It is estimated that each year 260,000 Americans sustain an ankle
fracture, and that 25% of them undergo surgical
stabilization1. Many
small case studies, retrospective reviews, and case-control studies have
focused on the unique features of ankle fracture in patients with diabetes
mellitus that affect management and the appropriate use of surgical
intervention, such as the association between poor wound-healing and the need
for increased care when mobilizing a
patient1-8.
These studies have demonstrated that careful surgical fixation in conjunction
with close attention to postoperative blood glucose levels as well as other
general medical issues (e.g., vigilant wound care) can reduce morbidity and
mortality. However, these studies have all been relatively small, and it is
difficult to extrapolate a clear picture of the nationwide impact of diabetes
mellitus on the cost, mortality, length of hospital stay, and rates of
in-hospital complications and non-routine discharge associated with the
surgical repair of an ankle fracture.
The objective of this investigation was to analyze a nationally
representative sample to compare the length of hospital stay, the rates of
in-hospital complications and mortality, the total cost, and the rate of
non-routine discharge in a cohort of diabetic patients with ankle fractures of
varying severity with those factors in a non-diabetic cohort with ankle
fractures. We hypothesized that the values for all of the dependent variables
would be significantly worse for the patients with diabetes than for the
patients without diabetes.
Database Description and Sample Selection
We utilized the Nationwide Inpatient Sample (NIS) databases for the
years 1988 through 2000 to establish a retrospective cohort of individuals
with an ankle fracture. The NIS databases provide demographic data, admission
and discharge dates, records of inpatient stay including information on
clinical and resource use, and discharge status. ICD-9 CM (International
Classification of Diseases, Ninth Revision, Clinical Modification) procedure
and diagnostic codes are used in these databases. The NIS is part of the
Healthcare Cost and Utilization Project (HCUP), sponsored by the Agency for
Healthcare Research and Quality (AHRQ), and contains information on 5 to 8
million hospital stays at about 1000 randomly selected hospitals to
approximate a 20% stratified sample of hospitals in the United States.
Sampling strata were employed for the creation of the NIS on the basis of five
hospital characteristics (geographic region, type of ownership, location,
teaching status, and bed size) to ensure maximal representativeness of the
United States
population9. The
quality control and reliability of the NIS were examined by reviewing the
univariate statistics for all numeric data elements, with checking of ranges
against norms, examination of frequency distributions for all categorical and
some continuous data elements, and performance of edit checks that identify
inconsistencies between related data
elements10. The NIS
has been shown to be as accurate, if not more accurate for many estimates, as
the National Hospital Discharge Survey
(NHDS)11.
We identified patients with ICD-9 CM diagnostic codes for ankle fracture
and ICD-9 CM procedure codes for surgical intervention (79.00, 79.06, 79.09,
79.10, 79.16, 79.19, 79.20, 79.26, 79.29, 79.30, 79.76, 79.79, 79.80, 79.86,
79.89, 79.90, 79.96, 79.99, 824.0-824.9, and 837.0-837.1) in the NIS for the
years 1988 through 2000. This selection process led to a sample size of
248,028 hospitalizations. Once the data had been collected for use in our
study, they were examined for valid coding—i.e., the ICD-9 CM codes were
checked for correct form and were modified if necessary. We excluded from the
analysis hospitalizations of patients with a diagnosis other than ankle
fracture with subsequent surgical intervention including those with polytrauma
(ICD-9 CM codes 79.00, 79.06, 79.09, 79.70, 79.76, 79.79, 800-823.9,
825-836.69, 838.00-844.9, 845.1-890.2, 892.0-909.9, 925.1-928.1, 928.11,
928.20, 928.3-958.2, and 958.4-959.9). This set of exclusions removed 76,647
hospitalizations, for a remaining total of 171,381. Patients younger than the
age of eighteen years (10,783 hospitalizations) were also excluded from the
analysis. The final number of hospitalizations included in the sample was thus
160,598 (Fig. 1).
Data Elements
Dependent variables analyzed in this study included the length of hospital
stay, in-hospital complications, in-hospital mortality, rate of non-routine
discharge, and fracture severity.
Length of hospital stay: This variable, which was measured in
days, was defined as the difference between the patient's date of admission
and his or her date of discharge. The length of stay was coded as 0 for
patients discharged within twenty-four hours after admission.
In-hospital complications: We examined all non-fatal in-hospital
morbidity, with any cause, on the basis of the ICD-9 CM codes. Since the NIS
contains in-patient data only, complications occurring after hospital
discharge were not included in our analysis.
In-hospital mortality: The NIS databases include information on
patients' vital status while they are in the hospital. However, these
databases do not contain information on patients after their release from the
hospital; therefore, deaths occurring after hospital discharge were not
included in our analysis.
Rate of non-routine discharge: The NIS provides information about
the patient's discharge status in the form of eight classifications: (1)
routine discharge, (2) short-term hospital stay, (3) skilled nursing facility,
(4) intermediate care, (5) another type of facility, (6) home health care, (7)
against medical advice, (8) death. The 274 patients who died during
hospitalization were excluded from the analysis of this specific end point.
(Mortality information is presented separately in
Table III.) The remaining
patients were grouped according to whether they had a routine discharge (1) or
a non-routine discharge (2 through 7).
Evaluation of end points stratified by fracture severity: We
assessed the end points in stratified analyses according to ankle fracture
severity (ICD-9 CM diagnosis code 824). Stratified analyses were performed
because open fractures and fracture-dislocations are important clinical
features that could substantially affect the end points. Unimalleolar closed
fractures were classified as the least severe, followed by bimalleolar or
trimalleolar closed fractures, dislocations, and finally open
fractures12.
Main effect: The main effect variable in the present investigation
was the presence or absence of a diagnosis of diabetes mellitus (either
controlled or uncontrolled Type-I or Type-II diabetes mellitus; ICD-9 CM
diagnosis codes 250.00 through 250.03).
Covariates: The covariates considered in this analysis included
age, gender, household income, hospital volume, and severity of the ankle
fracture. Household income was defined as the median household income in the
area defined by the patient's ZIP code. The information pertaining to income
in the NIS database originated from United States Census
data13. When the
data were first received from the NIS database, this variable was originally
divided into eight categories: 1 = $0 to $15,000, 2 = $15,001 to $20,000, 3 =
$20,001 to $25,000, 4 = $25,001 to $30,000, 5 = $30,001 to $35,000, 6 =
$35,001 to $40,000, 7 = $40,001 to $45,000, 8 = $45,001. Hospital volume
was defined as the total number of discharges per year. Ankle fracture
severity was defined as described above.
Statistical Analyses
All statistical analyses were performed with use of Stata, version 8.0, for
Unix (StataCorp, College Station, Texas). We used means and standard
deviations for descriptive analysis of continuous variables and frequencies
and percentages for categorical variables. The association between categorical
dependent variables and the main predictor was examined with use of chi-square
tests. The relationship between the presence or absence of diabetes and
confounders, including hospital volume, age, gender, and income, was analyzed
with use of t tests and chi-square tests. The same tests were used to examine
the relationship between dependent variables and confounders.
Multiple linear regression models were used to examine the risk-adjusted
association between diabetic and non-diabetic patients with fractures of
varying severity and length of hospital stay as well as total charges. Length
of hospital stay, a continuous variable, was modeled with use of
log-transformed length of stay. The estimated risk-adjusted median length of
stay was predicted by creating an exponential of the estimated log-transformed
length of stay. Total charges were also modeled with use of its logarithmic
transformation and predicted by creating an exponential of the estimated
log-transformed total charges.
Multiple logistic regression analyses were used to assess the risk-adjusted
impact of the presence of diabetes on the rate of non-routine discharge, the
occurrence of complications, and mortality. Linear and logistic regression
models were adjusted for hospital volume, age greater than sixty-five years,
gender, and income. After the initial analysis, the open-fracture and
fracture-dislocation groups were merged for all tables in order to provide
enough observations in this category. Unimalleolar and
bimalleolar/trimalleolar fracture groups were also merged because of the
extremely small numbers of observations for some of the dependent variables
that were examined. In the regression models, the median-income variable was
analyzed as a four-level indicator variable in which the first two original
income designations were combined to create the indicator variable. The lowest
income group was classified as the referent. In the regression models, a
dichotomized variable was used to indicate an age of sixty-five years or older
and an age of less than sixty-five years.
This study demonstrated that diabetic patients with an ankle
fracture had a significantly increased mortality rate, rate of postoperative
complications, length of hospital stay, rate of non-routine hospital
discharge, and cost associated with the hospital stay compared with
non-diabetic patients with an ankle fracture. These results are all consistent
with one another and are supported by the literature. Many studies have
demonstrated that patients with diabetes have a higher rate of complications
and more severe complications than do other
patients1-8,14,15.
However, the specific outcomes in these studies are difficult to compare
directly with one another because different definitions were utilized for
postoperative complications. In addition, the designs of the studies have
varied, with some authors examining both operative and nonoperative
intervention and others analyzing only surgical treatment. Despite these
problems, the rates of postoperative complications in these studies have
clearly been higher than those found in our study (in which postoperative
complications were considered in their broadest sense, with inclusion of
infection and amputation). This difference may be partly due to our inability
to determine whether postoperative complications developed following discharge
of our patients. However, it should be noted that, to our knowledge, much
smaller sample sizes, generally less than a few hundred patients, were
analyzed in all other studies in the literature. In addition, those studies do
not represent patients from across the nation; instead, the authors usually
relied on data gathered at their home institution over a handful of years.
Our study had some limitations. First, it was based on data from a large
administrative database, which may have some level of miscoding. In addition
to frank miscoding, there may be instances in which diagnosis is subjective,
resulting in alternate coding and/or reporting. However, some factors that
were analyzed, such as length of hospital stay, in-hospital mortality, and
type of discharge, are exempt from considerations of subjectivity. A second
limitation of the NIS database is that it records only in-hospital mortality
and complications; outcomes that occur after discharge are not contained in
the database. Also, the NIS provides information on hospitalizations, not
particular patients; therefore, multiple hospitalizations of individual
patients could cause a lack of independence in portions of the data, resulting
in improper assumptions in the statistical analysis that was used. However,
the likelihood of a significant number of multiple hospitalizations in the
analyzed data is
small16. Finally,
the NIS database is primarily an administrative one and thus does not include
extensive patient information. Specifically, the use of ICD-9 CM codes to
define ankle fractures does not allow more precise grouping of specific types
of ankle fractures that may be of interest to the clinician. Although we tried
to account for all variables that may affect the results on which information
is available in the NIS database, additional studies are needed to more
completely evaluate the effect of additional factors, such as the severity
and/or duration of the diabetes mellitus, on outcomes.
Despite these limitations, our study had the advantage of including a large
number of hospitalizations that were representative of the entire United
States inpatient hospital population, and it was larger than previous studies
in which similar end points were
examined1,2,5.
We also analyzed many years' worth of information, so that we were able to
study the changing impact of diabetes mellitus on ankle fractures at different
points in time. For instance, although we did not focus extensively on trend
analysis, information such as the increasing percentage of patients, of both
genders, with diabetes in the United States population over time can have a
substantial impact on the health-care system and should be explored further in
future studies. In addition, the approximately equal distribution of the
sample across all income categories in our study means that its findings are
more likely to be applicable to a wide range of income levels in the
population.
We found that patients with diabetes have a significantly increased
mortality rate, rate of postoperative complications, length of hospital stay,
rate of non-routine discharge, and hospital costs than patients without
diabetes. In addition, although fracture severity was not a particular focus
of research in this study, our findings support the previously noted
implication in the literature that patients with diabetes may be at increased
risk of sustaining a more severe ankle fracture than patients without
diabetes1-8,14,15,17.
(In our study, patients with diabetes had a higher risk of sustaining a closed
bimalleolar or trimalleolar fracture than of sustaining a closed unimalleolar
fracture.)
These findings may have important implications for health care.
Specifically, they may be used to justify particular resource-allocation
protocols for the purpose of reducing morbidity and mortality of both diabetic
and non-diabetic patients to the lowest levels possible. Our finding of small
absolute differences in in-hospital mortality between patients with diabetes
and those without diabetes and our finding of lower rates of postoperative
complications in diabetic patients than have been reported in previously
published studies should be explored in greater depth. Additional studies that
incorporate variables such as quality-of-life assessments into models, so that
the true cost of these negative outcomes can be established in human terms
rather than in dollar figures alone, should be pursued. ?