Fig. 1

Conceptual overview of process used to transform pharmacy invoices into a database suitable for analysis. A single dispensing pharmacy provided several years of invoice data for nursing homes in the same healthcare corporation. We began with aggregating the data, which varied in the number and names of columns. This process revealed large gaps or irregularities in the data, such as a nursing home that entered or left the corporation during the study period; these nursing homes were removed. This yielded a single aggregated database with a consistent format. Data cleaning was a multi-step process leading to a dataset consisting of a single row to describe each prescription (including refills) for systemic medications administered on a regular schedule. This was the dataset used to analyze antibiotic use across a large number of nursing homes. Please see the text for further details