Data sources and study population
This retrospective cohort study used the data from the US National (formerly, Nationwide) Inpatient Sample (NIS), a 20% stratified sample of all discharges from US community hospitals, provided by the Agency for Healthcare Research and Quality (AHRQ) Healthcare Cost and Utilization Project (HCUP) . The University of Alabama at Birmingham’s Institutional Review Board (IRB-120207004) approved this study and waived the need for informed consent. Study methods and results are reported in accordance with the Strengthening of Reporting in Observational studies in Epidemiology (STROBE) statement  (Additional file 1).
We included all hospitalizations for primary THA in the US from 1998 to 2014, identified by the presence of International Classification of Diseases, ninth revision, common modification (ICD-9-CM) code of 81.51, as the primary procedure code for hospitalization. The code-based algorithm is a validated approach to identify THA cohorts with positive predictive values of 98–99% . The underlying diagnosis for THA was listed as the primary diagnosis ICD-9-CM code.
Exposure of interest, study outcomes, and covariates
The exposure of interest, AKI, was identified by the presence of the ICD-9-CM code of 584.5, 584.6, 584.7, 584.8, or 584.9 as a non-primary (i.e., secondary) diagnosis. This code-based algorithm for AKI is also a validated approach with high positive predictive value of 98% .
Study outcomes were healthcare utilization and in-hospital post-operative complications. We assessed the length of hospital stay (above/below the median; 2.7 days rounded off to 3 days), the total hospital charges in US dollars (above/below median for each year), and the discharge disposition, i.e., to home vs. a rehabilitation facility which included short- or long-term care hospital, skilled nursing facility (SNF), intermediate care facility, or a certified nursing facility. Patients who left the hospital against medical advice were considered missing for discharge disposition. The key in-hospital post-operative complications include implant infection, transfusion, and THA revision and mortality.
Important covariates/potential confounders which included demographics (age, sex, race/ethnicity, income), the underlying diagnosis (osteoarthritis (OA), rheumatoid arthritis (RA), fracture, avascular necrosis of bone or other), comorbidity (Deyo-Charlson comorbidity index, a validated measure that included 17 comorbidities, based on the presence of ICD-9-CM codes), insurance payer (Medicare, Medicaid, private insurance, self-pay or other), and hospital characteristics were assessed. Hospital location/teaching status was categorized as rural, urban non-teaching, or urban teaching hospital. Hospital bed size was classified as small, medium, or large, using the NIS cutoffs that vary by the year. Hospital region was categorized as Northeast, Midwest, South, and West.
We performed a separate multivariable-adjusted logistic regression models to assess each clinical (infection, transfusion, THA revision, and mortality) and healthcare utilization outcome (total hospital charges above/below the median, discharge to a rehabilitation facility, length of stay above/below the median of 3 days), adjusting for important potential predictors of post-THA complications and healthcare utilization and/or potential confounders of AKI outcomes . We calculated the odds ratios (OR) and 95% confidence intervals (CI). We performed sensitivity analyses by additionally adjusting the main analyses for hospital characteristics. We used SAS 9.3 (Cary, NC) for all analyses.