Add the capability to fit multi-state models in PROC PHREG.
H. Putter, M. Fiocco1 and R. B. Geskus (2007) "Competing risks and multi-state models" Statistics in Medicine; 26:2389–2430
"The class of multi-state models forms an extension to that of competing risks models. Competing risks models deal with one initial state and several mutually exclusive absorbing states. Typically, the disease or recovery process of a patient will also consist of intermediate events that can neither be classified as initial states nor as final states. This type of models is called multi-state models. In many cancer studies, after surgery of the primary tumour, the tumour may recur in the vicinity of the primary tumour (local recurrence), or at distant locations (distant metastasis). These events may occur in any order (although local recurrence usually precedes distant metastasis) and patients may die before or after experiencing local recurrence or distant metastasis.
Markov, semi-Markov and extended Markov models.
A property that is often assumed in practice is that the multi-state model is a Markov model. Loosely speaking, the Markov property states that the future depends on the history only through the present. For a multi-state model this means that, given the present state and the event history of a patient, the next state to be visited and the time at which this will occur will only depend on the present state. Strictly speaking, only ‘clock forward’ models can be Markov models; for ‘clock reset’ models the Markov property cannot hold since the time scale itself depends on the history through the time since the current state was reached. However, if it is assumed that the sojourn times depend on the history of the process only through the present state and the time since entry of that state, the resulting multi-state model forms a sequence of embedded Markov models, called a Markov renewal model (see e.g. References [45–48]), or also a semi-Markov model. Note that competing risks models are always Markovian, since there is no event history. "
Thanks to Douglas Gregory, Ph.D. of Cardiovascular Clinical Studies for this suggestion and the above information.