Abstract:
|
Data collected longitudinally as part of usual health care is becoming increasingly available for research. However, if the frequency of follow-up is determined by the patient's health, failure to account for the observation process can result in biased inferences. For example, in a study of growth among newborns where data is collected from patient charts, infants who are slow to recover their birthweight will be monitored more closely, leading to under-estimation of the mean rate of growth if the observation process is ignored. This talk will discuss the circumstances under which the observation process can create bias, together with both design and analytic strategies to mitigate this bias.
|