Biomedical Engineering Reference
In-Depth Information
of failure. However, data are often available not only on the status of an
individual{that is, whether they have failed or not at the time of observation|
but also on the cause of failure. A classic example in the clinical setting is that
of a woman's age at menopause, where the outcome of interest is whether
menopause has occurred, U is the age of the woman, and the two competing
causes for menopause are either natural or operative.
More generally, consider a system with K (finite) components that will
fail as soon as one of its component fails. Let T be time to failure, Y be
the index of the component that fails, and U be the (random) observation
time. Thus, (T;Y ) has a joint distribution that is completely specied by
the sub-distribution functions fF
0i
g
i=1
, where F
0i
(t) = P(T t;Y = i).
The distribution function of T, say F
+
, is simply
P
i=1
F
0i
and the survival
function of T is S(t) = 1 F
+
(t). Apart from U, we observe a vector of
indicators = (
1
;
2
;:::;
K+1
), where
i
= 1fT U;Y = ig for i =
1; 2;:::;K and
K+1
= 1
P
j=1
j
= 1fT > Ug. A natural goal is to
estimate the sub-distribution functions, as well as F
+
. Competing risks in
the more general setting of interval-censored data was considered by Hudgens
et al. (2001) and the more specific case of current status data was investigated
by Jewell et al. (2003). In what follows, I restrict to two competing causes
(K = 2) for simplicity of notation and understanding; everything extends
readily to more general (finite) K but the case of infinitely many competing
risks, the so-called \continuous marks model," is dramatically dierent and I
touch upon it later.
Under the assumption that U is independent of (T;Y ), the likelihood func-
tion for the data that comprise n i.i.d. observations f
j
;U
j
g
j=1
, in terms of
generic sub-distribution functions F
1
;F
2
is
Y
F
1
(U
i
)
j
1
;F
2
(U
i
)
j
2
S(U
i
)
j
3
;
L
n
(F
1
;F
2
) =
j=1
this follows easily from the observation that the conditional distribution of
given U is multinomial. Maximization of the above likelihood function is
Search WWH ::
Custom Search