Precise definition of the disease(s) to be identified, as the cohort is followed in time, must be made and one must be as sure as possible that those who apparently do not have the disease are truly free of the disease in question. If the definition is not precise, there is scope for misclassification and this may dilute the differences between the exposed and non-exposed cohorts and thereby diminish the estimate of risk.
Cohort studies are subject to a number of biases including those caused by treatment selection and differential follow-up. An example of treatment selection bias is if the purpose of a cohort study is to estimate the rate of cardiovascular disease in men sterilised by vasectomy, it is necessary to have a comparison group of non-vasectomised men. However, as we have noted with 'ever' users of OCs, comparisons between such groups may be biased as it is not feasible to randomise men to 'sterilisation' or 'no-sterilisation' groups. It is clear that men who are seeking sterilisation would certainly not accept the 'no-sterilisation' option. As a consequence, the final comparison is made between those men who opt for sterilisation against those who do not. This may introduce inherent biases as, for example, the vasectomised men may be fitter than the non-vasectomised men and this may influence their cardiovascular disease rates.
Follow-up bias can arise when follow-up is poor, or when it is more complete for one group than for the other. Thus it is possible that, for example, subjects exposed to radiation are more likely to be carefully monitored than subjects not exposed.
Was this article helpful?