Statistical Methods for Irregularly Measured Longitudinal Data

点击翻译按钮获取中文摘要

We live in a data-rich society, and a particularly useful form of data is when the same questions or measures are collected repeatedly over time on the same subjects. This allows researchers to quantify trends over time. Often practical constraints lead to variation in the times at which measurements are taken. Indeed, sometimes study organisers may request that measurements be taken more often after an abnormal reading is observed. The end result is that the times at which measurements are taken is associated with the measurements themselves.**In order to capture changes in outcomes over time correctly, analysis must account for the potential for the number and timing of measurements to be related to the outcomes; failure to do so could lead to a seriously distorted picture of the dynamics of the process over time. This work will focus on statistical methods that account for the fact that the timings of measurements, as well as the measurements taken at each visit, give important information about the outcomes of interest.**Particular areas of focus will be improving the validity of methods, improving their efficiency, and improving their accessibility. **All methods for this type of data make assumptions. The current approach to analysis risks violating some of these assumptions because it disregards important information on the visit process when setting up models. We propose an approach to analysis that makes better use of the available information and so reduces the risk of bias in the results.**The most popular method of analysis for this type of data is known to be inefficient. That is, it does not extract as much information from the data as it could. This is wasteful because data are often expensive to obtain. There is a class of methods known as doubly robust methods that are generally more efficient, and we will aim to expand this class of methods to handle longitudinal data measured at irregular times.**Finally, to be useful, methods need to be accessible. That is, they need to be simple enough that scientists doing routine analyses of data can use them. We propose to develop a new approach to analysis that is straightforward to use in practice, and likely also to be more efficient than the approaches that are currently used.**The proposed work will thus strengthen the analysis of longitudinal data measured at irregular times by providing methods that improve validity, efficiency and accessibility.