A Guideline for Handling Longitudinal Data Subject to Irregular Observation

Longitudinal data are an important resource in describing disease course and impact of exposures, yet often feature a stochastic observation process, resulting in irregular sampling times that differ both within and between patients. Furthermore, observation times are often related to the outcome of interest. Irregular observation times in longitudinal data are common, especially when data collection occurs as part of usual care, for example through electronic health records. The increasing availability of this type of data for research makes this a pressing problem. Methods for handling irregular observation times have been developed, yet the issue of irregular observation is widely ignored in practice. This stems in part from a lack of awareness that it is an issue, and also from a lack of guidance on how to handle the issue. We have created a guideline outlining best practices for designing and analysing longitudinal studies which have the potential for irregular observation times. This panel discussion will introduce the guideline and provide an opportunity for stakeholder input. The panel consists of guideline developers as well as experts not involved in its development. The role of the external expert panelists is to provide constructive critique, while panelists who were involved in the development of the guideline will address questions about the rationale and process used to create the guideline. The chair will facilitate a discussion among attendees and panelists. We anticipate that this session will be of interest to those working on methodology for handling irregular observation and to those using longitudinal data in practical applications. It will raise awareness of the issue of irregular observation, point users to tools to address it, and give attendees a voice in outlining best practices.