A great deal of modern health research aims to exploit data from disease registries or administrative health records in order to supplement or replace the more costly collection of information from prospective cohort studies. Understanding sampling mechanisms and the factors governing the nature and duration of follow-up are critical for such data sources, however, to ensure valid inference and interpretable findings. This talk will discuss statistical challenges from, and approaches for dealing with, dependent delayed-entry, dependent intermittent observation schemes, and dependent loss to follow-up. Multistate models will be used to represent the sampling and observation processes, communicate the assumptions for standard analyses, and provide a framework for joint modelling which enables one to mitigate biases through simultaneous model fitting or two-stage procedures using inverse probability weighting.