Additionally, small samples are more likely to be
influenced by extreme values.
However, we performed a sensitivity
analysis deleting those values detected in the model diagnostics,
and no important changes in the coefficients were observed.
Due
to the study design, we cannot exclude a potential reversed-causality
in the associations found, due to a potential increase in vegetables
and cereal consumption in relation to meat after cancer
diagnosis.
Although we adjusted the model for all the confounders
identified, we cannot exclude the influence of third variables not
taken into account.
Furthermore, controls were not subjected to
biopsy and therefore, there is a possibility of some of them being
in a pre-clinical stage.
However, these would be more likely
decrease the magnitude of the associations found.