Geographic Variation in Anticoagulant Use and Resident,Nursing Home,and County Characteristics Associated With Treatment Among US Nursing Home Residents

ObjectivesTo quantify geographic variation in anticoagulant use and explore what resident, nursing home, and county characteristics were associated with anticoagulant use in a clinically complex population.DesignA repeated cross-sectional design was used to estimate current oral anticoagulant use on December 31, 2014, 2015, and 2016.Setting and ParticipantsSecondary data for United States nursing home residents during the period 2014-2016 were drawn from the Minimum Data Set 3.0 and Medicare Parts A and D. Nursing home residents (≥65 years) with a diagnosis of atrial fibrillation and ≥6 months of Medicare fee-for-service enrollment were eligible for inclusion. Residents in a coma or on hospice were excluded.MethodsMultilevel logistic models evaluated the extent to which variation in anticoagulant use between counties could be explained by resident, nursing home, and county characteristics and state of residence. Proportional changes in cluster variation (PCVs), intraclass correlation coefficients (ICCs), and adjusted odds ratios (aORs) were estimated.ResultsAmong 86,736 nursing home residents from 11,860 nursing homes and 1694 counties, 45% used oral anticoagulants. The odds of oral anticoagulant use were 18% higher in 2016 than 2014 (aOR: 1.18; 95% confidence interval: 1.14-1.22). Most states had counties in the highest (51.3-58.9%) and lowest (31.1%-41.4%) deciles of anticoagulant use. Compared with the null model, adjustment for resident characteristics explained one-third of the variation between counties (PCV: 34.8%). The full model explained 65.5% of between-county variation. Within-county correlation was a small proportion (ICC < 2.2%) of total variation.Conclusions and ImplicationsIn this older adult population at high risk for ischemic stroke, less than half of the residents received treatment with anticoagulants. Variation in treatment across counties was partially attributable to the characteristics of residents, nursing homes, and counties. Comparative evidence and refinement of predictive algorithms specific to the nursing home setting may be warranted.