Rounding up the usual suspects: Confirmation bias in epidemiological research
summaryBiases may arise not only from which data are collected, but also from commonalities in the variables embedded in available data sources, which lead to confirmation bias. The possibility that spurious relationships are being perpetuated should be considered when a relationship is reproduced in multiple datasets with substantially overlapping variables that do not include all plausibly confounding variables. When findings are reproduced, it is important to ask if the datasets have similar constellations of measured and unmeasured variables or related concepts, because the reproducibility may be an artefact of biased availability of variables. Future work is warranted to study whether biased availability of variables underlies the phenomenon of multiple observational studies with consistent causal inferences being refuted by subsequent randomized trials.