lity thattheobservedassociationwasduetounmeasuredor
residual confoundingshouldbeconsidered,andinadequatecon-
trol forconfoundersmightbiastheresultsineitherdirection,
towardexaggerationorunderestimationofriskestimates.Second,
severalstudieswithfewdeathnumbersoflivercancerwere
included, whichmightincreasethesamplingerror