Data dredging, bias, or confounding
Authors
Discussion
Loading...
Member takes
Implication
Methodological review demonstrates frequent discrepancies across observational epidemiology in general populations, highlighting distortion from data dredging and confounding.
Key Points
- To examine why observational epidemiological studies frequently produce contradictory health claims and assess how data dredging, bias, and confounding undermine reproducibility.
- Critically analyzed high-profile contradictions in observational epidemiological publications and media reporting concerning breast cancer, alcohol, and tobacco.
- Evaluated discordance between observational associations and subsequent evidence from randomized controlled trials, focusing on cases like hormone replacement therapy.
- Observational studies reported mutually contradictory conclusions within weeks regarding the association of moderate alcohol intake and early smoking with breast cancer risk.
- Randomized controlled trials failed to confirm seemingly robust observational findings, including a 1991 meta-analysis reporting that hormone replacement therapy halved coronary heart disease risk (relative risk 0.50).
Cite This Study
George Davey Smith (2002) studied this question.