Win ratio: On interpretation and handling of ties...Jiuzhou (Joe) Wang, ImmunoGen Inc. Marc...
Transcript of Win ratio: On interpretation and handling of ties...Jiuzhou (Joe) Wang, ImmunoGen Inc. Marc...
![Page 1: Win ratio: On interpretation and handling of ties...Jiuzhou (Joe) Wang, ImmunoGen Inc. Marc Vandemeulebroecke, Novartis Pharma AG. Regulatory-Industry Workshop, Sep. 25, 2019. Win](https://reader033.fdocuments.us/reader033/viewer/2022060517/604ae5bf1b0add2166327963/html5/thumbnails/1.jpg)
Gaohong Dong, iStats Inc.Junshan Qiu, FDA/CDER
David C. Hoaglin, Univ. of Mass. Medical SchoolRoland Matsuoaka, Duke Univ. Sch. of Medicine
Yu-Wei (Victoria) Chang, AbbvieJiuzhou (Joe) Wang, ImmunoGen Inc.
Marc Vandemeulebroecke, Novartis Pharma AG
Regulatory-Industry Workshop, Sep. 25, 2019
Win ratio: On interpretation and handling of ties
![Page 2: Win ratio: On interpretation and handling of ties...Jiuzhou (Joe) Wang, ImmunoGen Inc. Marc Vandemeulebroecke, Novartis Pharma AG. Regulatory-Industry Workshop, Sep. 25, 2019. Win](https://reader033.fdocuments.us/reader033/viewer/2022060517/604ae5bf1b0add2166327963/html5/thumbnails/2.jpg)
Post-Workshop note
| Sep. 25, 2019 | 2019 ASA Biopharmaceutical Section Regulatory-Industry Statistics Workshop 2
The following were added per some questions raised duringthe session:
Algorithms (rules) defining winners (losers) and ties
Win ratio estimands
Details of these two aspects can be found in the paper βThewin ratio: On interpretation and handling of tiesβ (Dong et al.,2019)
![Page 3: Win ratio: On interpretation and handling of ties...Jiuzhou (Joe) Wang, ImmunoGen Inc. Marc Vandemeulebroecke, Novartis Pharma AG. Regulatory-Industry Workshop, Sep. 25, 2019. Win](https://reader033.fdocuments.us/reader033/viewer/2022060517/604ae5bf1b0add2166327963/html5/thumbnails/3.jpg)
| Sep. 25, 2019 | 2019 ASA Biopharmaceutical Section Regulatory-Industry Statistics Workshop 3
![Page 4: Win ratio: On interpretation and handling of ties...Jiuzhou (Joe) Wang, ImmunoGen Inc. Marc Vandemeulebroecke, Novartis Pharma AG. Regulatory-Industry Workshop, Sep. 25, 2019. Win](https://reader033.fdocuments.us/reader033/viewer/2022060517/604ae5bf1b0add2166327963/html5/thumbnails/4.jpg)
Outline
| Sep. 25, 2019 | 2019 ASA Biopharmaceutical Section Regulatory-Industry Statistics Workshop 4
Analyses of composite endpoints (two examples)
Win ratio and stratified win ratio
Algorithms (rules) defining winners (losers) and ties
Interpretation: Win probability/proportion
Interpretation: Connection to commonly used statistics
Handling of ties and the win odds
Win ratio estimands
Best use of the win ratio
Summary
![Page 5: Win ratio: On interpretation and handling of ties...Jiuzhou (Joe) Wang, ImmunoGen Inc. Marc Vandemeulebroecke, Novartis Pharma AG. Regulatory-Industry Workshop, Sep. 25, 2019. Win](https://reader033.fdocuments.us/reader033/viewer/2022060517/604ae5bf1b0add2166327963/html5/thumbnails/5.jpg)
Example 1: Phase III Liver Transplant Trial
| Sep. 25, 2019 | 2019 ASA Biopharmaceutical Section Regulatory-Industry Statistics Workshop 5
Phase III liver transplant trial: 488 = 243+245 patientsrandomized into the following two treatment groupsβ’ Treatment: EVR + reduced TAC exposureβ’ Control: standard TAC exposure
Primary endpoint is a composite ofβ’ Treated biopsy-proven acute rejection (tBPAR)β’ Graft loss (GL)β’ Death (D)
This trial was completed in 2013
![Page 6: Win ratio: On interpretation and handling of ties...Jiuzhou (Joe) Wang, ImmunoGen Inc. Marc Vandemeulebroecke, Novartis Pharma AG. Regulatory-Industry Workshop, Sep. 25, 2019. Win](https://reader033.fdocuments.us/reader033/viewer/2022060517/604ae5bf1b0add2166327963/html5/thumbnails/6.jpg)
Example 1: Phase III Liver Transplant Trial (cont.)
Q1: Was the numerical difference in GL/D a chance finding? Q2: Could we put more weight on more important outcomes GL and D?| Sep. 25, 2019 | 2019 ASA Biopharmaceutical Section Regulatory-Industry Statistics Workshop 6
Source: Dong et al. (2016), Saliba et al. (2013),De Simone et al. (2012), ClincalTrials.gov (NCT00622869)
![Page 7: Win ratio: On interpretation and handling of ties...Jiuzhou (Joe) Wang, ImmunoGen Inc. Marc Vandemeulebroecke, Novartis Pharma AG. Regulatory-Industry Workshop, Sep. 25, 2019. Win](https://reader033.fdocuments.us/reader033/viewer/2022060517/604ae5bf1b0add2166327963/html5/thumbnails/7.jpg)
Example 2: CHARM program
| Sep. 25, 2019 | 2019 ASA Biopharmaceutical Section Regulatory-Industry Statistics Workshop 7
CHARM program: included 3 separate randomized trialscomparing candesartan with placebo in subjects withchronic heart failure (CHF).
Primary endpoint: Composite of cardiovascular (CV)death or hospitalization for CHF.
The three CHARM trials were completed in 2003 with7599 subjects with median follow-up 3.14 years
![Page 8: Win ratio: On interpretation and handling of ties...Jiuzhou (Joe) Wang, ImmunoGen Inc. Marc Vandemeulebroecke, Novartis Pharma AG. Regulatory-Industry Workshop, Sep. 25, 2019. Win](https://reader033.fdocuments.us/reader033/viewer/2022060517/604ae5bf1b0add2166327963/html5/thumbnails/8.jpg)
Example 2: CHARM program (cont.)
| Sep. 25, 2019 | 2019 ASA Biopharmaceutical Section Regulatory-Industry Statistics Workshop 8
Only 54% of CV deaths contributed to the compositeQ: Could all CV deaths be considered for the analysis?
Source: Pocock et al. (2012)
![Page 9: Win ratio: On interpretation and handling of ties...Jiuzhou (Joe) Wang, ImmunoGen Inc. Marc Vandemeulebroecke, Novartis Pharma AG. Regulatory-Industry Workshop, Sep. 25, 2019. Win](https://reader033.fdocuments.us/reader033/viewer/2022060517/604ae5bf1b0add2166327963/html5/thumbnails/9.jpg)
Composite endpoint
| Sep. 25, 2019 | 2019 ASA Biopharmaceutical Section Regulatory-Industry Statistics Workshop 9
Advantages:- Event rate is higher than those of the components alone.
==> Sample size can be reduced- Avoid multiplicity and competing risk- Captures multiple aspects of treatment effect
Disadvantages:- All components are treated equally important- First event analyzed may not be the most important- Some components could show an opposite treatment effect- A less important component could be dominant
![Page 10: Win ratio: On interpretation and handling of ties...Jiuzhou (Joe) Wang, ImmunoGen Inc. Marc Vandemeulebroecke, Novartis Pharma AG. Regulatory-Industry Workshop, Sep. 25, 2019. Win](https://reader033.fdocuments.us/reader033/viewer/2022060517/604ae5bf1b0add2166327963/html5/thumbnails/10.jpg)
Win ratio
| Sep. 25, 2019 | 2019 ASA Biopharmaceutical Section Regulatory-Industry Statistics Workshop 10
Win ratio (Pocock et al., 2012)- Consider the most important outcome (e.g. death) first,
then next important event, β¦ etc.- Can handle a composite of multiple outcomes in any data
type (e.g., time-to-event, ordinal, continuous, β¦)- Can handle non-proportional hazards situations- Enable project specific rules defining winners (losers) and
ties- The name of βwin ratioβ is intuitive
Two closely related methods:- Net benefit (proportion in favor of treatment) by Buyse
(2010).- Finkelstein and Schoenfeld (1999) : Combining mortality &
longitudinal outcome
![Page 11: Win ratio: On interpretation and handling of ties...Jiuzhou (Joe) Wang, ImmunoGen Inc. Marc Vandemeulebroecke, Novartis Pharma AG. Regulatory-Industry Workshop, Sep. 25, 2019. Win](https://reader033.fdocuments.us/reader033/viewer/2022060517/604ae5bf1b0add2166327963/html5/thumbnails/11.jpg)
Recent methodological developments of the win ratio
β’ Pocock et al. (2012): Original proposalβ’ Luo et al. (2015): Closed-form variance estimator for a particular way of
defining wins (or losses) and tiesβ’ Bebu and Lachin (2016) and Dong et al (2016): Closed-form variance
estimator for any way of defining wins (or losses) and tiesβ’ Wang and Pocock (2016): Win ratio for non-normal outcomesβ’ Oakes (2016): Win and loss probabilities in a limited time cβ’ Luo et al. (2017): Weighted win and lossβ’ Dong et al. (2018): The stratified win ratioβ’ Finkelstein and Schoenfeld (2019): Win ratio as a function of time t.β’ Mao (2019): on hypothesis of the win ratioβ’ Dong et al (2019): Win ratio: interpretation and handling of tiesβ’ Dong et al (in press): Win ratio: Impact of censoring and follow-up time
and use with non-proportional hazards
| Sep. 25, 2019 | 2019 ASA Biopharmaceutical Section Regulatory-Industry Statistics Workshop 11
![Page 12: Win ratio: On interpretation and handling of ties...Jiuzhou (Joe) Wang, ImmunoGen Inc. Marc Vandemeulebroecke, Novartis Pharma AG. Regulatory-Industry Workshop, Sep. 25, 2019. Win](https://reader033.fdocuments.us/reader033/viewer/2022060517/604ae5bf1b0add2166327963/html5/thumbnails/12.jpg)
Win ratio
Based on pairwise comparisons: each patient in the Treatmentgroup is compared with every patient in the Control group.
Treatment Patient 1
Treatment Patient Nt
β¦ β¦
Control Patient 1
Control Patient Nc
Control Patient 2β¦ β¦
Control Patient 1
Control Patient Nc
Control Patient 2β¦ β¦
TRT win Con win Tied
β¦ β¦
| Sep. 25, 2019 | 2019 ASA Biopharmaceutical Section Regulatory-Industry Statistics Workshop 12
![Page 13: Win ratio: On interpretation and handling of ties...Jiuzhou (Joe) Wang, ImmunoGen Inc. Marc Vandemeulebroecke, Novartis Pharma AG. Regulatory-Industry Workshop, Sep. 25, 2019. Win](https://reader033.fdocuments.us/reader033/viewer/2022060517/604ae5bf1b0add2166327963/html5/thumbnails/13.jpg)
Win ratio (cont.)
For each pair,
Mostimportant outcome
nextimportant outcome
Win (loss)
Win (loss)
Can not determine
Wππππ ππππππππππ =ππππππππππππ ππππ π€π€πππππ€π€ ππππππ ππππππππππππππππππππππππππππππππ ππππ π€π€πππππ€π€ ππππππ πΆπΆπππππππππππΆπΆ
| Sep. 25, 2019 | 2019 ASA Biopharmaceutical Section Regulatory-Industry Statistics Workshop 13
![Page 14: Win ratio: On interpretation and handling of ties...Jiuzhou (Joe) Wang, ImmunoGen Inc. Marc Vandemeulebroecke, Novartis Pharma AG. Regulatory-Industry Workshop, Sep. 25, 2019. Win](https://reader033.fdocuments.us/reader033/viewer/2022060517/604ae5bf1b0add2166327963/html5/thumbnails/14.jpg)
Win ratio(cont.)
| Sep. 25, 2019 | 2019 ASA Biopharmaceutical Section Regulatory-Industry Statistics Workshop 14
Patient A
Time
Death
Death
Hosp
HospPatient B
Who wins?β’ Win ratio: Patient B wins on Deathβ’ First-event analysis: Patient A wins on Hospitalization
![Page 15: Win ratio: On interpretation and handling of ties...Jiuzhou (Joe) Wang, ImmunoGen Inc. Marc Vandemeulebroecke, Novartis Pharma AG. Regulatory-Industry Workshop, Sep. 25, 2019. Win](https://reader033.fdocuments.us/reader033/viewer/2022060517/604ae5bf1b0add2166327963/html5/thumbnails/15.jpg)
Algorithms (rules) defining winners (losers) and ties
| Sep. 25, 2019 | 2019 ASA Biopharmaceutical Section Regulatory-Industry Statistics Workshop 15
No general rule can fit all disease indications Who is the winner?
β’ Death on Day 100 vs. censored on Day 98β’ Hospitalized through the death on Day 100 vs. heathy and died
on Day 98 Examples:
β’ Any difference is a differenceβ’ A minimal difference is required to be a win (Peron et al., 2016)β’ Dropouts were assumed alive given low mortality rate in the
phase III liver transplant trial (Dong et al., 2016) Rules should be project specific Defining rules β a joint work of statistician and clinical team Rules should be prespecified in the analysis plan
![Page 16: Win ratio: On interpretation and handling of ties...Jiuzhou (Joe) Wang, ImmunoGen Inc. Marc Vandemeulebroecke, Novartis Pharma AG. Regulatory-Industry Workshop, Sep. 25, 2019. Win](https://reader033.fdocuments.us/reader033/viewer/2022060517/604ae5bf1b0add2166327963/html5/thumbnails/16.jpg)
| Sep. 25, 2019 | 2019 ASA Biopharmaceutical Section Regulatory-Industry Statistics Workshop 16
Win ratio and stratified win ratio
Let
β’ ππππ : patient ππ in Treatment ππ = 1, β¦ ,πππ‘π‘
β’ ππππ : patient ππ in Control ππ = 1, β¦ ,ππππ
Define
K(Xi ,Yj) = 1 if Xi wins over Yj
= 0 otherwise
L(Xi ,Yj) = 1 if Yj wins over Xi
= 0 otherwise
# of wins for the Treatment # of wins for the Control
πππ‘π‘ = οΏ½ππ=1
πππ‘π‘
οΏ½ππ=1
ππππ
[πΎπΎ ππππ ,ππππ = 1] ππππ = οΏ½ππ=1
πππ‘π‘
οΏ½ππ=1
ππππ
[πΏπΏ ππππ ,ππππ = 1]
![Page 17: Win ratio: On interpretation and handling of ties...Jiuzhou (Joe) Wang, ImmunoGen Inc. Marc Vandemeulebroecke, Novartis Pharma AG. Regulatory-Industry Workshop, Sep. 25, 2019. Win](https://reader033.fdocuments.us/reader033/viewer/2022060517/604ae5bf1b0add2166327963/html5/thumbnails/17.jpg)
| Sep. 25, 2019 | 2019 ASA Biopharmaceutical Section Regulatory-Industry Statistics Workshop 17
Win ratio and stratified win ratio (cont.)
K(Xi ,Yj) = 1 if Xi wins over Yj
= 0 otherwise
L(Xi ,Yj) = 1 if Yj wins over Xi
= 0 otherwise
# of wins for the Treatment # of wins for the Control
Win ratio: WR = nt / nc
Stratified win ratio:
where w(m) is the weight for mth stratum (= 1, 2, ..., M).
πππ‘π‘ = οΏ½ππ=1
πππ‘π‘
οΏ½ππ=1
ππππ
[πΎπΎ ππππ ,ππππ = 1] ππππ = οΏ½ππ=1
πππ‘π‘
οΏ½ππ=1
ππππ
[πΏπΏ ππππ ,ππππ = 1]
( ) ( )
( ) ( )mc
M
mm
M
mm
tm
nw
nwWR
ββ
=
==1
1
![Page 18: Win ratio: On interpretation and handling of ties...Jiuzhou (Joe) Wang, ImmunoGen Inc. Marc Vandemeulebroecke, Novartis Pharma AG. Regulatory-Industry Workshop, Sep. 25, 2019. Win](https://reader033.fdocuments.us/reader033/viewer/2022060517/604ae5bf1b0add2166327963/html5/thumbnails/18.jpg)
Stratified win ratio (cont.)
| Sep. 25, 2019 | 2019 ASA Biopharmaceutical Section Regulatory-Industry Statistics Workshop 18
We propose the stratified win ratio with w(m) = 1/N(m), motivatedby the Mantel-Haenszel method :
Why?β’ The win ratio can reduce to the odds ratio.β’ This stratified win ratio can reduce to the Mantel-Haenszel
stratified odds ratio.β’ Mantel-Haenszel stratified odds ratio is similar or robust
than other stratified odds ratios (e.g. logic, MLE).
( ) ( )
( ) ( )ββ
=
== M
mmm
c
M
mmm
t
Nn
NnWR
1
1
![Page 19: Win ratio: On interpretation and handling of ties...Jiuzhou (Joe) Wang, ImmunoGen Inc. Marc Vandemeulebroecke, Novartis Pharma AG. Regulatory-Industry Workshop, Sep. 25, 2019. Win](https://reader033.fdocuments.us/reader033/viewer/2022060517/604ae5bf1b0add2166327963/html5/thumbnails/19.jpg)
| Sep. 25, 2019 | 2019 ASA Biopharmaceutical Section Regulatory-Industry Statistics Workshop 19
Interpretation: Connections to commonly used statistics
Single outcome Commonly used statistic Connection
Binary Odds ratio (OR) WR = 1/OR
Time-to-event Hazard ratio (HR) WR = 1/HR, when the proportionalhazards assumption is met.
Normallydistributedcontinuous
Mean difference (Ξ΄)WR = f(Ξ΄)** Acion et al. (2006) and Buyse (2010)
Non-normalcontinuous
Mann-Whitney U The number of wins (nt and nc) in thewin ratio = Mann-Whitney U, when thereare no ties.
![Page 20: Win ratio: On interpretation and handling of ties...Jiuzhou (Joe) Wang, ImmunoGen Inc. Marc Vandemeulebroecke, Novartis Pharma AG. Regulatory-Industry Workshop, Sep. 25, 2019. Win](https://reader033.fdocuments.us/reader033/viewer/2022060517/604ae5bf1b0add2166327963/html5/thumbnails/20.jpg)
Win probability (Oakes, 2016)πππ‘π‘(π₯π₯) = ππππππππ ππ π‘π‘ > ππ ππ = βοΏ½
0
π₯π₯πΉπΉ(π‘π‘)πππΉπΉ(ππ)
ππππ π₯π₯ =πππ‘π‘(π₯π₯)ππππ(π₯π₯)
=ππππππππ ππ π‘π‘ > ππ ππ
ππππππππ ππ ππ > ππ π‘π‘ =ββ«0
π₯π₯ πΉπΉ(π‘π‘)πππΉπΉ(ππ)
ββ«0π₯π₯ πΉπΉ(ππ)πππΉπΉ(π‘π‘)
Win proportions: estimates of win probabilities
οΏ½πππ‘π‘ = πππ‘π‘ = πππ‘π‘/πππ‘π‘ππππβ² οΏ½ππππ = ππππ = ππππ/πππ‘π‘πππποΏ½ππππ = πππ‘π‘/ππππ
Proportions of βwinsβ are more readily interpretable than thenumber of βwinsβ
Example: WR=1.2 means that a patient in the treatment group is20% more likely to βwinβ than a patient in the control group
| Sep. 25, 2019 | 2019 ASA Biopharmaceutical Section Regulatory-Industry Statistics Workshop 20
Interpretation: Win probablity/proportion
![Page 21: Win ratio: On interpretation and handling of ties...Jiuzhou (Joe) Wang, ImmunoGen Inc. Marc Vandemeulebroecke, Novartis Pharma AG. Regulatory-Industry Workshop, Sep. 25, 2019. Win](https://reader033.fdocuments.us/reader033/viewer/2022060517/604ae5bf1b0add2166327963/html5/thumbnails/21.jpg)
| Sep. 25, 2019 | 2019 ASA Biopharmaceutical Section Regulatory-Industry Statistics Workshop 21
Example 1 of Phase III liver transplant trial: 243 vs. 245patients randomized into Treatment vs. Control groups
Number of pairs: NtNc= 243 Γ 245 = 59535
Numbers of wins and ties:
nt = 5393, nc = 3761, and ntie = 50381,
Good to present win proportions instead:
πππ‘π‘ = πππ‘π‘πππ‘π‘ππππ
= 539359535
= 9.1% and ππππ = 6.3%
There are too many ties, but ignored in the win ratio calculation
Interpretation: Win probablity/proportion(cont.)
οΏ½ππππ =πππ‘π‘ππππ
= 1.4
![Page 22: Win ratio: On interpretation and handling of ties...Jiuzhou (Joe) Wang, ImmunoGen Inc. Marc Vandemeulebroecke, Novartis Pharma AG. Regulatory-Industry Workshop, Sep. 25, 2019. Win](https://reader033.fdocuments.us/reader033/viewer/2022060517/604ae5bf1b0add2166327963/html5/thumbnails/22.jpg)
| Sep. 25, 2019 | 2019 ASA Biopharmaceutical Section Regulatory-Industry Statistics Workshop 22
Mann-Whitney U:
Win odds (WO) β an extension of the Mann-Whitney odds
Handling of ties and the win odds
ββ= =
β+=
t cN
i
N
jjiji YXIYXIU
1 1)(
21)(
( )ββ= =
ββ+=
t cN
i
N
jjijiji YXLYXKYXK
1 1),(),(1
21),(
ct NNUU =2
tiec
tiet
tiec
tiet
PPPP
nnnn
UUWO
5.05.0
5.05.0
1 2
2
++
=++
=β
=
![Page 23: Win ratio: On interpretation and handling of ties...Jiuzhou (Joe) Wang, ImmunoGen Inc. Marc Vandemeulebroecke, Novartis Pharma AG. Regulatory-Industry Workshop, Sep. 25, 2019. Win](https://reader033.fdocuments.us/reader033/viewer/2022060517/604ae5bf1b0add2166327963/html5/thumbnails/23.jpg)
| Sep. 25, 2019 | 2019 ASA Biopharmaceutical Section Regulatory-Industry Statistics Workshop 23
Win Ratio vs. Win Odds
Win Ratio Win Odds
Wππ =πππ‘π‘ππππ
=πππ‘π‘ππππ
WR=1.2 means that a patientin the treatment group is 20%more likely to βwinβ than apatient in the control group.
WO=1.2 means that a patient in the treatment group is 20% more likely to βwinβ or βtieβ than a patient in the control group.
tiec
tiet
tiec
tiet
PPPP
nnnnWO
5.05.0
5.05.0
++
=++
=
![Page 24: Win ratio: On interpretation and handling of ties...Jiuzhou (Joe) Wang, ImmunoGen Inc. Marc Vandemeulebroecke, Novartis Pharma AG. Regulatory-Industry Workshop, Sep. 25, 2019. Win](https://reader033.fdocuments.us/reader033/viewer/2022060517/604ae5bf1b0add2166327963/html5/thumbnails/24.jpg)
| Sep. 25, 2019 | 2019 ASA Biopharmaceutical Section Regulatory-Industry Statistics Workshop 24
PEACE trial: a cardiovascular study (reported in 2004)with low event rates and long follow-up time.
Composite of CV death, MI, CABG, and PTCA.
Example 3: PEACE trial
GroupStratum-
specific win Proportion
Stratified win
Prop.
Stratified win ratio(95% CI)
Stratified win odds(95% CI)
Hazard ratio in favor of
treatment (95% CI)Female Male
Trandolapril
16.5% 18.1% 18.0% 1.03(0.94, 1.13)
1.01(0.98,1.04)
1.04(0.94, 1.14)
Placebo 16.4% 17.5% 17.4%
![Page 25: Win ratio: On interpretation and handling of ties...Jiuzhou (Joe) Wang, ImmunoGen Inc. Marc Vandemeulebroecke, Novartis Pharma AG. Regulatory-Industry Workshop, Sep. 25, 2019. Win](https://reader033.fdocuments.us/reader033/viewer/2022060517/604ae5bf1b0add2166327963/html5/thumbnails/25.jpg)
ATTR-ACT study: Compare tafamidis with placebo for thetreatment of TTR-CM (transthyretin cardiomyopathy)
Stratified by TTR genotype (variant vs. wild-type) and baselinedisease severity (NYHA Class I/II vs. III)
Primary composite endpoint of
- All-cause mortality
- Frequency of cardiovascular-related hospitalization
Started: Dec 2013; Completed : Feb 2018
Stratified win ratio (primary analysis with equal weight): 1.7 (95%CI:1.26, 2.29; P=0.0006 per Finkelstein and Schoenfeld (1999))
Drug approval: Vyndaqel/Vyndamax (tafamidis) in early 2019
| Sep. 25, 2019 | 2019 ASA Biopharmaceutical Section Regulatory-Industry Statistics Workshop 25
Example 4: ATTR-ACT trial (Pfizer)
Source: Maurer et al. (2018), Pocock and Collier (2019)
![Page 26: Win ratio: On interpretation and handling of ties...Jiuzhou (Joe) Wang, ImmunoGen Inc. Marc Vandemeulebroecke, Novartis Pharma AG. Regulatory-Industry Workshop, Sep. 25, 2019. Win](https://reader033.fdocuments.us/reader033/viewer/2022060517/604ae5bf1b0add2166327963/html5/thumbnails/26.jpg)
Win ratio estimands
| Sep. 25, 2019 | 2019 ASA Biopharmaceutical Section Regulatory-Industry Statistics Workshop 26
Censoring impact on the win ratio
What treatment effect is to be estimated
Addendum of ICH E9 Guidelines
Attributes of win ratio estimands:
1) Population: defined per inclusion/exclusion criteria
2) Variable (endpoint or outcome): a composite endpoint ofprioritized multiple outcomes of interest
3) Intercurrent event: events that occur after treatmentinitiation, e.g., rescue medication, discontinuation, β¦etc,may cause informative censoring
4) Summary measure: the (stratified or weighted) win ratiostatistic to compare the two treatment groups
![Page 27: Win ratio: On interpretation and handling of ties...Jiuzhou (Joe) Wang, ImmunoGen Inc. Marc Vandemeulebroecke, Novartis Pharma AG. Regulatory-Industry Workshop, Sep. 25, 2019. Win](https://reader033.fdocuments.us/reader033/viewer/2022060517/604ae5bf1b0add2166327963/html5/thumbnails/27.jpg)
Win ratio estimands (cont.)
| Sep. 25, 2019 | 2019 ASA Biopharmaceutical Section Regulatory-Industry Statistics Workshop 27
Intercurrent events:a) Dropouts:
β’ Consider discontinuations as outcomes in the composite(composite strategy, ICH E9 Addendum)
β’ Different discontinuations (e.g., AE vs. lack of efficacy) may beconsidered in different ways
b) Rescue medication and switching study treatments: maybe considered similarly to that for treatment discontinuation
c) Missing endpoints: Perform imputations first, then calculatethe win ratio
d) etc..
![Page 28: Win ratio: On interpretation and handling of ties...Jiuzhou (Joe) Wang, ImmunoGen Inc. Marc Vandemeulebroecke, Novartis Pharma AG. Regulatory-Industry Workshop, Sep. 25, 2019. Win](https://reader033.fdocuments.us/reader033/viewer/2022060517/604ae5bf1b0add2166327963/html5/thumbnails/28.jpg)
When the more important endpoints tend to occur laterand are reasonably frequent
- May provide a considerable benefit, when a substantialnumber of patients experience more than one endpoint
- Would have minimal advantage, when more importantoutcomes are infrequent.
- The follow-up time should be long enough to yield a usefulnumber of occurrences of the more important endpoints
Censoring times are similar in the two groups, unlessthe event times follow a proportional-hazards model.
| Sep. 25, 2019 | 2019 ASA Biopharmaceutical Section Regulatory-Industry Statistics Workshop 28
Best use of the win ratio
Source: Dong et al. (in press): Win ratio: Impact of censoring and follow-up time and use with non-proportional hazards. Pharmaceutical Statistics
![Page 29: Win ratio: On interpretation and handling of ties...Jiuzhou (Joe) Wang, ImmunoGen Inc. Marc Vandemeulebroecke, Novartis Pharma AG. Regulatory-Industry Workshop, Sep. 25, 2019. Win](https://reader033.fdocuments.us/reader033/viewer/2022060517/604ae5bf1b0add2166327963/html5/thumbnails/29.jpg)
Conventional analysis uses time to the first event. The first eventanalyzed may not be the most important outcome
Win ratio is the ratio of numbers of wins for two groups, it is alsothe ratio of win probabilities (proportions) for two groups. It has aconnection to some commonly used statistics.
Win odds assigns 50% of ties to the numerator and thedenominator, it is always closer to the null value 1.0, especiallywhen the number of ties is large.
Win ratio and the win odds are for time to the worst event. Theyconsider the importance order of multiple outcomes. Theyprovide an alternative way to analyze composite endpoints.
Win ratio has been applied for study designs; one drug approvalhas been obtained, with the win ratio/Finkelstein-Schoenfeld testas the primary analysis.
Summary
| Sep. 25, 2019 | 2019 ASA Biopharmaceutical Section Regulatory-Industry Statistics Workshop 29
![Page 30: Win ratio: On interpretation and handling of ties...Jiuzhou (Joe) Wang, ImmunoGen Inc. Marc Vandemeulebroecke, Novartis Pharma AG. Regulatory-Industry Workshop, Sep. 25, 2019. Win](https://reader033.fdocuments.us/reader033/viewer/2022060517/604ae5bf1b0add2166327963/html5/thumbnails/30.jpg)
Acknowledgements
| Sep. 25, 2019 | 2019 ASA Biopharmaceutical Section Regulatory-Industry Statistics Workshop 30
L.J. Wei (Harvard Univ. School of Public Health) for myfirst win ratio work
Bo Huang (Pfizer)
James Song (Beigene)
Yodit Seifu (Merck)