RT Journal
A1 Livingston EH, McNutt RA
T1 CRoss-level bias and variations in care—reply
JF JAMA
JO JAMA
YR 2011
FD November 16
VO 306
IS 19
SP 2096
OP 2097
DO 10.1001/jama.2011.1663
UL http://dx.doi.org/10.1001/jama.2011.1663
AB 
In Reply: Dr Finney and colleagues raise an important point regarding cross-level bias. All studies based on administrative data or clinical registries will be at risk for drawing invalid conclusions because of unmeasured confounding and misspecification errors. These data sources have limited amounts of information, and the codes used to identify diseases and procedures lack clear definition or specificity. When attempting to assess individuals aggregated into groups, an additional level of cross-level bias results from covariate heterogeneity across groups and effect modification.