Competing Risk: 1 Method One. Cox PH Model

Competing risk
Ana Teresa Yanes Musetti, Edgar Alan Muro Jimenez and Abel Camacho Guardian
Statistic Seminar.
23-05-2011
"Competing risk occur when there are at least two possible ways that a person can fail, but
only one such failure type can actually occur."[1]
Objective for competing risks[1]

Asses the relationship of relevant predictors to the failure rate or corresponding survival
probability of any of the possible events allowing for the competing risks of the other way
to fail.
Compare the failure rates for two or more possible events, controlling for relevant predictors.
1 Method one. Cox PH model.
Assumption : Non informative censoring, we need to check the PH assumptions.

For every type of event, we adjust a Cox PH model treating the other events as censored.
Cause specic hazard function

P (t ≤ Tc < t + ∆t|Tc ≥ t)
hc (t) = lim (1)
∆t→0 ∆t
Where Tc = time-to-event of type c. c ∈ {1, ..., C}. In this case we have C dierent types of
events.
Cox PH cause specic model for event type c.

Xp
hc (t, X) = h0c (t)exp( βic Xi ) (2)
i=1
Where c ∈ {1, ..., C} and p is the number of predictors 1
Lunn Mc Neil Approach

Only one Cox PH model for all type of events. It gives identical results as adjusting Cox PH for
each event type, but here we have the exibility to perform statistical inference.
1. βic can be dierent for dierent event types.
1
For C competing risks
h∗g (t, X) = h∗0g (t) exp[β1 X1 + ... + β1 Xp
+δ21 D2 X1 + ... + δ2p D2 Xp
+δ31 D3 X1 + ... + δ3p D3 Xp
+...
+δC1 DC X1 + ... + δCp DC Xp ]
Dc is a dummy variable which is equal to 1 for event type c and 0 otherwise, c ∈ {1, ..., C}.
In this case there is no dummy variable for event type 1, because it is the reference event.
For type of event 2, we get :
h∗2 (t, X) = h∗02 (t) exp[(β1 + δ21 )X1 + ... + (β1 + δ2p )Xp ]
The formula for the HR for the eect of X1 adjusted for the covariates is
HR2 (X1 = 1 vs. X1 = 0) = exp (β1 + δ21 ) (3)
Alternative Lunn-Mcneil Approach.
h∗g (t, X) = h∗0g (t) exp[δ1 X1 + ... + δ1 Xp

0 0
+δ21 D2 X1 + ... + δ2p D2 Xp
0 0
+δ31 D3 X1 + ... + δ3p D3 Xp
+...
0 0
+δC1 DC X1 + ... + δCp DC Xp ]
The dierence between the two Lunn Mc-Neil approaches is that in the alternative, for event
0 0
type one, we have products of the form D1 X1 , D1 X2 , ..., D1 Xp with coecients δ11 , ..., δp1
0
HR2 (X1 = 1 vs. X1 = 0) = exp (δ2 ) (4)
2 Non-informative censoring.
Denition :[7] Non informative censoring is when each subject has a censoring time that is
statistically independent of their failure time.
[5] The probability of being censored for any subject in the risk set at time t does not depend
on that subject's prognosis for failure at time t.
(More general)[5] Each subject is representative of subjects in the risk set with the same values
of predictors.
Example. Our event of interest is time when a death of cancer occurs and our competing risk
is time when a death of cardiovascular occurs. Let Harry and Barry be two individuals in an
homogeneous group, there is non-informative censoring if the probability of being censored is the
same for both individuals, in particular this means that the probability of dying of the competing
risk is equal for both individuals.
It is not possible to determine whether the competing risks are independent.
2
3 Cumulative Incidence Curve.
Mathematical denition. Competing risks as bivariate random variable.
(T, C)
T time at which the event of type i occurred, when C = i
T time at which the observation was censored, when C = 0.
Cumulative incidence function : The Probability that an event of type i occurs at or
before time t.
Fi (t) = P (T ≤ t, C = i)
CIC : estimates the "marginal probability" of an event type i in the presence of competing
risks. It is not required the independence of non-informative censoring.
X
F̂i (t) = hˆij Ŝ(tj−1 ) (5)
∀j:tj ≤t
dij
with hˆij = nj
It is required that
ht = hc1 (t) + hc2 (t) + ... + hck (t) (6)
The above equality is satised if the type of events are mutually exclusive (disjoint) and they
are non recurrent.
Conditional Probability Function.

The probability of experiencing an event c by time t, given that an individual has not expe-
rienced any of the other competing risks by time t.
CP Cc = P (T ≤ T, C = c| no other type of event by t ) (7)
CICc
CP Cc = (8)
1 − CICc0
0
c denotes all type of events, but except c.
[1] Lunn M, McNeil D. Applying Cox regression to competing risks. Biometrics 1995 ; 51 :524532
[2] Pintilie M. Competing Risks : A Practical Perspective. John Wiley and Sons : New York,
2006, 240pp.
[3] J.Klein and M. Moeschberg, Survival Analysis : Techniques for Censored and Truncated
Data, Spring, second ed., 2003.
[4] G. Rodriguez, Competing Risks, 2005.
[5] Survival analysis : A self-learning text, Kleinbaum D.G. and Klein M., Springer 2005.
[7] http ://en.wikipedia.org/wiki/Censoring_(statistics)

Competing Risk: 1 Method One. Cox PH Model

Uploaded by

Copyright:

Available Formats

Competing Risk: 1 Method One. Cox PH Model

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Competing Risk: 1 Method One. Cox PH Model

Uploaded by

Copyright:

Available Formats

Competing risk

Objective for competing risks[1]

1 Method one. Cox PH model.

Assumption : Non informative censoring, we need to check the PH assumptions.

Cause specic hazard function

Cox PH cause specic model for event type c.

Where c ∈ {1, ..., C} and p is the number of predictors 1

Lunn Mc Neil Approach

1. βic can be dierent for dierent event types.

Alternative Lunn-Mcneil Approach.

h∗g (t, X) = h∗0g (t) exp[δ1 X1 + ... + δ1 Xp

Mathematical denition. Competing risks as bivariate random variable.

Conditional Probability Function.

CP Cc = P (T ≤ T, C = c| no other type of event by t ) (7)

You might also like

Competing Risk: 1 Method One. Cox PH Model

Uploaded by

Copyright:

Available Formats

Competing Risk: 1 Method One. Cox PH Model

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Competing Risk: 1 Method One. Cox PH Model

Uploaded by

Copyright:

Available Formats

Competing risk

Objective for competing risks[1]

1 Method one. Cox PH model.

Assumption : Non informative censoring, we need to check the PH assumptions.

Cause specic hazard function

Cox PH cause specic model for event type c.

Where c ∈ {1, ..., C} and p is the number of predictors 1

Lunn Mc Neil Approach

1. βic can be dierent for dierent event types.

Alternative Lunn-Mcneil Approach.

h∗g (t, X) = h∗0g (t) exp[δ1 X1 + ... + δ1 Xp

Mathematical denition. Competing risks as bivariate random variable.

Conditional Probability Function.

CP Cc = P (T ≤ T, C = c| no other type of event by t ) (7)

You might also like

Cause specic hazard function

Cox PH cause specic model for event type c.

1. βic can be dierent for dierent event types.

Mathematical denition. Competing risks as bivariate random variable.