Skip to Main Content

Evaluating Bias in Biomedical Research

December 01, 2015
by Denise Meyer

Michael Bracken, ’70 MPH, ’74 PhD, the Susan Dwight Bliss Professor of Epidemiology (Chronic Diseases) and professor of neurology and of obstetrics, gynecology, and reproductive sciences, has led generations of Yale School of Public Health Students toward a greater understanding of how to craft truly objective epidemiological studies.

In his 2013 book, Risk, Chance and Causation – Investigating the Origins and Treatment of Disease, published by Yale University Press, he distilled these concepts for the general public.

In a discussion about the book for the Association of Yale Alumni Assembly on November 19, Bracken described the gold standard for epidemiological studies — double-blind, randomized control trials — where neither the patient nor the clinician knows who is receiving a treatment and who is receiving a placebo. However designing such studies can be tricky and bias can creep in at many levels. “We even mask the statisticians in our trials,” said Bracken, who is also founder and co-director of the Yale Center for Perinatal, Pediatric and Environmental Epidemiology.

In many studies, small sample size may create a bias because it is not large enough to be representative. How participants are recruited may skew the results, as can the methods by which they are randomized. In large trials, algorithms for balancing groups are needed to offset the chance of having one group have too many people with one characteristic, such as older age. Sometimes, doctors figure out or guess who is receiving the placebo and feel compelled to offer compensating treatments. In studies of harm, observed associations are rarely causal and causal connections must be attributed very cautiously. In other studies, researchers do not screen deeply enough or measure conditions that can confound the disease associations.

The counter measure to these and other types of bias is replication and meta-analysis, said Bracken. You need to see if the results can be shown in several studies, and then evaluate the quality of a group of related studies and compare their findings. Meta-analyses are much more newsworthy than the reports of individual studies, he said. Overall, Bracken only has confidence in about 60 percent of trials and ten percent of observational studies.

Professor Bracken has served as chair of the Department of Chronic Disease Epidemiology, director of Graduate Studies and as past president of the American College of Epidemiology and the Society for Epidemiological Research.

Submitted by Denise Meyer on December 01, 2015