Bias due to incomplete follow up in a cohort study.

AIM To investigate the bias introduced by incomplete follow up in a cohort study of ocular outcome after premature birth. METHODS A geographically defined cohort of children born before 32 weeks' gestation was prospectively recruited at birth to study the ocular outcome at 2 years. On the basis of attendance at 2 years, the children's families were allocated to one of three groups: group 1 attended for follow up, group 2 were difficult to trace, and group 3 were very reluctant for assessment. All children were examined by a single ophthalmologist, masked to these groupings. RESULTS 558 children (98.8% of study group) were examined, of whom 505 were in group 1, 20 in group 2, and 33 in group 3. The groups which were more difficult to study (groups 2 and 3) showed a significantly higher prevalence of ocular abnormalities, including strabismus (p=0. 02) and cicatricial retinopathy of prematurity (p=0.002) compared with those attending for follow up. Further, not all of these cases could have been identified by review of the children's previous records. Ocular abnormalities would be underestimated by 16% (11.3% in group 1 compared with 13.4% in the total cohort, p=0.77). CONCLUSIONS This study suggests that the prevalence of abnormalities would be underestimated by incomplete follow up, as those subjects who were most difficult to obtain for study had a significantly higher prevalence of abnormalities.