|
Abstract:
|
In a clinical trial or natural history study on a disease, longitudinal data on multiple outcomes are typically collected, each of which represents one aspect of the disease process. In order to better understand disease progression or treatment response from such heterogeneous data, it is desirable to partition patient sample into relatively homogeneous subgroups. In general, two approaches exist for clustering of multivariate longitudinal data, one is model-based and the other is data-driven. For the former a popular approach is based on the mixture modeling of multivariate mixed-effects model, whereas for the latter some widely used clustering methods such as k-mean clustering can be performed based on summary measures of multivariate longitudinal data. Each approach has its pros and cons, but no systematically comparison has been made on their performance. Motivated by the need for clustering patient trajectories from a large dataset from a phase III clinical trial, this research will focus on the evaluation of these two clustering approaches through extensive simulation studies. Recommendations are given based on the simulation results.
|
