BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:BSU Seminar: &quot\;Bayesian latent multi-state modelling for long
 itudinal health trajectories&quot\; - Yu Luo\, Kings College London
DTSTART:20241015T130000Z
DTEND:20241015T140000Z
UID:TALK220204@talks.cam.ac.uk
CONTACT:Alison Quenault
DESCRIPTION:In medical research\, understanding changes in outcome measure
 ments is crucial for inferring shifts in a patient's underlying condition.
  While data from clinical and administrative systems hold promise for adva
 ncing this understanding\, traditional methods for modelling disease progr
 ession struggle with analyzing a huge volume of longitudinal data collecte
 d irregularly and do not account for the phenomenon where the poorer an in
 dividual's health\, the more frequently they interact with the healthcare 
 system. In addition\, data from the claim and health care system provide n
 o information for terminating event\, such as death. To address these chal
 lenges\, we develop a Bayesian approach for the continuous-time hidden Mar
 kov model (CTHMM) to understand disease progression by modelling the obser
 ved data as an outcome whose distribution depends on the state of a latent
  Markov chain representing the underlying health state. Subsequently\, we 
 extend the model and allow the underlying health state to influence the ti
 mings of the observations via a point process and the unobserved death as 
 an informative censoring whose rate depends on the latent state of the Mar
 kov chain. This extension allows us to model disease severity and death no
 t only based on the types of care received but also on the temporal and fr
 equency aspects of different observed events.  In addition\, we also consi
 der the death as missing data\, and model it as an informative censoring. 
 We present an exact Gibbs sampler procedure that integrates the observatio
 n process by modelling observed data as an outcome process and the timing 
 of observations as a point process\, jointly with the CTHMM\, given the un
 derlying latent health state. We facilitate our inference by recovering th
 e trajectories from the final observed record to the end of the observatio
 n period\, thereby mitigating the bias caused by the lack of information f
 rom early death patients. Finally\, we apply our method to health care cla
 im data from a Canadian cohort.
LOCATION:MRC Biostatistics Unit\, East Forvie Building\, Forvie Site Robin
 son Way Cambridge CB2 0SR.
END:VEVENT
END:VCALENDAR