Discretised values of monocyte counts, and ratings of level of psychosis simulated from a model fitted to a data set consisting of observations made on a number of patients from the Northern District Health Board system. The real data must be kept confidential due to ethics constraints.
data("ccprSim")The object ccprSim is a list of length 1258. Each entry
of this list is to be considered to correspond to an individual
subject. The entries consist of matrices having two columns
named cellCount and psychosisRating. The number
of rows of these matrices varies from entry to entry of the list
(i.e. from subject to subject).
Most of the entries of these matrices are NA. The entries
are temporally ordered and correspond to the number of weeks
from the start of observation. Observations in the real
data set were made only when the patient in question visted a
physician and so weeks in which no visit was made resulted in an
“observation” of NA. The object ccprSim
was simulated in such a way as to imitate this characteristic.
The fraction of missing observations in each variate (i.e.
cellCount and psychosisRating is roughly
commensurate with the corresponding fractions in the real data.
The values in the first column of each matrix (the
cellCount column consist of integers from 1 to 5 and
are to be interpreted as indicators of cell counts in units of
\(10^9\) cells per litre, discretised according to the
following scale:
\(0.0 \leq c \leq 0.3\) \(\leftrightarrow\) 1
\(0.3 < c \leq 0.5\) \(\leftrightarrow\) 2
\(0.5 < c \leq 0.7\) \(\leftrightarrow\) 3
\(0.7 < c \leq 1.0\) \(\leftrightarrow\) 4
\(1.0 < c \leq 2.0\) \(\leftrightarrow\) 5
where \(c\) represents “count”.
The values in the second column of each matrix (the
psychosisRating column consist of integers from 0 to
4 and are to be interpreted as indicators of a physician's
assessment of the level of pschosis of the patient. A value of
0 corresponds to “no symptoms”; a value of 4 corresponds
to “severe”.
The question of essential interest in respect of the real data was “Is there any association between the cell count values and the psychosis ratings?” More specifically it was “Can the level of psychosis be predicted from the cell counts?”
# NOT RUN {
# Takes a long time:
fit <- hmm(ccprSim,K=2,indep=FALSE,itmax=500,verbose=TRUE)
# Throws an error; inadequate data:
subsetFit <- hmm(ccprsim[1021:1025],K=2,indep=FALSE,itmax=5)
# }
Run the code above in your browser using DataLab