Are Neighborhood Characteristics Associated With Outcomes After THA and TKA? Findings From a Large Healthcare System Database

Non-White patients have higher rates of discharge to an extended care facility, hospital readmission, and emergency department use after primary THA and TKA. The reasons for this are unknown. Place of residence, which can vary by race, has been linked to poorer healthcare outcomes for people with many health conditions. However, the potential relationship between place of residence and disparities in these joint arthroplasty outcomes is unclear. (1) Are neighborhood-level characteristics, including racial composition, marital proportions, residential vacancy, educational attainment, employment proportions, overall deprivation, access to medical care, and rurality associated with an increased risk of discharge to a facility, readmission, and emergency department use after elective THA and TKA? (2) Are the associations between neighborhood-level characteristics and discharge to a facility, readmission, and emergency department use the same among White and Black patients undergoing elective THA and TKA? Between 2007 and 2018, 34,008 records of elective primary THA or TKA for osteoarthritis, rheumatoid arthritis, or avascular necrosis in a regional healthcare system were identified. After exclusions for unicompartmental arthroplasty, bilateral surgery, concomitant procedures, inability to geocode a residential address, duplicate records, and deaths, 21,689 patients remained. Ninety-seven percent of patients in this cohort self-identified as either White or Black, so the remaining 659 patients were excluded due to small sample size. This left 21,030 total patients for analysis. Discharge destination, readmissions within 90 days of surgery, and emergency department visits within 90 days were identified. Each patient's street address was linked to neighborhood characteristics from the American Community Survey and Area Deprivation Index. A multilevel, multivariable logistic regression analysis was used to model each outcome of interest, controlling for clinical and individual sociodemographic factors and allowing for clustering at the neighborhood level. The models were then duplicated with the addition of neighborhood characteristics to determine the association between neighborhood-level factors and each outcome. The linear predictors from each of these models were used to determine the predicted risk of each outcome, with and without neighborhood characteristics, and divided into tenths. The change in predicted risk tenths based on the model containing neighbo