1. Challenges and Strategies in Pharmacometric Programming
Jing Su
Merck & Co., Inc.

2. PKPD Programming Forum
Sub-team under the Statistics and Pharmacometrics Interest Group (SxP)
Objectives
Establish a community for pharmacometric and pharmacokinetic programmers across industry and academics 
Collect and share best practices, challenges and examples for programmers to support pharmacometric and PK analysis
Share ideas of training new PK/PD programmers

3. Deliverables

4. More about Deliverables
Answer questions for:
Target selection
Compound selection
Go/no-go (discovery and clinical)
Dose optimization through exposure-response analysis (efficacy and safety) from phase I through III
PK characterization including dose recommendations for sub-populations (e.g. renal or hepatic impairment, obese patients, pediatric patients)
Immunogenicity evaluations
Study Design
Internal Decision Making
Drug Submission and Label
External Publications

5. PKNCA (ADPC) Data Support PK non-compartmental analysis (NCA) Include:
SUBJID
DAY
PTALD
CONC
DOSE
PKDTC
LSDTC 1
100
T07:30:00
1.5
T09:00:00
T08:00:00 2 3
T10:00:00 4
4.2
T12:00:00 12
3.6
T20:00:00
2.7
T07:55:00
6.5
T12:00:00
T08:00:00
3.9
T07:55:00
7.8
T12:00:00
T08:00:00
T07:55:00
8.1
T12:00:00
T08:00:00 5
4.3
T07:55:00
8.2
T12:00:00
T08:00:00
Include:
PK concentration
PK time
Dosing time
Dose amount
Covariates
Derived variables: actual time since first dose and time since last dose prior to the sample, etc.

6. Time since Last Dose is Crucial to PK Analysis
Correct time since last dose prior to PK sample depends on accuracy of both PK sample and dosing date and time
Incorrect time since last dose can cause incorrect analysis and result interpretation
In this example, correct Cmax occurred at 4hr post dose
If the time of the Cmax sample is incorrectly recorded as 6hr post dose, the Cmax value would appear later than expected. If the time of dosing is incorrectly recorded before the actual time, trough sample would become post dose sample whose value would be less than expected, and Cmax will appear later than expected.
Correct data point
incorrect data point
Incorrect dosing time

7. NONMEM PopPK Data In addition to PKNCA data, include
dosing records (EVID=1)
NONMEM reserved variables
other covariates
Ordered by subjid and time
CSV file format is required
Dosing records:
actual dosing date and time, and actual dose amount are needed for every dose event
Imputation is required when actual dosing time and dose amount are not collected for every dose by design. For example, oral daily dose late stage studies.
SUBJID
DAY
PTALD
CONC
ATSFD
EVID
AMT 1 .
100
1.5 2 3 4
4.2 12
3.6
2.7 24
6.5 28
3.9 48
7.8 52 72
8.1 76

8. Challenges Source data related Programming process related
Cross-functional collaboration
Multiple data sources and data formats
Individualized PopPK data specification
Lack of standard on raw PK data
Imputation on missing dosing time
Evolving data collection/mapping standards
Missing or incorrect time
Quick turn-around

9. Challenges with Source Data
Multiple data sources
Raw PK concentration data and other clinical data not in the same database
No consistent common variables for merging
Difficult to check integrity between two databases
Lack of standard on raw PK data in terms of file structure and/or variable names
Evolving data collection/mapping standards over years
Missing or incorrect PK sampling time and/or dosing time, especially for late stage studies

10. Challenges – Pm Programming Process
Cross-functional collaboration: PKPD programmers work with both pharmacometrician and biostatistician.
Communicate with pharmacometrician to get analysis/modeling purpose and programming requirements
Communicate with biostatistician to identify correct clinical data source based on modeling purpose, and get details about clinical data to ensure consistency between statistical analyses and Pm modeling
Individualized NONMEM PopPK data specification
Lack of standard and clarity
Hard to program, validate, review and share
Quick turn-around after DBL
Imputation on missing dosing time

11. Strategies Standardization Quality improvement Automation
Standardize source PK data – follow SDTM standards
Follow ADaM requirements for PKPD data if possible
Develop PopPK data standards – an ISoP initiative
Create guidelines on missing time imputation
Quality improvement
PK reconciliation/cleaning to improve source data quality
Streamline analysis/modeling dataset validation process
SAS outputs based on pre-defined validation checklist
Automation
More SAS macros to improve efficiency, reduce validation needs and enforce standardization
Build talent and keep experienced Pm programmers

12. Challenges = Opportunities

13. Acknowledgments
Bret Musser, Regeneron Qi Shen, Merck & Co., Inc. David Turner, Merck & Co., Inc.