Related Experiment Videos
Does low participation in cohort studies induce bias?
Ellen Aagaard Nohr1, Morten Frydenberg, Tine Brink Henriksen
1The Danish Epidemiology Science Centre, Institute of Public Health, University of Aarhus, Aarhus C, Denmark. ean@soci.au.dk
Epidemiology (Cambridge, Mass.)
|June 7, 2006
Summary
Declining participation in large cohort studies may not significantly bias risk estimates. This study found minimal bias in three key exposure-risk associations, suggesting results from similar cohorts are reliable.
Area of Science:
- Epidemiology
- Biostatistics
- Reproductive Health
Background:
- Participation rates in large cohort studies have declined over the past two decades.
- The impact of this trend on relative risk estimation remains unclear.
Purpose of the Study:
- To assess the impact of low participation (30%) on risk estimates in the Danish National Birth Cohort.
- To evaluate the bias in relative risk estimation due to nonparticipation.
Main Methods:
- Compared odds ratios (ORs) in the source population versus participants for three exposure-risk associations.
- Calculated relative odds ratios (RORs) to quantify the effect of nonparticipation.
- Assessed two methods for calculating confidence intervals for RORs.
Main Results:
- Nonparticipation had a small effect on the selected ORs, with RORs close to one.
- Bias due to nonparticipation was never larger than 16%.
- Two methods for confidence interval calculation yielded similar results, with good coverage probabilities.
Conclusions:
- The odds ratios for the three studied associations were not biased by nonparticipation.
- Findings are reassuring for the Danish National Birth Cohort and similar studies on pregnant women.
- The methodology for computing confidence intervals for RORs performed well.