

Patient and Procedural Characteristics

Patient and procedural factors stratified by a preoperative diagnosis of diabetes by VASQIP are summarized in Table 1. The mean age for the overall cohort was 63.7 years. Most patients were men (n = 20 348, 94.5%). Patients with diabetes who were insulin-dependent had a higher BMI (32.0 vs 30.1; P < .001), included more patients with an ASA classification of 4 to 5 (19.8% vs 13.3%; P < .001), experienced more postoperative complications (29.8% vs 26.6%; P < .001), and had more 30-day readmissions (16.6% vs 14.2%; P < .001). Patients with diabetes who were insulin-dependent also had higher preoperative glucose levels (166.1 vs 133.4 mg/dL; P < .001), initial postoperative glucose levels (190.7 vs 160.0 mg/dL; P < .001), and peak 48-hour postoperative glucose levels (230.8 vs 182.0 mg/dL; P < .001). The mean HbA_1c for the cohort was 6.7%, with 7.0% among patients with diabetes who were noninsulin-dependent and 8.0% among patients with diabetes who were insulin-dependent (P < .001). Of 10 545 patients who did not receive a preoperative diagnosis of diabetes, the mean HbA_1c was 5.9%, with 3986 patients (37.8%) having 5.7% or less (normal), 5420 (51.4%) ranging from 5.7% to 6.4% (prediabetic), and 1139 (10.8%) more than 6.5%, diagnostic of diabetes.

Postoperative Complications by HbA_1c and Glucose

Elevated preoperative HbA_1c was associated with decreased postoperative complications even after adjusting for age, sex, ASA class, wound class, work relative value units, postoperative hospital length of stay, diabetes diagnosis, and insulin use (Table 2). As compared with patients with normal HbA_1c (<5.7%), patients with HbA_1c more than 6.5% were significantly less likely to develop any postoperative complications (odds ratio [OR], 0.89; 95% CI, 0.80-0.99; P = .03) or to be readmitted within 30 days (OR, 0.85; 95% CI, 0.74-0.96; P = .01). The frequencies of each complication stratified by preoperative HbA_1c are summarized in Table 3.

By contrast, elevated peak 48-hour postoperative blood glucose levels were associated with increased odds for developing postoperative complications, but these associations were only marginally significant after adjusting for other patient characteristics (Table 2). After adjustment, only 30-day readmissions remained significantly associated with peak 48-hour postoperative glucose levels (>250 vs <120 mg/dL; OR, 1.20; 95% CI, 1.01-1.42).

The interaction of preoperative HbA_1c and postoperative glucose was of specific interest, but we failed to find a statistically significant interaction between the 2 terms in the adjusted model, and it was therefore not included in the final model.

Postoperative Glucose Level Surveillance

The mean total glucose level was 4.92 for patients with an HbA_1c of less than 5.7%, 6.89 for patients with an HbA_1c ranging from 5.7% to 6.4%, and 9.71 for patients with an HbA_1c of more than 6.5% (P < .001). The frequency of hyperglycemia surveillance by means of glucose level checks increased as the preoperative HbA_1c increased from 3% to 10% (Figure 2A).

Postoperative Hyperglycemia Treatment

Patients with preoperative HbA_1c more than 6.5% have lower thresholds for insulin use per postoperative glucose until their glucose level reaches 240 mg/dL, in which insulin use equalizes across all preoperative HbA_1c categories (Figure 2B). The closest preoperative HbA_1c and glucose levels, as well as the highest 48-hour glucose levels, were the strongest predictors of receiving insulin in the 48 hours following surgery. Overall, patients with an HbA_1c ranging from 5.7% to 6.4% were not more likely to receive insulin compared with patients with an HbA_1c of less than 5.7%. However, patients with an HbA_1c of more than 6.5% were 1 to 3 times more likely to receive insulin when their postoperative glucose values were less than 250 mg/dL. Specifically, 530 patients (11.5%) with an HbA_1c of 5.7% or less received insulin compared with 1422 (17.5%) with an HbA_1c ranging from 5.7% to 6.4% and 3895 (44.2%) with an HbA_1c of more than 6.4 (P < .01). Once the postoperative glucose level reaches 250 mg/dL, all patients have an equal opportunity to receive insulin regardless of postoperative glucose levels. Other factors associated with insulin use are age, sex, BMI, ASA classification, preoperative diabetes, preoperative sepsis, and hypertensive medications.

Strengths and Limitations

Our study was limited to patients undergoing gastrointestinal surgery, among whom there are studies on the relationship of diabetes and hyperglycemia on postoperative outcomes. A strength of this study is the use of all glucose measurements and treatment in a national cohort of patients undergoing surgery to evaluate the association of perioperative glucose on complications and readmissions. However, it also has its limitations. We chose to categorize HbA_1c based on the American Diabetes Association cutoffs for normal (<5.7%), prediabetes (5.7 – 6.4%), and diabetes (>6.5%). This study does not delineate the outcomes for elevated HbA_1c of more than 7%, more than 8%, or even more than 10%, in which the power is lacking. In addition, because we excluded patients who died in the hospital to examine readmissions, patients with the most severe complications resulting in in-hospital death were excluded from this study. However, since we hypothesized that postoperative hyperglycemia was associated with worse outcomes, this limitation would result in bias toward the null. While all patients underwent a procedure in the general surgery specialty, they did not undergo the same procedure. Because we did not compare the specific type of surgery and patient perioperative glucose or HbA_1c, it is possible that the results were affected. In addition, there are several potential confounders, including laparoscopic vs open procedure, indications for surgery, and antibiotic use, which were not addressed and could limit the study. We also excluded patients without preoperative HbA_1c within 90 days of undergoing surgery or a postoperative glucose level assessment, which may have introduced selection bias (eTable 1 in the Supplement). Finally, because our data source is the VA, our patient population is predominantly male and as a retrospective study, we cannot assess causation.