Telephone versus postal surveys of general practitioners: methodological considerations.