Latent Class Models in Medicine
Clustering based on patient-centred information is commonplace in studies of medicine. The ability to carry this out has wide-reaching implications, including the identification of meaningful clinical phenotypes or at-risk patients and a basis for reasoning observed differences in treatment outcome. Nevertheless, clustering is often complicated by the complexity and variability observed between subjects. In many cases, differences between subjects are thought to be attributed to an underlying unobservable process. To this end, the use of statistical models that include an unobservable or latent variable may assist in discovering and learning more about such underlying processes.
The use of latent variables for this purpose has found an abundance of applications in medicine. Latent class models to identify subgroups of patients with symptom profiles have been used in Alzheimer research (Walsh 2006); migraine symptom groupings (Chen et al. 2009); for trajectory response patterns to identify responders and non-responders in clinical trials for interstitial cystitis (Leiby et al. 2009); and identification of trajectories for positive affect and negative events following myocardial infarction (Elliott et al. 2005).
In this chapter, we showcase the flexibility of latent class models ...