|
Abstract:
|
In longitudinal health claims data, patients’ visit patterns usually do not follow regular patterns. These patterns may provide information about patients’ health status, and may introduce bias. In diseases with temporal variability in disease status such as multiple sclerosis, this can have an important impact. In this talk, I will describe settings in which informative visit patterns can cause bias in estimation of disease risk, time to progression, and of relative risks comparing two treatments. I will present graphical methods to describe these potential biases, and some statistical tools involving inverse probability weighting to correct for bias.
|
