We developed a model that performs unsupervised clustering of survival times in a jointsurvival-longitudinal framework with known longitudinal trajectories and
all forms of censoring and truncation. The model allows for data that is observed, left censored, right censored,
interval censored, along with all forms of truncation.
From simulation studies, the model correctly identifies the parameter estimates in any level of censoring quickly.
When clustering is present, we use variations of AIC and BIC for identifying the correct number of clusters.
From simulation studies, we find that BIC correctly identifies
the right number of clusters within multiple levels of censoring
greater than 90% of the time along
with correctly estimating the parameter estimates.
All the analysis is performed in the R package currently being
developed, which performs the analysis relatively quickly.
We applied the model to the Study of Women's Health Across the Nation (SWAN) dataset.
We used this data set for detecting Alzheimer's disease and to
decipher what covariates are linked to an increased risk for developing Alzheimer's disease.