9. Background
Methods
Results
Summary
Rationale
Objective
Previous Work
Why is this Gap a Problem?
Overdiagnosis
1 Screen detected cancer that would have otherwise not been
diagnosed before death from other causes
2 Can lead to overtreatment
Overtreatment
Treatment for a cancer that is clinically not significant → low
tumour grade
3 / 35
10. Background
Methods
Results
Summary
Rationale
Objective
Previous Work
Why is this Gap a Problem?
Overdiagnosis
1 Screen detected cancer that would have otherwise not been
diagnosed before death from other causes
2 Can lead to overtreatment
Overtreatment
Treatment for a cancer that is clinically not significant → low
tumour grade
Morbidity associated with being diagnosed with cancer
3 / 35
11. Background
Methods
Results
Summary
Rationale
Objective
Previous Work
Why is this Gap a Problem?
Overdiagnosis
1 Screen detected cancer that would have otherwise not been
diagnosed before death from other causes
2 Can lead to overtreatment
Overtreatment
Treatment for a cancer that is clinically not significant → low
tumour grade
Morbidity associated with being diagnosed with cancer
Quality of life effects, incontinence, erectile dysfunction
3 / 35
12. Background
Methods
Results
Summary
Rationale
Objective
Previous Work
Why is this Gap a Problem?
Overdiagnosis
1 Screen detected cancer that would have otherwise not been
diagnosed before death from other causes
2 Can lead to overtreatment
Overtreatment
Treatment for a cancer that is clinically not significant → low
tumour grade
Morbidity associated with being diagnosed with cancer
Quality of life effects, incontinence, erectile dysfunction
Lower life expectancy
3 / 35
16. Background
Methods
Results
Summary
Rationale
Objective
Previous Work
Solution to Overdiagnosis and Overtreatment
Individualised Decision Making
Active surveillance for patients with favorable risk disease
(Gleason score ≤ 6)
Clinical tools that calculate life expectancy based on
predisposing factors such as age and comorbidities
4 / 35
20. Background
Methods
Results
Summary
Rationale
Objective
Previous Work
Wykes (2011)
Wykes (2011) proposed an ad hoc approach to estimating the
relative risk and baseline hazard function for the case cohort design:
Modified PH Model
S(t) = (S0(t))α
exp {αβ1 (age − µage) +
β2 CIRS-Gpros − µCIRS-Gpros
S0(t), β1, µage are derived from the full cohort
β2, µCIRS-Gpros are from case cohort
α = β1 from case cohort
β1 from full cohort
6 / 35
21. Case Cohort Study Design
40 45 50 55 60 65 70
(a) Cohort, no exposure info
40 45 50 55 60 65 70
(b) Case cohort sampling
40 45 50 55 60 65 70
(c) Add back failures
CIRS=10
CIRS=2
CIRS=7
CIRS=6
CIRS=14
CIRS=13
CIRS=1
CIRS=9
CIRS=9
CIRS=4
CIRS=4
CIRS=10
CIRS=1
CIRS=3
CIRS=6
CIRS=7
40 45 50 55 60 65 70
(d) Get exposure information
Figure: Each line represents how long a subject is at risk. A failure is represented by a
red dot. Subjects without a red dot are censored individuals. The axis represents age
23. Background
Methods
Results
Summary
Case Cohort Study Design
Population
Model
Software
Study Population
Population-based case cohort study of 2,740 patients
diagnosed with prostate cancer between 1990 and 1998 in
Ontario
8 / 35
24. Background
Methods
Results
Summary
Case Cohort Study Design
Population
Model
Software
Study Population
Population-based case cohort study of 2,740 patients
diagnosed with prostate cancer between 1990 and 1998 in
Ontario
Stratified by CCOR random sample
8 / 35
25. Background
Methods
Results
Summary
Case Cohort Study Design
Population
Model
Software
Study Population
Population-based case cohort study of 2,740 patients
diagnosed with prostate cancer between 1990 and 1998 in
Ontario
Stratified by CCOR random sample
Cases were sampled independently of subcohort
8 / 35
26. Background
Methods
Results
Summary
Case Cohort Study Design
Population
Model
Software
Study Population
Population-based case cohort study of 2,740 patients
diagnosed with prostate cancer between 1990 and 1998 in
Ontario
Stratified by CCOR random sample
Cases were sampled independently of subcohort
Death from prostate cancer and death from other causes
assessed → death clearance date December 31, 1999
8 / 35
27. Background
Methods
Results
Summary
Case Cohort Study Design
Population
Model
Software
Study Population
Population-based case cohort study of 2,740 patients
diagnosed with prostate cancer between 1990 and 1998 in
Ontario
Stratified by CCOR random sample
Cases were sampled independently of subcohort
Death from prostate cancer and death from other causes
assessed → death clearance date December 31, 1999
Death from any cause updated for subcohort only → death
clearance date December 31, 2008
8 / 35
29. Background
Methods
Results
Summary
Case Cohort Study Design
Population
Model
Software
Improvements
Stratified by CCOR Cox PH
Sj (t) = S0j (t)exp(β1·Age+β2·CIRS-Gpros )
, j = 1, . . . , 8
We improve on Wykes (2011) work in two ways:
9 / 35
30. Background
Methods
Results
Summary
Case Cohort Study Design
Population
Model
Software
Improvements
Stratified by CCOR Cox PH
Sj (t) = S0j (t)exp(β1·Age+β2·CIRS-Gpros )
, j = 1, . . . , 8
We improve on Wykes (2011) work in two ways:
we estimate both the baseline survival function and relative
risk parameters using the case cohort population
9 / 35
31. Background
Methods
Results
Summary
Case Cohort Study Design
Population
Model
Software
Improvements
Stratified by CCOR Cox PH
Sj (t) = S0j (t)exp(β1·Age+β2·CIRS-Gpros )
, j = 1, . . . , 8
We improve on Wykes (2011) work in two ways:
we estimate both the baseline survival function and relative
risk parameters using the case cohort population
we use a stratified by CCOR analysis, which assumes a
different baseline hazard for each region
9 / 35
32. Background
Methods
Results
Summary
Case Cohort Study Design
Population
Model
Software
Relative Risk Estimation
Prentice (1986) proposed a pseudolikelihood given by
˜L(β) =
D
j=1
exp zT
(j)β
k∈˜Rtj
exp zT
k β wk
10 / 35
33. Background
Methods
Results
Summary
Case Cohort Study Design
Population
Model
Software
Relative Risk Estimation
Prentice (1986) proposed a pseudolikelihood given by
˜L(β) =
D
j=1
exp zT
(j)β
k∈˜Rtj
exp zT
k β wk
t1, . . . , tD: D distinct failure times
10 / 35
34. Background
Methods
Results
Summary
Case Cohort Study Design
Population
Model
Software
Relative Risk Estimation
Prentice (1986) proposed a pseudolikelihood given by
˜L(β) =
D
j=1
exp zT
(j)β
k∈˜Rtj
exp zT
k β wk
t1, . . . , tD: D distinct failure times
˜Rtj : the risk set at failure time tj
10 / 35
35. Background
Methods
Results
Summary
Case Cohort Study Design
Population
Model
Software
Relative Risk Estimation
Prentice (1986) proposed a pseudolikelihood given by
˜L(β) =
D
j=1
exp zT
(j)β
k∈˜Rtj
exp zT
k β wk
t1, . . . , tD: D distinct failure times
˜Rtj : the risk set at failure time tj
wk: the weight applied to the risk set
10 / 35
36. Background
Methods
Results
Summary
Case Cohort Study Design
Population
Model
Software
Weighting Schemes
exp zT
i β
Yi (tj )wi (tj ) exp zT
i β +
k∈SC,k=i
Yk(tj )wk(tj ) exp zT
k β
Table: Comparing different values of wk
Outcome type and timing Prentice Self & Prentice Barlow
(1986) (1988) (1994)
Case outside SC before failure 0 0 0
Case outside SC at failure 1 0 1
Case in SC before failure 1 1 1/α
Case in SC at failure 1 1 1
SC control 1 1 1/α
11 / 35
37. Background
Methods
Results
Summary
Case Cohort Study Design
Population
Model
Software
Relative Risk Estimation
Therneau & Li (1999) showed that Var(˜β) can be computed by
adjusting the outputted variance from standard Cox model
programs:
˜I−1
+ (1 − α)DT
SC DSC
12 / 35
38. Background
Methods
Results
Summary
Case Cohort Study Design
Population
Model
Software
Relative Risk Estimation
Therneau & Li (1999) showed that Var(˜β) can be computed by
adjusting the outputted variance from standard Cox model
programs:
˜I−1
+ (1 − α)DT
SC DSC
˜I−1: variance returned by the Cox model program
12 / 35
39. Background
Methods
Results
Summary
Case Cohort Study Design
Population
Model
Software
Relative Risk Estimation
Therneau & Li (1999) showed that Var(˜β) can be computed by
adjusting the outputted variance from standard Cox model
programs:
˜I−1
+ (1 − α)DT
SC DSC
˜I−1: variance returned by the Cox model program
DSC : subset of matrix of dfbeta residuals that contains
subcohort members only
12 / 35
40. Background
Methods
Results
Summary
Case Cohort Study Design
Population
Model
Software
Absolute Risk Estimation
A weighted version of the Breslow estimator for the full cohort
with covariate history z0(u) is given by
d ˆΛ(u, z0(u)) = d ˆΛ0(u) exp zT
0 (u) ˆβ
=
n
i=1
dNi (u)
( n
m ) j∈˜Ri (u) exp zT
j (u) ˆβ
exp zT
0 (u) ˆβ
=
n
i=1
dNi (u)
( n
m ) j∈˜Ri (u) exp (zj(u) − z0(u))T ˆβ
=
n
i=1
m
n
dNi (u)
j∈˜Ri (u) exp (zj(u) − z0(u))T ˆβ
13 / 35
41. Background
Methods
Results
Summary
Case Cohort Study Design
Population
Model
Software
Absolute Risk Estimation
The sum of these increments can be used to calculate the risk
between two time points, say s and t, for s < t:
ˆΛ(s, t, z0) = ˆΛ(0, t, z0) − ˆΛ(0, s, z0) =
t
s
d ˆΛ(u, z0(u))
14 / 35
42. Background
Methods
Results
Summary
Case Cohort Study Design
Population
Model
Software
Absolute Risk Estimation
The sum of these increments can be used to calculate the risk
between two time points, say s and t, for s < t:
ˆΛ(s, t, z0) = ˆΛ(0, t, z0) − ˆΛ(0, s, z0) =
t
s
d ˆΛ(u, z0(u))
For this project, we will be only considering the case when s = 0.
The corresponding survival probability is
ˆS(0, t, z0) = exp − ˆΛ(0, t, z0)
14 / 35
45. Background
Methods
Results
Summary
Case Cohort Study Design
Population
Model
Software
Computational Methods for the Case Cohort Design
Standard Cox PH functions in SAS and R cannot directly
compute the relative and absolute risk estimates
16 / 35
46. Background
Methods
Results
Summary
Case Cohort Study Design
Population
Model
Software
Computational Methods for the Case Cohort Design
Standard Cox PH functions in SAS and R cannot directly
compute the relative and absolute risk estimates
Modifications must be made to the dataset to make use of
PROC PHREG and coxph functions
16 / 35
49. Background
Methods
Results
Summary
Relative Risk Estimates
Absolute Risk Estimates
Web Life Expectancy Tool
Table: All Cause mortality
All Cause
Covariate 1999 2008 Wykes (2011)
Age 1.04 (1.02, 1.05) 1.05 (1.04, 1.06) 1.02 (1.01, 1.03)
CIRS-Gpros 1.14 (1.10, 1.18) 1.13 (1.10, 1.16) 1.12 (1.09, 1.15)
18 / 35
50. Background
Methods
Results
Summary
Relative Risk Estimates
Absolute Risk Estimates
Web Life Expectancy Tool
Table: All Cause mortality
All Cause
Covariate 1999 2008 Wykes (2011)
Age 1.04 (1.02, 1.05) 1.05 (1.04, 1.06) 1.02 (1.01, 1.03)
CIRS-Gpros 1.14 (1.10, 1.18) 1.13 (1.10, 1.16) 1.12 (1.09, 1.15)
Table: Cause Specific mortality
Cause Specific
Covariate PCa (1999)a OC (1999)
Age 1.00 (0.99, 1.02) 1.07 (1.05, 1.09)
CIRS-Gpros 1.032 (0.99, 1.07) 1.25 (1.20, 1.30)
a
predictors not significant
18 / 35
51. Background
Methods
Results
Summary
Relative Risk Estimates
Absolute Risk Estimates
Web Life Expectancy Tool
qqqqqqqqqqqqqqqqqqqqqq
qqqqqqqqqqqqqqqqqqqqqqqq
qqqqqqqqqqqqqqqqqqqqqqqq
q
qqqqqqqqqqqqqqqqqqqqqqqqqq
qqqqqqqqqqqqqqqqqqqqqqqqqqq
qqqqqqqqqqqqqqqqqqqqqqqqqqqq
qqqqqqqqqqqqqqqqqqqqqqqqqqqqq
qqqqqqqqqqqqqqqqqqqqqqqqqqqqq
q
qqqqqqqqqqqqqqqqqqqqqqqqqqqqqq
q
qqqqqqqqqqqqqqqqqqqqqqqqqqqqqqq
q
qqqqqqqqqqqqqqqqqqqqqq
qqqqqqqqqqqqqqqqqqqqqqqqqqqqqqq
qqqqqqqqqqqqqqqqqqqqqqqqqqqqqqq
qqqqqqqqqqqqqqqqqqqqqqqqqqqqqqq
qqqqqqqqqqqqqqqqqqqqqqqqqqqqqqq
qqqqqqqqqqqqqqqqqqqqqqqqqqqqqqq
qqqqqqqqqqqqqqqqqqqqqqqqqqqqqqq
qqqqqqqqqqqqqqqqqqqqqqqqqqqqqqq
qqqqqqqqqqqqqqqqqqqqqqqqqqqqqqq
qqqqqqqqqqqqqqqqqqqqqqqqqqqqqqq
qqqqqqqqqqqqqqqqqqqqqqqqqqqqqqq
q
q
qqqqqqqqq
q
qq
qqqqqqqqqqqqqqqq
q
qqqqqqqqqqqqqqqqqqqqqqq
q
q
q
q
q
qqqqqqqqqqqqqqqqqqqqqqqqq
q
q
qqqqqqqqqqqqqqqqqqqqqqqqqqqqqq
qqqqqqqqqqqqqqqqqqqqqqqqqqqqqqq
qqqqqqqqqqqqqqqqqqqqqqqqqqqqqqq
qqqqqqqqqqqqqq
qqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqq
qqqqqqqqqqqq
qqqqqqqqqqqqqq
qqqqqqqqqqqqqqq
qqqqqqqqqqqqqq
qqqqqqqqqqqqqqqq qqqqqqqqqqqqqqqqq
qqqqqqqqqqqqqqqqqqqqqq
qqqqqqqqqqqqqqqqqqqqqqq
qqqqqqqqqqqqqqqqqqqqqqqqq
qqqqqqqqqqqqqqqqqqqqqqqqqq
qqqqqqqqqqqqqqqqqqqqqqqqqqqq
qqqqqqqqqqqqqqqqqqqqqqqqqqqqqq
qqqqqqqqqqqqqqqqqqqqqqqqqqqqqq
qqqqqqqqqqqqqqqqqqqqqqqqqqqqqq
qqqqqqqqqqqqqqqqqqqqqqqqqqqqqq
qqqqqqqqqqqqqqqqqqqqqqqqqqqqqq
All cause (1999) All cause (2008) OC (1999)
PCa (1999) Wykes (2011)
10
20
30
10
20
30
0 3 4 5 6 7 8 9101112 0 3 4 5 6 7 8 9101112
CIRS−G_pros
lifeexpectancy(years)
50
60
70
80
Age
Figure: Estimates from East Region (Wykes estimates based on all cause unstratified
subcohort at 2008) 19 / 35
52. Background
Methods
Results
Summary
Relative Risk Estimates
Absolute Risk Estimates
Web Life Expectancy Tool
5
10
15
20
25
0 3 4 5 6 7 8 9 10 11 12
CIRS−G_pros
lifeexpectancy(years)
Age
55
65
75
Model
All cause (2008)
Wykes (2011)
Figure: Estimates from East Region (Wykes estimates based on all cause unstratified
subcohort at 2008) 20 / 35
53. 0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95
age (years)
survivalprobability
CCO North West region
(a) All cause (1999)
0.4
0.5
0.6
0.7
0.8
0.9
1.0
0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95
age (years)
survivalprobability
CCO North West region
(b) Other cause(1999)
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90
age (years)
survivalprobability
CCO North West region
(c) Prostate cancer(1999)
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 100
age (years)
survivalprobability
CCO North West region
(d) All cause (2008)
65. Background
Methods
Results
Summary
Contributions
Future Work
References
References I
M. Center et al.
International Variation in Prostate Cancer Incidence and
Mortality Rates
European Association of Urology, 61:2012
Bryan Langholz and Jenny Jiao
Computational methods for case cohort studies
Computational Statistics & Data Analysis, 51:2007
Therneau, T and Li, H
Computing the Cox model for case cohort designs
Lifetime data analysis, 2(5):1999
26 / 35
66. Background
Methods
Results
Summary
Contributions
Future Work
References
References II
PRENTICE, R. L.
A case-cohort design for epidemiologic cohort studies and
disease prevention trials
Biometrika, 73(1):1986
RStudio and Inc.
shiny: Web Application Framework for R
R package version 0.6.0, 17(25):2012
T. Lumley
survey: analysis of complex survey samples
R package version 3.28-2,2012
27 / 35
68. Background
Methods
Results
Summary
Contributions
Future Work
References
Influential Analysis
Dfbeta residuals
Are the approximate changes in the parameter estimates (ˆβ − ˆβ(j))
when the jth observation is omitted. These variables are a
weighted transform of the score residual variables and are useful in
assessing local influence and in computing approximate and robust
variance estimates
Standardized dfbeta residuals:
(ˆβ − ˆβ(j))
sd(ˆβ − ˆβ(j))
29 / 35
69. Background
Methods
Results
Summary
Contributions
Future Work
References
Why is the PsL not a Partial Likelihood
Not a PL → Asymptotic distribution theory for the PL
estimators breaks down because cases outside the subcohort
induce non-nesting of the σ-fields, or sets on which the
counting process probability measure is defined
cannot use Likelihood ratio tests since PsL are not real
likelihoods
30 / 35
70. Background
Methods
Results
Summary
Contributions
Future Work
References
Consistent Estimators
An estimator ˆθ will perform better and better as we obtain more
samples. If at the limit n → ∞ the estimator tends to be always
right, it is said to be consistent. This notion is equivalent to
convergence in probability
Convergence in Probability
Let X1, . . . , Xn be a sequence of iid RVs drawn from a distribution
with parameter θ and ˆθ an estimator for θ. ˆθ is consistent as an
estimator of θ if ˆθ
p
→ θ
lim
n→∞
P(|ˆθ − θ| > ) = 0, ∀ > 0
31 / 35
71. Background
Methods
Results
Summary
Contributions
Future Work
References
Competing Risks
when disease of interest has a delayed occurrence, other
events may preclude observation of disease of interest giving
rise to a competing risk situation
to analyze the event of interest taking into account the
competing risks, one has to model the hazard of
subdistribution. In the presence of competing risks, the
probability to observe the event of interest spans the interval
[0, p], p < 1. Because p does not reach 1, this function is not
a proper distribution function and it is called a subdistribution
Modify the partial likelihood in the Cox model to model
hazard of subdistribution, such that the individuals
experiencing competing risks are always in the risk set with
their contribution regulated by a weight
32 / 35
72. Background
Methods
Results
Summary
Contributions
Future Work
References
Competing Risks
Fine and Gray Partial Likelihood:
PL(β) =
n
j=1
exp {βxj }
r∈˜Rj
wrj exp {βxr }
δj
(1)
wrj =
ˆG(tj )
ˆG(min(tj , tr ))
(2)
Cj =
1 if the event of interest was observed at time tj
2 if the competing risk event was observed at time tj
0 if no event was observed at time tj
δj =
1 Cj = 1
0 Cj = 0 or Cj = 2
33 / 35
73. Background
Methods
Results
Summary
Contributions
Future Work
References
Competing Risks
ˆG(tj ): estimated probability of being censored at time tj by
either the event of interest or a competing risks event
˜Rj = {r; tr ≥ tj or Cr = 2}
individuals experiencing competing risk event are always
considered in the risk set at all time points
The weight controls how much a specific observation
participates in the sum in the denominator
If the observation at time tr is censored (Cr = 0) or has the
event of interest (Cr = 1), the weight is either 1 or 0
depending whether the observation is in the risk set (tr ≥ tj )
or it is not in the risk set (tr < tj )
If the observation has a competing risk event, then it is always
in the risk set and the weight is 1 if (tr ≥ tj ) or a quantity less
than 1 if tr < tj
This quantity decreases as the distance between tr and tj
34 / 35
74. Background
Methods
Results
Summary
Contributions
Future Work
References
Case-cohort in the presence of competing risks
PsL(β) =
n
j=1
exp {βxj }
r∈˜Rj
Irj wrj exp {βxr }
δj
δj =
1 Cj = 1
0 Cj = 0 or Cj = 2
Irj =
1 r is in subcohort or r is a case outside the subcohort and tr
0 r is a case outside the subcohort and tr = tj
wrj =
ˆG(tj )
ˆG(min(tj ,tr ))
, where ˆG(tj ) is the estimation of the probability
of censoring regardless of the fact that in a case-cohort study the
cases are oversampled
35 / 35