CSV, ODBC and other packages built on top of DataStreams currently use Nullable to handle missing values. This is different from the NA/DataArray approach that DataFrames uses by default. But it is still a DataFrame.
All this is unfortunate but it is being actively worked on. By Julia 0.7 all the packages should use a new approach that will be easier to work with - see this announcement.
In the meantime you can use DataTables that works more naturally with the Nullable data type or convert things manually.