Article Text

Download PDFPDF

2.2 Using cohorts to study lifecourse epidemiology Chair: Prof Cyrus Cooper, UK Discussant: Prof John Frank, UK
O2-2.1 Maximising the return from cohort studies
  1. A Leyland1,
  2. I White2,
  3. S Harding1,
  4. S Seaman2,
  5. C Booker3
  1. 1MRC|CSO Social and Public Health Sciences Unit, Glasgow, UK
  2. 2MRC Biostatistics Unit, Cambridge, UK
  3. 3Institute of Social and Economic Research, University of Essex, UK


Introduction Cohort studies are important for understanding the aetiology underlying differences in disease incidence. Selective attrition is problematic as those at greater risk of ill-health are more likely to drop out, resulting in homogenous study populations with limited generalisability. Selective attrition may also bias estimates of association. For the benefits of cohort studies to be realised efforts must be made to minimise attrition and statistical methods for the analysis of studies with missing data must be developed to minimise bias.

Methods We describe work examining methods used to maintain participation in cohort studies, identify best practice for reducing attrition and investigate methodologies suitable for the analysis of cohort studies with attrition under different circumstances.

Results A literature search identified factors associated with minimising attrition including study design, recruitment procedures, incentives and retention methods utilised. A questionnaire including such factors was sent to 32 UK-based cohort studies; 25 (78%) returned a questionnaire. Analysis suggested that no one method is most effective, rather it is the combination of methods and study setting which may dictate the overall retention of study participants.

Work with various cohort studies resulted in consideration of the appropriateness of methods for missing data. This has led to a review of inverse probability weighting (IPW) focusing on how and when to use it appropriately. The motivation for combining IPW and multiple imputation, and the theoretical justification for doing so, has been examined. A final development concerns the use of IPW when predictors of missingness are themselves missing.

Statistics from

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.