Data source
Data was extracted from the Nationwide Inpatient Sample (NIS), 2009–2011. NIS, considered
the largest publically available all-payer inpatient database in the United States,
includes data from all states that participate in the Healthcare Cost and Utilization
Project (HCUP), sponsored by the Agency for Healthcare Research and Quality (AHRQ).
An annual approximate sample of 8 million hospitalizations from 1,000 hospitals, reflecting
a 20% stratified sample of community hospitals in the nation are included in NIS.
NIS excludes short-term rehabilitation hospitals (starting 1998 data), long-term non-acute
care hospitals, psychiatric hospitals, and alcoholism/chemical dependency treatment
facilities. All hospitals in NIS are stratified based on five hospital characteristics:
ownership/control, bed size, teaching status, urban/rural location, and geographic
location. Starting 1988, NIS data is available yearly and further details of the dataset
are available elsewhere 17].
Data collection and study definitions
Our study sample included hospital primary discharges with pneumonia or UTI in adults
over the age of 17Â years. The International Classification of Diseases, 9
th
Revision, Clinical Modification (ICD-9-CM) was used to identify UTI (599.0) and CDI
(008.45). Pneumonia was assessed using Clinical Classification Software (CCS) code
of 112, representing cases not caused by tuberculosis or sexually transmitted disease
(as defined by NIS). Similar coding strategies have been utilized in previous literature
for CDI 11],13],14], pneumonia 18],19], and UTI 20],21]. NIS provides a primary discharge code in addition to additional codes of secondary
diagnoses. In this study, patients with secondary discharge code for CDI were identified
as CDI patients.
We further included both patient and hospital characteristics, to both control for
and evaluate if such characteristics negatively impact CDI patients. For example,
previous research has demonstrated both patient socioeconomic factors and hospital
characteristics, to be positively associated with length of stay and mortality in
other patient populations 11],13],14],22],23], and thus such variables were further accounted for in our study.
Patient characteristics included were: age (18–34 years, 35–49 years, 50–64 years,
65Â years or more), gender (men, women), race/ethnicity (White, Black, Hispanic, Asian
or Pacific Islander, Native American, other), primary payer type (Private including
HMO, Medicare, Medicaid, Self-pay, no charge, other), neighborhood income defined
as median household income quartiles by patient ZIP code ($1-$38999, $39000-$47999,
$48000-$62999, $63000 or more) and Charlson-Deyo Index (0, 1, 2 or more). The Deyo
modification of the Charlson comorbidity index was used, which creates a score representing
co-morbidities for each discharge utilizing the ICD-9-CM coding algorithms. The 17-item
index is a validated measure of comorbidity for administrative data 24]-26].
Hospital characteristics included in the study were: bed size tertile categories (small,
medium, large), ownership/control (private investor-own, private non-profit, government
nonfederal), setting/teaching status (rural, urban non-teaching, urban teaching),
and geographic location (Northeast, Midwest, South, West). In-hospital mortality was
defined as those who died during hospitalization versus those who did not. LOS and
total charges were used from NIS-provided variables, which were edited by AHRQ to
ensure uniformity between states. Total charges were adjusted quarterly for inflation
using the Gross Domestic Product (GDP) deflator available through the United States
Department of Commerce, Bureau of Economic Analysis with 2009 USD as the reference
year 27].
Statistical analyses
SAS 9.4 (SAS Institute, Inc., Cary, NC) was used for all statistical analyses except
for negative binomial regression, for which we used the STATA 12 package (Stata Corp
LP, College Station, TX). Given that existing data suggests potential gender differences
in CDI 28]-30] all statistical analyses were stratified by gender. Due to the large number of variables
and in turn multiple testing, a family-wise correction using the Bonferroni adjustment
was conducted to reduce type I error rate. As a result, P??0.0017 was set as the level of significance.
To assess CDI prevalence (for each patient population), in addition to patient and
hospital characteristic differences between each gender among pneumonia and UTI cases,
chi-square tests using design-based F values were used. The prevalence of secondary CDI was noted as cases per 1,000 discharges
for pneumonia and UTI groups, by gender. Next, independent survey-weighted multivariable
logistic regression analyses were performed to identify patient and hospital characteristics
associated with prevalence of secondary CDI in both primary pneumonia and UTI discharges.
In order to identify the impact of secondary CDI on in-hospital mortality among patients
hospitalized for primary pneumonia or UTI, chi-square tests were conducted followed
by survey-weighted logistic regression analyses. To assess the impact of secondary
CDI on resource utilization, Wilcoxon rank sum test was used, followed by survey-weighted
negative binomial regression and survey-weighted linear regression for LOS and total
charges, respectively. For all adjusted models in each regression analyses, control
variables of survey year, patient, and hospital characteristics were included. Since
the distribution of total charges was non-normal and skewed to the right, this variable
was natural log transformed for linear regression analyses. In addition, given that
descriptive analyses demonstrated a significantly higher percent of our population
as 65 or older and the elderly are more likely to have negative health impacts 31]-33], a sensitivity analysis was performed in the aforementioned adjusted models among
patients aged 65 and older. Model building for all analyses included assessment of
assumptions and relevant interaction terms (sociodemographic characteristics with
hospital characteristics), with significance established at P??.05. The study was submitted to the University of Wisconsin-Madison Institutional
Review Board and was considered exempt from review.