We obtained data for all hospitalizations in California during the years 1995 through 2005 from California’s Office of Statewide Health Planning and Development (OSHPD). The OSHPD database is compiled annually and includes discharge abstracts from all licensed nonfederal hospitals in California [11
]. Each discharge abstract reports demographic information that includes age, gender, insurance type, and the race or ethnicity of the patient. In addition, International Classification of Diseases, 9th Revision (ICD-9) codes are entered into the record for each patient; the number of codes entered is not prespecified and the maximum allowed is up to 20 inpatient procedures and 24 diagnoses per hospitalization (Table ). Hospital characteristics are also reported, including the teaching status and whether a hospital is classified as rural in location. The OSHPD state inpatient database was initiated as a component of the Healthcare Cost and Utilization Project (HCUP) and is collected through mandatory reporting by all nonfederal hospitals in the state of California. Institutional Review Board approval was obtained for this study.
Demographics of patient sample
We identified 138,399 patients undergoing their first THA using the ICD-9 procedure code for primary THA (81.51) who met inclusion and exclusion criteria. A previously published coding algorithm was modified and used to exclude 20,291 patients with infection, pathologic fracture, or undergoing revision arthroplasty [4
] (Appendix 1). We also excluded 3,848 patients with a non-California zip code to decrease the probability of the patient having prior admissions meeting exclusion criteria or experiencing a subsequent complication treated outside of the state. The unit of analysis was hospital discharge for each patient. All patients had basic demographic data as mandated by the state reporting requirements so no patients were excluded for missing data. Baseline patient characteristics were recorded in the database and analyzed. The mean age of the patient sample was 66 years with 85% being white. The population was diverse with 4% being black, 7% Hispanic, and 2% Asian. Complicated diabetes is defined as diabetes associated with end-organ damage; uncomplicated diabetes was noted in 8%, whereas less than 1% of patients had complicated diabetes. A diagnosis of rheumatoid arthritis was noted in 4% of patients (Table ).
We selected the primary patient-based predictors: the Charlson comorbidity index [1
], age, race, gender, and income using zip code as a proxy as reported in the OSHPD database crossreferenced to US Census data. The Charlson comorbidity index assesses 19 comorbid conditions and has been validated for use in administrative database studies [1
]. This study uses the approach of Deyo et al. that adapted the Charlson index by defining the 19 comorbid conditions using ICD-9-CM coding and subsequently determining if the relevant codes are included in a patient record [1
]. In addition to the Charlson score, individual comorbidities were included for separate analysis consisting of diabetes, peripheral vascular disease, and rheumatoid arthritis.
Hospitals characteristics included surgical volume of THA, rural location, and teaching status. Teaching status and rural location are self-reported by the participating hospitals. Surgical volume was defined as the average number of primary THAs performed yearly during the study period. Hospitals were classified by their annual average volume as high-, intermediate-, or low-volume hospitals. Hospitals were categorized as low-volume if they were in the lowest 40th percentile by annual volume among hospitals where THA was performed. Intermediate-volume hospitals were defined as the next 40th percentile; high-volume hospitals were defined as the highest 20th percentile.
The outcomes analyzed as the dependent variables were the aggregate rate of short-term complications as well as the separately analyzed rates of individual complications, including mortality or readmission for the specific complications of infection, dislocation, revision surgery, perioperative fracture, neurologic injury, and thromboembolic disease at 90 days postoperatively. Previously published algorithms [4
] were adapted to detect codes consistent with a complication. The coding algorithms use ICD-9 nomenclature to identify patients undergoing total hip replacement using the 81.51 procedure code. Additional associated diagnoses, exclusion criteria, and complications are defined based on ICD-9 procedure and diagnoses codes judged by the authors to be consistent with the diagnoses or complications of interest. These algorithms were modified to correct for coding changes made during the study period [7
] (Appendix 1). Mortality was identified by the linkage of the California State Death Statistical Master File to the OSHPD database. This allowed us to identify hospital deaths occurring after discharge and the time elapsed before death in patients undergoing primary THA. The DSMF is a database of death certificates for all individuals dying in California and of those California residents who die outside of California’s borders but within the United States [13
We used multiple variable logistic regression models to determine the role of the patient and provider characteristics as independent variables in predicting the occurrence of the complications selected as dependent variables. This method allows us to report the odds ratio for each patient and provider independent variable adjusted for all of the other variables included in the model. The regression models included the patient characteristics of race/ethnicity, age, gender, income, specific comorbidities, and modified Charlson comorbidity index and the provider characteristics of hospital volume, rural location, and teaching status as independent variables. The strength of association between the risk of a complication and the patient and provider characteristics is reported as the odds ratio in relation to a reference group adjusted for all the other variables included in the model. P-values and 95% confidence intervals are reported with the odds ratios. All statistical analyses were conducted using Stata/SE 8.0 (Stata Corp, College Station, TX).