Does attrition affect estimates of association: A longitudinal study

Area of Science:

• Epidemiology
• Biostatistics
• Longitudinal Studies

Background:

• Survey research, particularly longitudinal studies, often experiences participant attrition due to refusals, inaccessibility, or loss to follow-up.
• Concerns exist that participants lost to follow-up may represent a select group, potentially biasing study findings.

Purpose of the Study:

• To investigate the extent to which selective loss to follow-up in a longitudinal study influences research findings.
• To compare measures of association at baseline with those derived from retained participants versus those lost to follow-up.

Main Methods:

• Utilized data from the Mater-University Study of Pregnancy, including 7718 women at baseline and follow-ups over 27 years.
• Compared baseline measures of association with those from retained (Group A) and lost-to-follow-up (Group B) participants at each follow-up phase.
• Employed univariate logistic regression to assess associations between maternal mental health, socio-demographic factors, and loss to follow-up rates.

Main Results:

• Estimates of association remained consistent across follow-ups, regardless of attrition rates or comparison group (retained vs. lost).
• No statistically significant differences were observed in 90.8% of baseline comparisons with Group A and 96.9% with Group B.
• Differential loss to follow-up demonstrated a minimal impact on the estimates of association in this study.

Conclusions:

• Differential loss to follow-up in longitudinal studies infrequently impacts the validity of association estimates.
• Misleading findings due to attrition are likely only when loss to follow-up is compounded by other sources of bias.