Arthritis and diabetes mellitus affect approximately 46 million and 20 million people in the United States, respectively1-3, and 52% of patients with diabetes mellitus have some form of physician-diagnosed arthritis4. Both diseases have been linked to aging, and currently half of the patients with diabetes mellitus are over the age of sixty years1,2. As the prevalence of diabetes mellitus is expected to increase, the number of patients diagnosed with arthritis is also projected to rise.
It has been estimated that >8% of patients undergoing primary and revision total hip arthroplasty and total knee arthroplasty in the United States carry the diagnosis of diabetes mellitus5. In general, diabetic patients undergoing arthroplasty have demonstrated substantial increases in the rates of common surgical and systemic complications and mortality during the initial hospital stay 5. It remains unclear whether glycemic control has any impact on the outcomes for diabetic patients who undergo arthroplasty.
While there are strict criteria for differentiating between Type-I and Type-II diabetes mellitus, to our knowledge, there are no established criteria that define controlled or uncontrolled disease. Presently, glycemic control is designated by a physician's assessment that is based on the American Diabetes Association (ADA) guidelines utilizing a combination of a patient's self-monitoring of blood glucose levels, the current hemoglobin A1c level, and the presence or severity of diabetes-related comorbidities6. Self-monitoring of blood glucose levels provides a periodic measurement of daily glucose concentrations in the blood, whereas hemoglobin A1c (glycosylated or glycated hemoglobin) is a serologic marker that provides an average glucose concentration in the bloodstream for the previous one to three months. The 2007 ADA position statement recommends that adult patients with diabetes mellitus have a hemoglobin A1c level of <7% (normal, 4% to 7%), a preprandial capillary plasma glucose level of 90 to 130 mg/dL (5.0 to 7.2 mmol/L), and a peak postprandial capillary plasma glucose level of <180 mg/dL (10.0 mmol/L). Therefore, the determination of glycemic control is made with use of a combination of short-term and long-term parameters.
Glycemic control has been shown to be associated with outcomes in acute medical, general surgical, and trauma environments7-12, and other studies have demonstrated that elective surgery is associated with physiologic stress that can alter the ability of both diabetic and nondiabetic patients to regulate glucose metabolism13,14.
In the present study, a large, nationally representative database was utilized to retrospectively compare various outcomes in patients with controlled diabetes mellitus, patients with uncontrolled diabetes mellitus, and patients without diabetes mellitus who had undergone hip and knee replacement in the United States. The purpose of the present study was to determine whether the quality of glycemic control in diabetic patients is associated with the prevalence of perioperative complications following total hip arthroplasty and total knee arthroplasty. We hypothesized that, regardless of the type of diabetes, patients with uncontrolled diabetes mellitus would demonstrate significantly more perioperative complications when compared with patients with controlled diabetes and those without diabetes.
The present study involved the use of a national database that is publicly available and devoid of all protected health information; therefore, on the basis of our institution's guidelines, no institutional review board approval was necessary to perform the study or to report the findings of the study.
Database
The present study utilized data from 1988 to 2005 that were found within the Nationwide Inpatient Sample (NIS) database. The Nationwide Inpatient Sample is part of the Healthcare Cost and Utilization Project sponsored by the Agency for Healthcare Research and Quality and comprises the largest nationwide all-payer hospital inpatient care database in the United States. In contrast to other large national databases, the Nationwide Inpatient Sample includes data for patients with Medicare, Medicaid, and private insurance as well as patients without insurance.
Each year, the Nationwide Inpatient Sample contains discharge data from approximately 7 to 8 million hospital stays at approximately 1000 hospitals randomly selected to approximate a 20% stratified sample of hospitals across the United States. Comorbidity data are reflective of discharge diagnoses, and coding for the database is in accordance with the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM)15. The quality control and reliability of the Nationwide Inpatient Sample have been previously described16, and the accuracy is at least as good as that of the National Hospital Discharge Survey17. On the basis of the characteristics of the database and the use of sampling weights for statistical analyses, the results of the present study can be generalized to the entire population of patients undergoing total hip or knee arthroplasty in the United States.
Sample Selection
ICD-9-CM primary procedure codes indicating primary total hip arthroplasty (81.51), revision total hip arthroplasty (81.53), primary total knee arthroplasty (81.54), or revision total knee arthroplasty (81.55) were utilized for inclusion in the present study. We excluded patients undergoing primary or revision arthroplasties with ICD-9-CM diagnosis codes for pathologic fracture; metastatic cancer; infection of the knee or thigh, including acute or chronic osteomyelitis; infections around a device, implant, or graft; or primary malignant bone neoplasm. In addition, patients with diagnosis codes indicating femoral neck fractures, internal device failures, or complications (996 codes) were excluded.
We then categorized the included patients into three groups: those with controlled diabetes mellitus, those with uncontrolled diabetes mellitus, and those with no report of diabetes mellitus (nondiabetic patients). All patients included with diabetes mellitus and some variation of the disease spectrum were required to have either controlled or uncontrolled diagnostic codes. Uncontrolled diabetes was coded on the basis of the Agency for Healthcare Research and Quality's Healthcare Cost and Utilization Project and its suggested Prevention Quality Indicators, which included uncontrolled diabetes without complications and diabetes with short-term complications (ketoacidosis, hyperosmolarity, and diabetic coma)16-18. For example, we included the codes for uncontrolled diabetes without complications (ICD-9-CM codes 250.02 to 250.03) and the codes for diabetes with the short-term complications of ketoacidosis (ICD-9-CM 250.10 to 250.13), hyperosmolarity (ICD-9-CM 250.20 to 250.23), and diabetes-associated coma (ICD-9-CM 250.3) that have been previously used to track the current burden of uncontrolled diabetes during hospitalization19. Any patient with diabetes without the aforementioned uncontrolled parameters was considered to have controlled disease. Therefore, patients with controlled diabetes were described and coded within the database as "not uncontrolled." A full list of coding parameters can be found in the Appendix.
Main Outcome Measures
The main outcome measures in the present study were perioperative complications, mortality, the length of hospital stay, the hospital disposition (i.e., routine or nonroutine discharge), and the hospital charges (adjusted for inflation) during the index hospitalization. Specific complications included cerebrovascular accident, pneumonia, myocardial infarction, thrombophlebitis, deep-vein thrombosis, urinary tract infection, ileus, altered mental status, operative infection, other wound problems, postoperative hemorrhage/shock, transfusions, and fractures. Hip dislocation and sciatic nerve injury were also considered to be specific complications of primary and revision total hip arthroplasty cases. Complications were identified in the Nationwide Inpatient Sample on the basis of ICD-9-CM diagnosis or procedure codes and were reported as a dichotomous variable.
Covariates
Patient-specific covariates included age, sex, race, and median household income according to ZIP code (postal code) for each patient. In addition, we included hospital region, expected primary payer source, hospital bed size, hospital location, surgeon volume, and hospital volume. We further modified the Charlson index by removing the diabetes element of scoring to improve the comparison of comorbidities of the three groups20. This index summarizes patients' comorbidities with use of ICD-9-CM diagnosis codes and takes into account the severity of the specific diagnoses by weighting.
Statistical Analysis
Comparative descriptive statistics for three cohorts (patients with uncontrolled diabetes mellitus, patients with controlled diabetes mellitus, and patients without diabetes) included sex, age, race, household income, hospital region, expected primary payer source, hospital bed size, hospital location, surgeon volume, hospital volume, and our modification of the Charlson index. We also analyzed differences in a number of perioperative complications, length of stay, discharge status, and inflation-adjusted charges between patients with uncontrolled diabetes, patients with controlled diabetes, and patients without diabetes. Multivariate nonparametric analyses, including Pearson chi-square analyses, the Kruskal-Wallis test, or the Fisher exact test (when appropriate) were used to measure differences in complication variables among the cohorts.
Regression modeling with use of adjustments for all descriptive covariates was used to examine odds ratios for controlled and uncontrolled diabetes as compared with a reference variable of no diabetes, uncontrolled diabetes as compared with a reference variable of controlled diabetes, uncontrolled Type-I diabetes as compared with a reference variable of controlled Type-I diabetes, and uncontrolled Type-II diabetes as compared with a reference variable of controlled Type-II diabetes. Median comparisons of length of stay and inflation-adjusted charges (after adjustments for covariates and log linear transformation) were made to compare the uncontrolled diabetes, controlled diabetes, and no diabetes cohorts; the controlled and uncontrolled diabetes cohorts; the controlled Type-I and uncontrolled Type-I diabetes cohorts; and the controlled Type-II and uncontrolled Type-II diabetes cohorts. For all comparisons and modeling, the level of significance was set at p < 0.05.
Source of Funding
This study was funded internally by the Adult Reconstruction section of the Division of Orthopaedic Surgery at Duke University Medical Center, which receives unrestricted institutional support from Zimmer and DePuy. In general, funding is necessary to access the Nationwide Inpatient Sample database, and to pay, in part, the salary of our statistical editor.
Among patients identified in the Nationwide Inpatient Sample database as undergoing primary or revision hip or knee arthroplasty, we identified 920,555 nondiabetic patients, 105,485 patients with controlled diabetes mellitus, and 3973 patients with uncontrolled diabetes mellitus. Significant demographic differences as well as significant differences in hospital region, hospital location, and the modified Charlson index20 were observed among the three groups (Table I).
Significant differences were observed among the three groups in terms of the unadjusted comparative analyses of length of stay, discharge status, mortality, inflation-adjusted charges, and the prevalence of perioperative complications, including cerebrovascular accident, myocardial infarction, urinary tract infection, ileus, thrombophlebitis, pneumonia, infection, postoperative hemorrhage, and transfusion (p = 0.001 for all) (Table II).
We performed regression modeling of controlled and uncontrolled diabetes against a reference variable of no diabetes while adjusting for the variables of sex, age, race, household income, hospital region, expected primary payer source, hospital bed size, hospital location, surgeon volume, hospital volume, and modified Charlson index20. Patients with controlled diabetes were significantly less likely to have a routine discharge (adjusted odds ratio = 0.82; 95% confidence interval = 0.80 to 0.85) and were more likely to have a complication related to urinary tract infection (adjusted odds ratio = 1.25; 95% confidence interval = 1.17 to 1.33) and transfusion (adjusted odds ratio = 1.09; 95% confidence interval = 1.06 to 1.13) than patients without diabetes (p < 0.001 for all). Patients with uncontrolled diabetes were significantly more likely to have a complication such as mortality (adjusted odds ratio = 2.70; 95% confidence interval = 1.65 to 4.43), cerebrovascular accident (adjusted odds ratio = 4.06; 95% confidence interval = 2.35 to 7.02), urinary tract infection (adjusted odds ratio = 2.48; 95% confidence interval = 2.04 to 3.01), ileus (adjusted odds ratio = 2.38; 95% confidence interval = 1.65 to 3.44), infection (adjusted odds ratio = 2.31; 95% confidence interval = 1.42 to 3.75), postoperative hemorrhage or shock (adjusted odds ratio = 1.81; 95% confidence interval = 1.27 to 2.56), and transfusion (adjusted odds ratio = 1.29; 95% confidence interval = 1.13 to 1.47) than patients without diabetes (p < 0.001 for all). Patients with controlled diabetes were significantly less likely to have a myocardial infarction as compared with patients without diabetes (adjusted odds ratio = 0.28; 95% confidence interval = 0.13 to 0.64; p = 0.002) (Table III). Significant differences were found among the three groups in terms of the median length of stay and inflation-adjusted charges, with patients with uncontrolled diabetes incurring the highest amounts in both areas (p < 0.0001) (Table IV).
Adjusted regression modeling comparing uncontrolled and controlled diabetes is presented in tables in the Appendix. Patients with uncontrolled disease had significantly greater odds of cerebrovascular accident (adjusted odds ratio = 3.42; 95% confidence interval = 1.87 to 6.25; p < 0.001), urinary tract infection (adjusted odds ratio = 1.97; 95% confidence interval = 1.61 to 2.42; p < 0.001), ileus (adjusted odds ratio = 2.47; 95% confidence interval = 1.67 to 3.64; p < 0.001), infection (adjusted odds ratio = 2.28; 95% confidence interval = 1.36 to 3.81; p = 0.002), postoperative hemorrhage or shock (adjusted odds ratio = 1.97; 95% confidence interval = 1.38 to 2.87; p < 0.001), transfusion (adjusted odds ratio = 1.19; 95% confidence interval = 1.04 to 1.36; p = 0.011), and death (adjusted odds ratio = 3.23; 95% confidence interval = 1.87 to 5.57; p < 0.001) than those with controlled disease. Patients with uncontrolled diabetes also had a significantly longer length of stay (adjusted median length of stay, 5.12 compared with 4.28 days; p < 0.0001).
The adjusted regression models comparing patients with Type-I and Type-II diabetes, with use of controlled disease as the reference variable for each model, are presented in the tables in the Appendix. Patients with uncontrolled Type-I diabetes were more likely to have a cerebrovascular accident (adjusted odds ratio = 3.88; 95% confidence interval = 1.25 to 12.04; p = 0.019), a urinary tract infection (adjusted odds ratio = 2.16; 95% confidence interval = 1.51 to 3.11; p < 0.001), and a longer length of stay (adjusted median length of stay, 5.781 compared with 5.103 days; p < 0.001) than patients with controlled Type-I diabetes. Patients with uncontrolled Type-II diabetes were significantly more likely to have a cerebrovascular accident (adjusted odds ratio = 3.22; 95% confidence interval = 1.53 to 6.77; p = 0.002), a urinary tract infection (adjusted odds ratio = 1.78; 95% confidence interval = 1.39 to 2.30; p < 0.001), ileus (adjusted odds ratio = 2.98; 95% confidence interval = 1.97 to 4.53; p < 0.001), thrombophlebitis (adjusted odds ratio = 2.46; 95% confidence interval = 1.12 to 5.40; p = 0.025), postoperative hemorrhage (adjusted odds ratio = 2.00; 95% confidence interval = 1.28 to 3.13; p = 0.002), an infection (adjusted odds ratio = 3.17; 95% confidence interval = 1.78 to 5.65; p < 0.001), and a longer length of stay (adjusted median length of stay, 4.965 compared with 4.181 days; p < 0.001) as compared with patients with controlled Type-II diabetes; they were also more likely to die (adjusted odds ratio = 3.48; 95% confidence interval = 1.86 to 6.48; p < 0.001).
As changes in the blood sugar concentration in a patient with diabetes mellitus can acutely affect physiologic stability, serum glucose concentration at the time of admission has been shown to be an important predictor of outcomes in hospitalized patients. In particular, hyperglycemia at the time of admission is an independent predictor of morbidity and/or mortality in patients who are admitted for the treatment of acute surgical and medical emergencies such as stroke, acute coronary syndrome, heart failure, pneumonia, and trauma7-12. In contrast, the hemoglobin A1c level, which represents an average blood glucose concentration over a one to three-month period, has not been shown to be as consistent an outcomes predictor when used exclusively in the acute surgical and intensive care environments. However, a recent article by Dronge et al.21 reviewing the outcomes of several types of noncardiac surgery suggested that hemoglobin A1c levels of <7% were associated with a significantly lower risk of postoperative infections.
Several authors have reported that a diagnosis of diabetes mellitus is associated with an increased prevalence of adverse perioperative outcomes in patients undergoing total joint arthroplasty22-30. Those investigators stratified patients according to whether or not they were insulin-dependent; however, the delineation of subgroups on the basis of insulin dependency seemed to be focused toward the chronicity of disease or diabetic type as opposed to glycemic control.
Our review of the current literature on total joint arthroplasty suggested that only three reports have included a discussion of diabetes-control parameters in their methods section. Moeckel et al.24 reviewed the results of ninety-three hip replacements in seventy-eight diabetic patients after an average duration of follow-up of 4.1 years. They reported the mean serum glucose level for their cohort but provided no additional information regarding its impact on clinical outcomes. They noted, however, that twelve previously non-insulin-dependent patients required insulin during the perioperative period and that six of the twelve remained insulin-dependent at the time of the latest evaluation. The authors concluded that the continued insulin dependence at the time of the latest evaluation was due to deterioration of the diabetic condition. Therefore, stress-induced hyperglycemia occurs even in patients undergoing elective arthroplasty surgery and can potentially affect systemic outcomes in both the short term and the long term.
Papagelopoulos et al.26 evaluated long-term functional outcomes and implant survival in a study of sixty-eight total knee arthroplasties in fifty-one diabetic patients. When the study group was compared directly with a matched population of nondiabetic patients undergoing total knee arthroplasty, the authors reported a greater frequency of complications in the diabetic cohort (12% compared with 2%; p < 0.05). However, there was no statistical analysis of the surgical or systemic complications because of the small size of the groups. With additional univariate analysis, the authors found no association between the presence of diabetes mellitus-related systemic disease manifestations and implant survival. They suggested that systemic manifestations of diabetes mellitus affected clinical results as reflected by lower outcome and postoperative knee scores, but they did not statistically analyze the relationship of diabetes-related illness and functional outcomes or perioperative morbidity.
Chiu et al. evaluated the risk of postoperative infection in diabetic patients managed with cefuroxime-impregnated cement for primary total knee arthroplasty29. In their small cohort (seventy-eight knees), they found no significant increase in the risk of infection in association with disease type, duration of treatment, or preoperative or postoperative blood glucose concentrations.
To our knowledge, the present study is the first study that has evaluated glycemic control as an outcomes predictor in diabetic patients undergoing hip or knee arthroplasty. The results confirm our hypothesis that, regardless of diabetes type, patients with uncontrolled diabetes demonstrate significantly more perioperative complications as compared with patients with controlled diabetes or patients without diabetes. There were significant differences within the comparative analysis among the three groups with regard to most of the complication variables (Table II). On closer examination, regression modeling with adjustment for confounders demonstrated significant differences between both cohorts of diabetic patients and the cohort of nondiabetic patients. While patients with controlled diabetes demonstrated more frequent complications related to urinary tract infection, transfusion, and nonroutine discharge as compared with nondiabetic patients, patients with uncontrolled diabetes demonstrated higher odds of cerebrovascular accident, urinary tract infection, ileus, infection, postoperative hemorrhage, transfusion, and death. In a direct comparison between patients with uncontrolled and controlled diabetes with use of adjusted regression modeling, patients with uncontrolled disease clearly had a significantly greater propensity for systemic complications, death, and a longer length of stay.
When the cohort of diabetic patients was further stratified on the basis of the type of diabetes, adjusted regression analysis determined that the odds of cerebrovascular accident and urinary tract infection were significantly higher in patients with uncontrolled Type-I diabetes as compared with patients with controlled Type-I disease. In patients with uncontrolled Type-II diabetes, the odds were significantly higher not only for cerebrovascular accident and urinary tract infection but also for ileus, thrombophlebitis, postoperative hemorrhage, infection, and death. As glycemic control appears to be more critical in patients with Type-II diabetes with regard to surgical and systemic morbidity, additional studies will be necessary to evaluate the effect of diabetic type on perioperative morbidity in patients managed with arthroplasty.
Despite the large sample size provided by the Nationwide Inpatient Sample, the present study has several potential limitations. The coding has not been directly validated against clinical data, and access to patients' charts is not possible. While it is unlikely that substantial amounts of miscoding occurred, it is possible that certain data could have been underrepresented or even misrepresented, creating bias. For example, it is well established that race is not a reliable demographic, as certain hospitals do not supply those data to the Nationwide Inpatient Sample. The most obvious clinical parameter of concern would be the designation of diabetic control. While based on objective criteria, the designation of "controlled diabetes" is a subjective one. Also, there is clearly no one specific parameter that defines uncontrolled diabetes. Therefore, we cannot accurately determine whether the number of patients with uncontrolled diabetes is underrepresented or overrepresented.
Despite a large amount of health information within the Nationwide Inpatient Sample database, the database lacks reliable data concerning patient height and weight, and, therefore, analyses attempting to explore the role of obesity and glycemic control cannot be performed with use of these data. This limitation is particularly important given the relationship between Type-II diabetes and obesity. Outliers also pose a challenge as there is no way to confirm coding assessments. There was one outlier regarding patient outcomes in the current study. Nondiabetic patients had significantly higher odds of experiencing a postoperative myocardial infarction as compared with patients with controlled diabetes. It is not possible to glean from the database what preoperative screening protocols were in place, whether the diabetic patients were more likely to have a cardiology workup prior to surgery, or if the diagnosis was missed secondary to a silent event.
Although we adjusted for the variability in demographic and conditional values between the three groups during regression analyses, adjustments are no substitute for similarity in sampling. Therefore, the population differences are still considered to be a limitation.
Finally, the Nationwide Inpatient Sample only provides information related to hospital discharge. Because of these limitations, adverse events such as infections and thromboembolic disease are certainly underrepresented in the current study despite the achievement of significance in several comparisons, and, unless coded specifically as a postoperative complication, the temporal effect of selected disorders remains unknown. Nevertheless, high-quality datasets like the Nationwide Inpatient Sample are able to provide nationally representative epidemiologic studies over an extended period of time.
It has been well established that a diagnosis of diabetes mellitus is associated with increased perioperative morbidity and mortality in patients managed with arthroplasty5,22-30. The present study demonstrated that uncontrolled diabetes mellitus is a significant outcomes predictor for surgical and systemic complications, mortality, increased length of stay, and higher hospital charges, particularly in patients with Type-II diabetes. To further validate these results, a prospective study that compares glycemic control in diabetic patients and nondiabetic patients is warranted. We recommend that such a study include perioperative optimization of blood-sugar management as part of its study protocol.