Compute subject-specific transition probabilities using a convolution algorithm based on the Fast Fourier transform.
probtrans_fft(initial_state, cumhaz, max_time, nr_steps = 10000)An object of class 'probtrans'. See the 'value'
section in the help page of mstate::probtrans.
The present function estimates state occupation probabilities from the state given in this argument.
An msfit object created by running
mstate or mstate_generic.
The maximum time for which transition probabilities are estimated.
The number of steps in the convolution algorithm (larger increases precision but makes it slower)
Rui Costa
The time argument is crucial for precision.
The density of time points and
the upper time limit should
be increased until the estimated curves become stable.
A useful rule of thumb is to set the upper time limit
to a time point in which the
probability of each transient state is zero and the probability of
each absorbing state is constant.
For the same approximation grid, probtrans_fft doesn’t
always yield the same result as probtrans_ebmstate
(semi-Markov version), even though they are meant to approximate
exactly the same convolution. probtrans_ebmstate is
sensitive to the grid interval size, but not such much to the
maximum grid time. probtrans_fft is sensitive to both
these parameters, as referred above.
The algorithm behind probtrans_ebmstate is based
on the convolution of density and survival functions and
is suitable for processes with a tree-like transition
structure only.
probtrans; probtrans_ebmstate