Shipments during this time period were sent overnight to their destination (regardless of distance), to arrive when receiving locations within the state were open. We categorized shipments (over 75%) by the type of provider through a series of targeted queries we generated. Thus, we calculated proportion of shipments or doses PTP to providers focused on children, primary care, county health departments, unclassified ABT-199 cost medical doctors, internists, specialists, long-term care, veterans, urgent care, hospitals, clinics, pharmacies, jails, military, government, universities, and nursing homes. The category of “specialists”

includes providers that we could identify as associated with caring for the ACIP population categorized as high-risk because of health conditions such as asthma, heart disease, diabetes, etc. We also combined these in several subgroupings

driven by like characteristics that might explain differences in coverage: selleck chemical e.g., general internists and specialists combined (internists and specialists can be grouped because both serve adults; however, while internists may provide primary care, adults may be less likely to visit internists or specialists during a short campaign); targeted access (doses sent to long term care, internists, specialists, nursing homes, and children); and general access locations (primary care, MDs that could not be classified by specialization, counties, hospitals, urgent care, clinics, or pharmacies). Using cross-sectional

data, we developed a regression model to predict vaccine coverage in adults, as of the end of January 2010, for DC and each state [1]. In a separate analysis, we constructed distinct models for children (6 m to 17 y) and high others risk adults (25–64 with a chronic condition) because we expected factors affecting coverage to differ across groups; we present those analyses in a separate paper. We calculated simple descriptive statistics (means, standard deviations, proportions, and measures of association including Pearson’s correlation). The primary technique used for modeling was multivariate linear regression (ordinary least squares) with transformations specified when used. Data were linearly scaled to values in [0,1] before performing regressions. Variable selection is a challenging problem [32], and our analysis poses additional difficulties because of high correlations among variables. Statistical research [33] and [34] sets basic principles for dealing with these problems. We performed stepwise selection of variables to better prevent introducing high correlations in the model.