Does Low Participation in Cohort Studies Induce Bias?

Publikation: Bidrag til tidsskrift/Konferencebidrag i tidsskrift /Bidrag til avisTidsskriftartikelForskningpeer review

  • Afdeling for Biostatistik
  • Afdeling for Epidemiologi
  • Børneafdeling, SKS
Background: Participation rates in large cohort studies have dropped during the last two decades. The consequences of this trend for relative risk estimation are unknown. Methods: The impact of a low participation rate (30%) on the Danish National Birth Cohort was examined among 49,751 women from the source population, including 15,373 participants in the cohort study. Based on independent data collection, we estimated odds ratios (ORs) in the source population and among participants for three exposure-risk associations: a) in vitro fertilization and preterm birth, b) smoking during pregnancy and birth of a small-for-gestational-age infant, and c) prepregnancy body mass index and antepartum stillbirth. The effect of non-participation was described by a relative odds ratio, ROR, calculated as the OR(participants) / OR(source population). Two methods for calculation of confidence intervals for the relative odds ratio were also assessed. Results: The effect of non-participation on the selected odds ratios was small. The relative odds ratios were close to one and the bias was never larger than 16%, although some of the confidence intervals were wide. The two methods for calculation of confidence intervals gave very similar results and a small simulation study showed that the coverage probabilities were close to the 95% nominal level Conclusion: For the three chosen associations the odds ratios were not biased by non-participation. The results are reassuring for studies based on the Danish cohort and similar cohorts of pregnant women. The methodology used to compute confidence intervals for the relative odds ratios performed well in the scenarios considered.
OriginalsprogEngelsk
TidsskriftEpidemiology
Vol/bind17
Nummer4
Sider (fra-til)413-18
Antal sider6
ISSN1044-3983
StatusUdgivet - 2006

Se relationer på Aarhus Universitet Citationsformater

ID: 1401457