Demonstrating Clinical Effectiveness of a Follow-On Companion Diagnostic Test When a New Clinical Trial is Unfeasible

Abha Sharma, Roche Molecular Systems, May 2015

cobas® KRAS mutation test

Overview of the Presentation

2

Background

Approach for follow- on Diagnostic Test • NDMC Assumption

Bridge 1 : NDMC Criteria for cobas Test

Bridge 2 : NDMC Criterion for FDA-approved test

Bridge 3: Influence Criteria evaluation

Bridge 4: Covariate assessment

Bridge 5: Sensitivity and Robustness Analysis

Background

• Design(s) to demonstrate clinical utility of the first companion diagnostic test

Using the final In-Vitro Diagnostic (IVD) version of the test to select patients

Bridging from Clinical Trial Assay (CTA) or Lab Developed Test (LDT) to the final IVD Test

• Follow on companion diagnostic test : Re-test samples from a previously conducted trial for first

companion Diagnostic

-samples may not be available Conduct a new prospective trial with the same drug with a

placebo arm

- unethical to give placebo to patients while an effective approved drug is available.

3

Approach for Follow on Diagnostic Test and NDMC Assumption

1. Test Samples from another clinical study cohort by– The First companion Diagnostic Test – the follow on companion diagnostic test, and– the reference method (A sequencing method)

2. Calculate agreement between follow on companion diagnostic test and the other two tests.

3. “Transport” Results of drug efficacy from the pivotal study for the first companion diagnostic test to the follow on companion diagnostic test assuming Non-differential Misclassification (NDMC).

4

Background: Cetuximab for mCRC patients Pivotal Clinical Study: KRAS Mutation

5

572 Patients with Advanced

Colorectal Cancer

Cetuximab + BSC

Best Supportive Care (BSC)

Tested by Sanger Sequencing

KRAS Mutation Detected(MD or POS)

KRAS Wild-Type (WT or NEG)

HR- 0.98 (0.70, 1.37)No Effect of Treatment

HR – 0.55 (0.41,0.74) Treatment Effective

*Karpetis et. al. NEJM 2008

NDMC Assumption: Given the comparator method result, clinical efficacy is assumed not to depend on the cobas® KRAS Mutation Test result Sanger Sequencing

cobas® KRAS Test

S = 0 (Neg)

S = 1(Pos)

R = 0 (Neg)

δ00 δ01

R = 1(Pos)

δ10 δ11

δ.0 δ.1

6

Sanger Sequencing

cobas® KRAS Test

S = 0 (Neg)

S = 1(Pos)

R = 0 (Neg)

δ.0 δ.1

R = 1(Pos)

δ.0 δ.1

When Clinical outcome is available

Applying NDMC Assumption

δr. = δr0 (1- πr ) + δr1 πr = (NDMC) δ.0 (1- πr ) + δ.1 πr

Here , π1 = PPV and π0 = 1-NPV; δ1. = Log Hazard Ratio in R=1(Pos); δ0. = Log Hazard Ratio in R=0 (Neg )

Background: Cetuximab Study and FDA-Approved KRAS Test

7

572 Patients with Advanced

Colorectal Cancer

Cetuximab + BSC

Best Supportive Care (BSC)

Tested by FDA-Approved KRAS test

KRAS Mutation Detected(MD)

KRAS Wild-Type (WT)

HR- 0.91 (0.67, 1.24)No Effect of Treatment

HR – 0.63 (0.47,0.84) Treatment Effective

8

XELOXA Samples + supplemental mCRC samples (n=461)

Sanger Sequencing

cobas® KRAS Test

Calculate NPV+PPV-1

Calculate NPV+PPV-1

• Under NDMC: E(h|R=0) –E(h|R=1) = [E(h|S=0) – E(h|S=1)( NPV+PPV-1)

• difference in log-hazard ratio for cobas test = difference in log-hazard ratio for Sanger ×(NPV+PPV-1)

NDMC Assumption Implication and Analysis (Criterion 1 and 2)

NPV+PPV-1 is defined as the “attenuation factor”

FDA Approved Test

cobas® KRAS Test

Attenuation Factor (NPV+ PPV-1) calculations

9

Here, τ is the prevalence of Pr(S=1) in the Pivotal Study, and

Comparison of the cobas® KRAS Mutation Test with Comparator Methods for Detection of KRAS Mutations in Codon 12/13

cobas® KRAS

Mutation Test

Comparator Method

Sanger Sequencing FDA-approved IVD test

MD NMD Invalid Total MD NMD Invalid Total

MD 124 34 5 163 139 9 15 163

NMD 4 268 2 274 10 248 16 274

Invalid 0 19 5 24 0 5 19 24

Total 128 321 12 461 149 262 50 461

PPA(95% CI) 96.9% (92.2%, 98.8%) 93.3% (88.1%, 96.3%)

NPA(95% CI) 88.7% (84.7%, 91.8%) 96.5% (93.5%, 98.1%)

10

Attenuation Factor (NPV+ PPV-1) calculations

for cobas® test and FDA Approved Test

11

Comparator PPV

(95% CI)

NPV

(95% CI)Attenuation Factor

(95% CI)

Sanger Sequencing

0.858

(0.811, 0.902)

0.975

(0.946, 0.994)

83.3%

(77.7, 88.3)

Table 1: Attenuation Factors for cobas® KRAS mutation test

Table 2: Attenuation Factor for FDA Approved test

Comparator PPV

(95% CI)

NPV

(95% CI)Attenuation Factor

(95% CI)

Sanger Sequencing

0.840

(0.790, 0.888)

0.956

(0.918, 0.986)

79.5%

(73.4, 85.2)

Comparator PPV

(95% CI)

NPV

(95% CI)Attenuation Factor

(95% CI)

FDA Approved

Test

0.957

(0.927, 0.981)

0.945

(0.909, 0.978)

90.2%

(85.6, 94.4)

Table 3: Attenuation Factor with respect to FDA Approved test

Five Criteria to Establish Clinical Utility of cobas® KRAS Test

1. NDMC (Non-Differential Misclassification) criterion for cobas® Test

2. NDMC (Non-Differential Misclassification) criterion for FDA Approved Test

3. Influence Condition Evaluation

4. Covariate Assessment

5. Sensitivity Analysis

12

Influence Condition Evaluation: (3)

Evaluate Influence Condition: To enable Overall population labeling, the beneficial effect of the drug must not be limited to only the predefined subpopulation

– Our Objective is to show that for this study overall Population labeling does not apply; Influence condition is false

– i.e. 95% CI for the hazard ratio in the Mutation positive subset includes 1, and

– the 95% CI for the hazard ratio in the Mutation Negative subset excludes 1.

13

Influence Condition Evaluation:

• Log Hazard Ratios for cobas® test can be calculated using following relationships between hazard ratios based on NDMC assumption

-For Mutation Negative subsetδ0. = δ00 (1- π0 ) + δ01 π0 = (NDMC) δ.0 (1- π0 ) + δ.1 π0

Here , π0 = 1-NPV

-For Mutation Positive subsetδ1. = δ10 (1- π1 ) + δ11 π1 = (NDMC) δ.0 (1- π1 ) + δ.1 π1

Here , π1 = PPV

14

Influence Condition Evaluation Results

Drug Efficacy

cobas®

KRAS Mutation Test

Status

Samples Tested

(N)

Hazard Ratio (HR)

Estimate 95% CI

Overall Survival (OS)

No Mutation Detected 272 0.558 (0.422, 0.752)

Mutation Detected 158 0.908 (0.670, 1.209)

Progression Free Survival

(PFS)

No Mutation Detected 272 0.413 (0.304, 0.550)

Mutation Detected 158 0.869 (0.670, 1.138)

15

Covariates Comparison between the two studies: Patient Characteristics (4)

Characteristic Study Cohort (N=437)

Pivotal Clinical study(N=453) P-value*

Sex       N 437 453   Female 204 (46.7%) 153 (33.8%) P1<0.0001 Male 233 (53.3%) 300 (66.2%)  Race       N 437 453   White 391 (89.5%) 414 (91.4%) P1=0.151 Non-White 46 (10.5%) 39 (8.6%)  Baseline ECOG       N 421 453   0 285 (67.7%) 110 (24.3%) P1<0.0001 1 125 (29.7%) 245 (54.1%)   2 11 (2.6%) 98 (21.6%)  Age       N 437 453   Median 61.0 63.2 P2=0.004 Min - Max 26.0 - 89.0 28.6 – 88.1  BSA       N 361 453   Median 1.9 1.8 P2<0.0001 Min - Max 1.3 - 2.8 1.3 – 2.5  * P1=p-value from Chi-Square Goodness-of-Fit Test; P2=p-value from 2-sided one sample Sign Test.Note: ECOG=Eastern Cooperative Oncology Group; BSA= Body surface area BSA=[(height in cm*weight in Kg)/3600]1/2

16

Covariates Comparison between the two studiesDisease Characteristics (4)

Disease Characteristics  Study Cohort (N=437)

Therascreen (N=453) P-valuea

Disease Stage       Duke’s Stage Total 430 59   A 0 (0.0%) 1 (1.7%) P1<0.0001 B 0 (0.0%) 16 (27.1%)   C 363 (84.4%) 38 (64.4%)   D 67 (15.6%) 4 (6.8%)  Tumor Type      N 432 448   Primary 420 (97.2%) 410 (91.5%) P1<0.0001 Metastatic 12 (2.8%) 38 (8.5%)  a; P1=p-value from Chi-Square Goodness-of-Fit Test.

17

Covariates Comparison between the two studiesSample Characteristics (4)

 Study Cohort

(N=437)

Pivotal Clinical study

(N=453) P-valuea

Tumor Content in Sample       N 437 453   Median 35.00 47.25 P2<0.0001 Min - Max 5.0 - 90.0 1.0 – 100.0  Macro Dissection of Samples       N 437 453   Tumor Content ≤20 98 (22.4%) 39 (8.6%) P1<0.0001 Tumor Content >20 339 (77.6%) 414 (91.4%)  Necrosis Score Within Tumor Area

     

N 437 453   0 - < 10% 304 (69.6%) 346 (76.4%) P1<0.0001 10 - 50% 131 (30.0%) 86 (19.0%)   >50% 2 (0.5%) 21 (4.6%)  KRAS Mutation Type       N 149 208   12ALA 12 (8.1%) 14 (6.7%) P1=0.317 12ARG 2 (1.3%) 2 (0.9%)   12ASP 42 (28.2%) 71 (34.1%)   12CYS 15 (10.1%) 16 (7.7%)   12SER 13 (8.7%) 11 (5.3%)   12VAL 35 (23.5%) 54 (25.9%)   13ASP 30 (20.1%) 40 (19.2%)  a; P1=p-value from Chi-Square Goodness-of-Fit Test; P2=p-value from 2-sided one sample Sign Test..

18

Hazard ratio Estimates for Significant CovariatesIf the covariate distribution was similar to observed in the original study

19

  Hazard Ratio (HR)  Wild Type Mutation Detected

Covariates Drug Efficacy Estimate 95% CI Estimate 95% CIAge OS 0.554 (0.426,0.730

)0.907 (0.667,1.207)

PFS 0.413 (0.308,0.555)

0.872 (0.660,1.156)

BSA OS 0.563 (0.416,0.760)

0.902 (0.681,1.202)

PFS 0.416 (0.313,0.560)

0.874 (0.667,1.142)

Duke’s Stage (<=C, >C)

OS 0.559 (0.422,0.757)

0.903 (0.678,1.217)

PFS 0.412 (0.311,0.552)

0.866 (0.656,1.131)

Baseline ECOG OS 0.563 (0.414,0.771)

0.904 (0.675,1.214)

PFS 0.407 (0.302,0.551)

0.860 (0.646,1.137)

Necrosis

(0 - <10%, 10 – 50%, >50%)

OS 0.562 (0.420,0.753)

0.890 (0.686,1.180)

PFS 0.408 (0.303,0.545)

0.844 (0.659,1.117)

Sex OS 0.552 (0.428,0.734)

0.898 (0.662,1.195)

PFS 0.408 (0.311,0.545)

0.867 (0.662,1.141)

Tumor Type OS 0.549 (0.419,0.726)

0.980 (0.699,1.335)

PFS 0.401 (0.299,0.531)

0.972 (0.726,1.326)

Tumor content (<=20 or >20)

OS 0.563 (0.425,0.755)

0.912 (0.677,1.214)

PFS 0.416 (0.308,0.554)

0.889 (0.665,1.187)

Tumor content (Num) OS 0.556 (0.405,0.734)

0.901 (0.673,1.221)

PFS 0.408 (0.306,0.541)

0.877 (0.668,1.149)

Sensitivity Analysis* (5)

• Sensitivity analysis was conducted to consider the robustness of the study results to the assumptions by simulating how many agreements between cobas® test and Sanger sequencing would have to be changed to disagreements before the study fails to show clinical effectiveness.

20

Sanger Sequencing

cobas® KRAS Test

Pos Neg

Pos a b

Neg c dk k

• ‘k’ patients will be randomly selected from ‘a’ cell and their status will be changed to Sanger Pos, cobas test Negative, similarly k’ patients will be randomly selected from ‘d’ cell and their status will be changed to Sanger Negative cobas test Positive. Estimate of Log Hazard ratio calculated for each value of ‘k’.

• The highest value of k at which the hazard ratio is still statistically significant will be determined. *Denne, Pennello et al. 2014, Statistics in Biopharmaceutical

Research

OS (HR) Changes by KRAS Status as Determined by the cobas® KRAS Mutation Test by Moving Subjects from Concordance to Discordance (Criterion 5)

21

when k = 45, which corresponds to 21% more discordance between the cobas® KRAS Mutation Test and Sanger sequencing for Mutation Positive subset

PFS (HR) Changes by KRAS Status as Determined by the cobas® KRAS Mutation Test by Moving Subjects from Concordance to Discordance (5)

when k = 27, which corresponds to 12.6% more discordance between the cobas® KRAS Mutation Test and Sanger sequencing in Mutation Positive subset

Five Bridges to Demonstrate Clinical Utility

23

XELOXA +

Supplemental samples

Cetuximab CO.17 Trial results

NPV+PPV-1 for cobas® test >83%

NPV+PPV-1 for cobas® test > NPV+PPV-1 for FDA

Approved Test

Influence condition is false

Covariate Assessment

Sensitivity Analysis

Acknowledgements

• Interactions with FDA

John Palma, Lesley Farrington, Allison Gannon, Tori Brophy, Sung Lee

• Clinical Study Coordination , conduct,

Karen Yu, Melody Chee, Sim Truong

• Clinical Study decisions, CSR and research

Sid Scudder

• Statistical Analysis

Guili Zhang, Shagufta Aslam, Ranga Yerram and SAS programming team

COBAS is a trademark of Roche. 24

Doing now what patients need next

25