Using routinely collected data from electronic health records offers a low-cost approach to investigating disease progression over multiple years of follow-up. However, this study design can lead to a biased sample since patients interact with the healthcare system more often when they are unwell and thus there is an overrepresentation of measurements on sicker patients. We show how the rigour of modelling of the irregularly observed disease trajectory over time can be enhanced by leveraging information that has never been used before but is often already recorded in patient charts: physician recommendations on when the next visit should occur. Specifically, we demonstrate how recommended intervals can be used in examining and classifying the irregular visit process, and in assessing the sensitivity of the results to visiting not at random (VNAR). We illustrate our approach using data from a clinic-based cohort of patients with juvenile dermatomyositis at the Hospital for Sick Children.