Scribd
Upload
62 views26 pages

Evaluating The Proportional Hazard (PH) Assumption: Jerry D.T. Purnomo, PH.D

1) Graphical techniques and goodness-of-fit tests can be used to evaluate the proportional hazards assumption in Cox regression models. Graphs such as log-log survival curves and residuals can check for parallelism over time. 2) Cox-Snell and martingale residuals were examined for a Cox model analyzing ovarian cancer data, showing reasonable fits. 3) The proportional hazards assumption was also formally tested for this model, and residuals were found to satisfy the assumption of proportional hazards over time.

Uploaded by

Nesia
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
62 views26 pages

Evaluating The Proportional Hazard (PH) Assumption: Jerry D.T. Purnomo, PH.D

1) Graphical techniques and goodness-of-fit tests can be used to evaluate the proportional hazards assumption in Cox regression models. Graphs such as log-log survival curves and residuals can check for parallelism over time. 2) Cox-Snell and martingale residuals were examined for a Cox model analyzing ovarian cancer data, showing reasonable fits. 3) The proportional hazards assumption was also formally tested for this model, and residuals were found to satisfy the assumption of proportional hazards over time.

Uploaded by

Nesia
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 26

Evaluating the Proportional

Hazard (PH) Assumption


Week 10

Jerry D.T. Purnomo, Ph.D.


2

Outline
 Graphical techniques
 Goodness of Fit
3

Graphical Techniques (1/2)


 The most popular of these involves comparing
estimated –ln(–ln) survival curves over different
(combinations of) categories of variables being
investigated.
 –ln(–ln) S curves parallel?
4

Problem (1/3)
 Problems with log–log survival curve approach: How
parallel is parallel?
 Recommend:
 subjective decision
 conservative strategy: assume PH is OK unless
strong evidence of nonparallelism
 many categories data “thins out”
 different categorizations may give different graphical
pictures
 small # of categories (2 or 3)
 meaningful choice
 reasonable balance
5

Problem (2/3)
 How to evaluate several variables simultaneously?
 Strategy:
 categorize variables separately
 form combinations of categories
 compare log–log curves on same graph
 Drawback:
 data “thins out”
 difficult to identify variables responsible for
nonparallelism
 Alternative Strategy: Adjust for predictors already
satisfying PH assumption, i.e., use adjusted log–log
survival curves
6

Problem (3/3)
 Alternative Strategy: Adjust for predictors
already satisfying PH assumption, i.e., use
adjusted log–log survival curves.
7

Data Example
> library(survival)
> data(ovarian) → ovarian cancer
The data consists of the following variables:
 futime: survival time (in days) after diagnosis of the
cancer
 fustat: 0 = censored, 1 = dead.
 age: age in years
 residual.dz: a measure of health condition after
chemotherapy.
 rx: 1 = treatment A, 2 = treatment B
 ecog.ps: measure of functioning of the ovaries
8

Cox PH Model (1/2)


 Hypothesis for overall test.
H 0 : β1 = β 2 = β 3 = β 4 = 0
H1: at least one of these β’s are nonzero
 We derive from coxph output (LRT) that p-
value=0.001896 lower than α=0.05 (the three of
statistics tests give the same result). It means that
H0 is rejected, or at least one of these β’s are
nonzero.
9

Cox PH Model (2/2)


10

-ln(–ln) Survival Curves (1/2)


 log #1:
 
ln S t , X   exp
p
i 1

 i X i  ln S0 t 
0  S t , X   1
 log #2:

ln ln S t , X  i 1 i X i  ln ln S0 t 


p

or
 ln ln S t , X  i 1 i X i  ln ln S0 t 
p
11

–ln(–ln) Survival Curves (2/2)

Figure 1. KM survival curve vs -ln(-ln) survival curve


12

Log Rank Test


Hypothesis:
H0: S1(t) = S2(t) for all t
H1: S1(t) ≠ S2(t) for at least one t
13

Graphical Techniques (2/2)


 Other graphical techniques that also commonly used
are Cox-Snell residual, and martingale residual.
14

Cox-Snell Residual (1/3)


 The ith Cox-Snell residual is defined as
 
rCi  Hˆ 0 (ti )  exp xTi βˆ
where Hˆ 0 (ti ) and β̂ are the MLE’s of the baseline
cumulative hazard function and coefficient vector,
respectively.
 The Cox-Snell residuals are most useful for
examining the overall fit of a model.
 This plot is generally used only as a rough
diagnostic.
15

Cox-Snell Residual (2/3)

Figure 2. Cox-Snell residual of ovarian data


16

Cox-Snell Residual (3/3)


 The final model gives a reasonable fit to the data.
 Overall the residuals fall on a straight line with an
intercept zero and a slope one.
 Further, there are no large departures from the
straight line and no large variation at the right-hand
tail.
17

Martingale Residual (1/3)


 The ith martingale residual is defined as
Mˆ i   i  rCi
The M̂ i take value in (,1] and are always
negative for censored observations.
 Used to check the linearity assumption of the
covariate.
 It is common practice in many medical studies to
discretize continuous covariate. The martingale
residual are useful for determining possible cut
points for such variables.
18

Martingale Residual (2/3)

Figure 3. Martingale residual of ovarian data


19

Martingale Residual (3/3)


 In the plot of the Martingale residuals, Figure 3,
there appears to be just little bit bump for age,
between 52 and 65. Moreover, the lines before
and after the bump nearly coincide. Therefore, a
linear form seems appropriate for age.
20

The GOF Testing Approach (1/2)


 Statistical test appealing
 Provides p-value
 More objective decision than when using graphical
approach
 Test of Harrel and Lee (1986)
 Variation of test of Schoenfeld
 Uses Schoenfeld residuals
 Schoenfeld residuals defined for
 Each predictor in model
 Every subject who has event
21

The GOF Testing Approach (2/2)


 p-value large ) PH satisfied
(e.g. P > 0.10)
 p-value small ) PH not satisfied
(e.g. P < 0.05)
22

GOF (Test PH)


23

Schoenfeld Residuals

Figure 4. Schoenfeld residual of ovarian data


24

Backward Method
25

Cox PH (Best Model)


26

Test PH (Best Model)

Figure 5. Schoenfeld residual of best model

You might also like