We used the Medicare 5% national sample administrative claims database to identify patients undergoing primary TKA between January 1, 1997, and December 31, 2006. Each year, this data set includes the healthcare records in inpatient, outpatient, Part B carrier, skilled nursing facility, hospice care, home health, and durable medical equipment analytic data files. To limit the study to the elective cohort, we excluded patients undergoing primary TKA who received their implant as a result of bone cancer, joint infection, or fracture using criteria identical to those used previously by Katz, Mahomed, and coworkers [9
]. Patients younger than 65 years old or health maintenance organization (HMO) enrollees were also excluded. An overall cohort of 82,362 patients undergoing primary TKA was initially identified between 1997 and 2006, from which 69,663 elective patients were included in our study. From this elective patient cohort, 1400 TKA infections were identified.
During the period that a patient is enrolled in Medicare, all the claims associated with the patient are collected in the database. Because each enrollee is assigned a unique, encrypted Medicare beneficiary identifier, this was used to follow each patient longitudinally by compiling all their claims throughout the 10-year study period. Patients’ Medicare entitlement status and mortality were tracked using a linked “denominator” file provided by the CMS that accompanied the analytic data sets. International Classification of Diseases, 9th Revision, Clinical Modification (ICD-9-CM: 81.54) and Current Procedural Terminology, 4th Edition (CPT-4: 27447) procedure codes were used to identify patients undergoing primary TKA.
PJI was identified with the ICD-9-CM diagnosis code 996.66 in any component of the Medicare data set. We also performed a post hoc analysis to determine the clinical setting (ie, inpatient versus outpatient care) in which the deep infection was diagnosed. Patients diagnosed with a PJI were classified as early onset if it appeared within eight quarters (2 years) of the primary procedure. Late infection was classified if it was diagnosed for the first time more than 2 years after implantation.
Kaplan-Meier survivorship curves were compiled with PJI as an end point, adjusted for comorbidities using the Charlson index, and censored appropriately for patient mortality during the study period. The Charlson comorbidity index was based on the implementation for administrative data sets recommended by Deyo and coworkers [5
]. Based on the diagnosis and surgeries indicated from the claims records, a “weight” with values of 1, 2, 3, or 6 is assigned to each disease category and the final index is a composite value representing the overall degree of comorbidity. For our analysis, the scores were grouped into: 0 (none), 1 to 2 (low), 3 to 4 (moderate), and 5 or greater (high). Anesthesia time from the claims data was used as a proxy for procedure duration with categories of less than 120, 120 to 150, 150 to 180, 180 to 210, and 210 or greater minutes [20
]. Cox regression was used to evaluate the effects of patient factors (age, race, gender, Medicare buy-in status), Census region, procedure duration, and hospital characteristics (urban versus rural location, teaching status, bed size, ownership) on the relative risk of PJI. The relative risks were computed by adjusting for these covariates. The Medicare buy-in status was an identifier of patients whose Medicare premiums and deductibles were subsidized by the state as a result of their financial status and was used as a proxy for the patient’s socioeconomic status. The classifications of the hospital characteristics were based on those denoted by CMS in the claims files. Statistical analyses were performed using SAS (Cary, NC).