1. Data Standards for Pharmacometric Analysis Data Sets
Neelima Thanneer
Bristol-Myers Squibb 1

2. Source data issues and Standard Rules for Imputation
As far as possible imputations should be avoided, but they may be necessary particularly for Population PK datasets as complete dosing history is generally not available in source data.
Dose date/time imputation typically occurs in below scenarios:
IV dosing where every dose date, time should be recorded
Oral doses where interval dosing is recorded

3. Standard Rules for Imputation
Dose clock time imputation for CRF designs where every single dose date and time is captured
If a dose date is available but time is missing:
If a trough sample was taken on the same day, use the trough time as the dose time (if IV use start time of infusion). One can add 5 min
If no troughs were taken, use the previous dose time. For BID/TID adjust the imputation based on frequency.
If the first dose time is missing then impute using day 1 lab time or impute based on post first dose sample time. For BID/TID adjust the imputation based on frequency.
The imputed clock time records should be flagged
Account for missing doses/dose interruptions based on the number of tablets or comment in text like variable

4. Standard Rules for Imputation
Infusion Duration Imputation
If infusion stop time is available but infusion start time is missing, the protocol defined duration (e.g. 1 hour or 30 min) is used to determine the start of infusion and vice versa if stop time is not available.
If both infusion start and stop times are missing on day 1, pre- dose sample time or end of infusion sample time is used along with nominal infusion duration to determine the start of infusion time
The imputed date/time records should be flagged

5. Standard Rules for Imputation
When CRF is designed to capture interval doses with start and stop dates recorded and only dose times relative to PK are recorded
Variables ADDL and II are derived to capture the not recorded doses
ADDL: Number of additional doses exactly like the current one
II: Interdose Interval
If the dose time relative to PK samples are not recorded then impute using the IV imputation rules
ADDL needs to be adjusted based on the recorded time deviations.

6. Influencing CRF Design
CRF’s are not consistently designed to capture the information need for Pharmacometric analysis
For BID and TID CRF should capture the prior 2 and 3 dose times relative to PK.
First dose time should be collected
Dose interuptions for BID and TID doses should indicate which doses are missing
Pharmacometricians and programmers can influence the CRF design by getting involved at the design stage.
For oral studies where patient takes drug at home CRF should be designed to collect the dose date time relative to the PK sample drawn.

7. Population PK Data Standards Initiative Sponsored by International Society of Pharmacometrics (ISoP)
Group of around 20 enthusiastic pharmacometricians and data programmers has come together at the beginning of 2016 to address the outlined situation and work on its improvement.
With the sponsorship of ISoP, the group envisions the development of PPK data standards for interchange and analysis. Some of the benefits of the work will directly reflect on improved:
Consistency and efficiency in pharmacometric datasets
Quality (fewer errors, enabling development of open-source tools for automatic data checking)
Enabling development of open-source tools for automatic graphical exploration
Regulatory compliance and audit readiness (enabling development of open-source tools for data submission)
Population PK analyses are most commonly included component of regulatory filings but the datasets for such analyses are prepared and documented inconsistently

8. Objective: Standardized datasets used directly by analysis software
Source data
Standardized dataset
SDTM
Population PK standard dataset
Analysis
ADaM
NONMEM, Monolix, R, Stan, Julia, etc can use standardized datasets directly
Already the case for R and some other tools
Common variable names
Common structure
Will always look the same
Suitable for collaboration
Typically SAS datasets
Radivojevic, et al. ACoP 2016

9. What we have done Draft standard for input data
Classified by information group
Source dataset and variable name
Required input
from the pharmacometrician?
CDISC naming conventions and length requirements
Imputation and derivation guidance
Controlled terminology
General comments

10. What we have done Current status
Path to CDISC data standards for pharmacometrics
hand-over of solid draft data structure
review by CDISC teams and feasibility assessment
implementation guide
Preparing a “Perspectives” article to outline our thinking and objectives

11. For any other stakeholders
Next steps
Scripts for further conversion into tool dialects
Once the standard is established, capture the deliverable in a “White paper”
Longer term:
Start expanding to PKPD data
Develop and publish automation scripts
For this initiative to succeed, it must make things faster, easier and more efficient
For modelers
For programmers
For any other stakeholders