Multi-state hazard models (MSH) have been a key part of the R survival package, and play a significant role in our group's medical research work. This is a library extends these methods to interval censored data. One subset or extension of interval censored MSH are hidden Markov models (HMM), this the root of our combined acronym for the package name: MSH + HMM = mshmm.
Three things have spurred the development of the package. The primary one has been the Mayo Clinic study of aging (MCSA), and the work therin to understand the patterns and components of cognitive decline, including the impacts thereon of amyloid and tau depostion (an important driver of Alzheimer's disease, a dementia subtype), of accumulated cerbral vascular compromise (itself multifactorial), neuron and neural function loss,clinical risk factors such as age, sex, hypertension, etc., and measurement of the cognitive decline itself. This dictates many of types of model we want to fit, in particular that direct measurement of many of these underlying processes is only available during postmortem pathologic examination of the brain, so we must rely on indirect ones. (Unlike a potential breast or prostate cancer, say, a needle biospy of the living brain is not a diagnostic option.) A second is that our more complex models have required more sophisticated numerical methods, e.g., Levenberg-Marquardt and trust region additions to the iteration, imposition of constraints, and for some, an MCMC approach. The optim routine underlying the msm package was simply not up to the task. The third was to make these often quite complex models easier for the user to specify by extending user interface ideas worked out for the semi-parametric MSH methods found in the survival package, and in the process make the similarities of the two MSH approaches clear.
This current version is intended for sharing within a small group of developers, until we have worked all the test cases.