1. Safety analysis of clinical trials in NDA submissions JSM 2018, Jul
Safety analysis of clinical trials in NDA submissions JSM 2018, Jul. 28 – Aug. 2 Linyun Zhou

2. Outline Modules in NDA (BLA) Submission Document (eCTD)
Core Safety Analysis in NDA (BLA/MAA/ROW) Submissions
Compound Specific Safety Analysis
Challenges and Research Areas

3. Modules in NDA Submission Document (eCTD) (FDA website)
The eCTD contains an electronic table of contents also referred to as a backbone that manages all the metadata for an application.
This backbone is broken down into five modules. Documents are placed appropriately into modules, which are graphically presented as the CTD Triangle.
Module 1 references regional information such as forms, cover letter, labeling, and investigational brochures.
Module 2 references summaries such as quality, clinical, and non-clinical summaries.
Module 3 references quality information.
Module 4 references non-clinical information.
Module 5 references clinical information.
The next screen shows the CTD Triangle.

4. Modules in NDA Submission Document (eCTD) (FDA website)
2.5
2.7.4

5. Modules in NDA Submission Document (eCTD) (FDA website)
Section 2.5 (~30 pages)

6. Modules in NDA Submission Document (eCTD) (FDA website)
Section (~50 – 400 pages)

7. Modules in NDA Submission Document (eCTD) (FDA website)
section (no space limitation)

8. Core Safety Analysis in NDA Submissions
1. Disposition, Exposure, Demographic
2. AE - TEAE/TRAE, SAE, AE leading to discontinuation, Most Common AE, AE special interest, AE by severity or relationship, Time to Event , AE by time interval
3. Labs - Mean Change From Baseline, Markedly Abnormal Laboratory Values, Shift from Normal to High or Low
4. Vital Signs, ECGs, Others - Mean Change From Baseline, Markedly Abnormal Laboratory Values.
5. Sub-group analysis – selected main table by Demog variables, MHs, CMs …

9. for an NDA (BLA/MAA/ROW) submission?
Are these enough
for an NDA (BLA/MAA/ROW) submission?

10. Compound Specific Safety Analysis
Hepatic
Renal
Hematology CV ….

11. Compound Specific Safety Analysis
Hepatic
Summary table for elevated liver function test values - ALT, AST and Total Bili
Shift scatter plot of max postbaseline vs baseline for ALT
eDISH plot: Total Bili vs ALT. One time point per subject using the common algorithm
Patient profile graphs for subjects based on eDISH plot
Shift in LFTs by baseline values 

12. Hepatic - Compound Specific Safety Analysis Summary table for elevated liver function test values - ALT, AST and Total Bili

13. Hepatic - Compound Specific Safety Analysis eDISH plots
ALT (xULN)
Drug A Drug B Drug C Drug D

14. Hepatic - Compound Specific Safety Analysis Patient profile graphs
HbA1C
LFT tests
Drug A
AEs

15. Hepatic - Compound Specific Safety Analysis Shift in LFTs by baseline values
During Treatment

16. Compound Specific Safety Analysis
Renal
Shift scatter plot of max postbaseline vs baseline for Creatinine. Lines for 0%, 30% and 50% increase
Summary table of Creatinine Increase From Baseline at max Post-Baseline
Patient profile graphs for subjects with 50% increase in creatinine
Efficacy vs Safety

17. Renal – Compound Specific Safety Analysis Shift scatter plot of max postbaseline vs baseline for Creatinine. Lines for 0%, 30% and 50% increase
Peak Post-Baseline
Baseline

18. Renal - Compound Specific Safety Analysis
Summary table of Creatinine Increase From Baseline at max Post-Baseline
p-values
Drug A/B/C vs
Placebo
Placebo
Drug A
Drug B
Drug C

19. Renal – Compound Specific Safety Analysis Efficacy vs safety
SBP
sCr

20. Compound Specific Safety Analysis
Hematology
Shift scatter plot of min postbaseline vs baseline for Hemoglobin by gender
Shift scatter plot of max postbaseline vs baseline for WBC
Shift scatter plot of min postbaseline vs baseline for WBC
Shifts from Baseline in Laboratory Variables Relative to Normal Ranges  

21. Hematology - Compound Specific Safety Analysis Shift scatter plot of min postbaseline vs baseline for Hemoglobin by gender
Min Post-Baseline
Baseline

22. Compound Specific Safety AnalysisCV
HR/RR/OR and 95% CI Tables and Graphs (RR, FDA CV guidance 2009)
Shift scatter plot of max postbaseline vs baseline for CK (no CPK)
Scatter plot of max postbaseline CK vs AST/uln at same timepoint
Shifts from Baseline in Laboratory Variables Relative to Normal Ranges

23. CV - Compound Specific Safety Analysis HR and 95%CI
Favor Drug A Favor Drug B

24. Challenges Rare Events Adjudicated Events Events vs Subjects
Recurrent Events
Concurrent Events
Clinical results vs post-market results
Real-World Evidences
Safety signal detection
Benefit vs Risk
Meta-analysis …

25. Favor Drug A Favor Drug B
Safety Signal Detection - AE screening
P-risk plot for all adverse events of Drug A vs Drug B
Favor Drug A Favor Drug B
1 page plot vs 200+page Table
Risk Difference (%)

26. Drug A Drug B Drug C Drug D
Safety Signal Detection - AE Duration
Drug A Drug B Drug C Drug D

27. Safety Signal Detection – Time to Event
Event Onset

28. Benefit vs Risk – Mean difference

29. Meta analysis - Brief Overview of Key Effect Sizes
Studies
Placebo
Drug A
Drug B
40 mg
80 mg 01 63 62 64 02
142
281
284
282 03
155
280
283
567 04
294
293 05
327
329
326

30. Meta analysis - Brief Overview of Key Effect Sizes
A simple 2x2 table for ith study
Event
No-Event
Total
Drug A Ai Bi
Ai + Bi
Drug B Ci Di
Ci + Di
Ai + Ci
Bi + Di
Ai+Bi+Ci+Di

31. Meta Analysis - Brief Overview of Key Effect Sizes Example: Drug A vs Drug B Cardiovascular Adverse Events
Studies
Drug B
Drug A
OR (95% CI)
RR (95% CI)
RD (95% CI) N
Event 01 63
126 1
0.01 ( -0.01, 0.02) 02
282
565
0.50 ( 0.03, 7.99)
0.50 ( 0.03, 7.95)
-0.00 ( -0.01, 0.01) 03
567 2
563
1.01 ( 0.14, 7.17)
1.01 ( 0.14, 7.12)
0.00 ( -0.01, 0.01) 04
293 6
587
0.49 ( 0.16, 1.55)
0.50 ( 0.16, 1.53)
-0.01 ( -0.03, 0.01) 05
326
656 4
0.99 ( 0.18, 5.45)
0.99 ( 0.18, 5.40)

32. Meta Analysis - Brief Overview of Key Effect Sizes
A simple 2x2 table for ith study
Event
No-Event
Total
Drug A Ai Bi
Ai + Bi
Drug B Ci Di
Ci + Di
Ai + Ci
Bi + Di
Ai+Bi+Ci+Di

33. Meta Analysis - Brief Review of Standard/Common Methods (1)
Mantel-Haenszel Method for performing a meta-analysison on odds ratios
Assuming the data from each study are presented as 2×2 tables, for ith study Ai (Bi )and Ci (Di )are event (non-event) counts for treatment and control group respectively, then the odds ratio from ith study is defined as
𝑌 𝑖 = 𝐴 𝑖 𝐷 𝑖 𝐵 𝑖 𝐶 𝑖
the Mantel-Haenszel Method assigned the weight for ith study as
𝑊 𝑖 = 𝐵 𝑖 𝐶 𝑖 𝐴 𝑖 + 𝐵 𝑖 + 𝐶 𝑖 + 𝐷 𝑖

34. Meta Analysis - Brief Review of Standard/Common Methods (2)

35. Meta Analysis - Brief Review of Standard/Common Methods (3)
Mantel-Haenszel Method for performing a meta-analysison on odds ratios (cont’d)
𝐹 𝑖 = 𝐴 𝑖 + 𝐷 𝑖 𝐵 𝑖 𝐶 𝑖 𝑛 𝑖 2 , 𝐺 𝑖 = 𝐵 𝑖 + 𝐶 𝑖 𝐴 𝑖 𝐷 𝑖 𝑛 𝑖 2 , 𝐻 𝑖 = 𝐵 𝑖 + 𝐶 𝑖 𝐵 𝑖 𝐶 𝑖 𝑛 𝑖 2 ,
then 𝑉 𝑙𝑛𝑂 𝑅 𝑀𝐻 = 𝐸 𝑖 𝑅 𝑖 𝐹 𝑖 𝐺 𝑖 𝑅 𝑖 × 𝑆 𝑖 𝐻 𝑖 𝑆 𝑖 2
(all summation are from i=1 to k), the 95% CI for ln( 𝑂𝑅 𝑀𝐻 ) is estimated as
ln( 𝑂𝑅 𝑀𝐻 )± 𝑉 𝑙𝑛𝑂 𝑅 𝑀𝐻
and the z-score is estimated by
𝑍= ln( 𝑂𝑅 𝑀𝐻 ) 𝑉 𝑙𝑛𝑂 𝑅 𝑀𝐻

36. Meta Analysis - Brief Review of Standard/Common Methods (4)
One-step (Peto) formula for odds ratios
The log odds ratio for ith study is estimated as
𝑌 𝑖 = 𝐴 𝑖 − 𝐸 𝑖 𝐼 𝑖
where
𝐸 𝑖 = 𝐴 𝑖 + 𝐵 𝑖 × 𝐴 𝑖 + 𝐶 𝑖 𝐴 𝑖 + 𝐵 𝑖 + 𝐶 𝑖 + 𝐷 𝑖
and
𝐼 𝑖 = 𝐴 𝑖 + 𝐵 𝑖 × 𝐶 𝑖 + 𝐷 𝑖 × 𝐴 𝑖 + 𝐶 𝑖 × 𝐵 𝑖 + 𝐷 𝑖 ( 𝐴 𝑖 + 𝐵 𝑖 + 𝐶 𝑖 + 𝐷 𝑖 ) 2 × 𝐴 𝑖 + 𝐵 𝑖 + 𝐶 𝑖 + 𝐷 𝑖 −1

37. Meta Analysis - Brief Review of Standard/Common Methods (5)

38. Brief Review of Standard/Common Methods (6)
One-step (Peto) formula for odds ratios (cont’d) the 95% CI for ln 𝑂𝑅 𝑜𝑛𝑒𝑠𝑡𝑒𝑝 is estimated as ln 𝑂𝑅 𝑜𝑛𝑒𝑠𝑡𝑒𝑝 ±1.96 𝑉 𝑙𝑛𝑂𝑅 𝑜𝑛𝑒𝑠𝑡𝑒𝑝 and the z-score is estimated by 𝑍= ln 𝑂𝑅 𝑜𝑛𝑒𝑠𝑡𝑒𝑝 𝑉 𝑙𝑛𝑂𝑅 𝑜𝑛𝑒𝑠𝑡𝑒𝑝

39. Meta Analysis - Brief Review of Standard/Common Methods (7)
Random-effect model using inverse variance method to estimate the mean effect size
𝑀 ∗ = 𝑘=0 𝑛 𝑊 𝑖 ∗ 𝑌 𝑖 𝑖=1 𝑘 𝑊 𝑖 ∗ , 𝑊 𝑖 ∗ = 1 V 𝑌 𝑖 ∗ , 𝑉 𝑖 ∗ = V 𝑌 𝑖 + ԏ 2
where V 𝑌 𝑖 is the within study variance for study 𝑌 𝑖 and ԏ 2 is the cross studies variance. A special situation for this model is to ignore the cross studies variance, and get so-called fixed effect model.

40. Meta Analysis - Brief Review of Standard/Common Methods (8)
There are various methods to estimate ԏ 2 , one of them is the method of moment (the DerSimonian and Laird):
ԏ 2 = 𝑖=1 𝑘 𝑊 𝑖 − 𝑖=1 𝑘 𝑊 𝑖 𝑌 𝑖 𝑖=1 𝑘 𝑊 𝑖 −𝑘 𝑊 𝑖 − 𝑊 𝑖 2 𝑊 𝑖

41. Meta Analysis - Brief Review of Standard/Common Methods Example: Drug A vs Drug B Cardiovascular Adverse Events
METHOD
OR (95% CI)
Mantel-Haenszel
0.71 ( 0.32, 1.58)
Peto
0.70 ( 0.31, 1.60)
Inverse Variance
0.66 ( 0.29, 1.48)
DerSimonian and Laird

42. Thanks!