1. The material in this part of the presentation is prepared using
Hands on session
Pravin Jadhav, Pharmacometrics
The material in this part of the presentation is prepared using
NONMEM help guides
ACCP website (
October 09, 2008
Hands on session

2. Where to find help to learn NONMEM software?
Depending on your installation version
Version VI- C:\nmvi\guides and C:\nmvi\help
Version V- C:\nmv\guides and C:\nmv\help
NMUSERS
Listserve:
ACCP website
NONMEM VI features
October 09, 2008
Hands on session

3. Introduction to NONMEM
What is NONMEM?
A computer program to analyze data using a NONlinear Mixed Effects Model
Focus on Population Pharmacokinetic data, however, can handle several other problem sets as well
Developed by the NONMEM Project Group at the University of California, San Francisco (UCSF)
Written in ANSI FORTRAN 77
Two NONMEM programs available- SINGLE and DOUBLE (default) precision
October 09, 2008
Hands on session

4. File of NM-TRAN records Several Input files for NONMEM
NONMEM Architecture
File of NM-TRAN records
Begins with “$”
Data File
ASCII format
Two major components are distributed with NONMEM
NM-TRAN (NONMEM Translator)
a preprocessor to NONMEM that translates inputs specified in a more user-friendly way to the formats required by NONMEM
PREDPP (“PRED for Population Pharmacokinetics”)
Specialized PRED subroutine for use with NONMEM for pharmacokinetic data
Collection of FORTRAN subroutines
NM-TRAN
Several Input files for NONMEM
NONMEM
PREDPP
ADVAN and TRANS PK
ERROR
NONMEM Output
October 09, 2008
Hands on session

5. Model input file (NM-TRAN Control Records)
Independent variable
Dependent variable
Covariates
NM-TRAN Control Language Translation service provides flexible order of these records and allows abbreviations of the record names, for example, $EST ($ESTIMATION), $COV ($COVARIANCE) or option names METH (METHOD), SIG (SIGDIGITS) etc.
No “Tabs” in NONMEM! ASCII format only
October 09, 2008
Hands on session

6. Data file (passed to NM-TRAN preprocessor)
Data item
Data record
When using PREDPP, data preprocessor can supply missing items such as MDV even if user does not supply in dataset/$INPUT record
Note:
TIME should be sorted chronologically and records are added according to events.
The dose (AMT) is specified at the actual dosing time. Don’t confuse with DOSE column that is intended to be used as a covariate
October 09, 2008
Hands on session

7. Building a control stream and data file
Record 1- description of the problem
Record 2- location of the dataset. If not the same directory, requires file path
NONMEM VI incorporates
More flexible use of IGNORE option- For example, IGNORE=(BW.GT.100)
Adds ACCEPT option- For example, ACCEPT=(BW.LE.100)
October 09, 2008
Hands on session

8. Building a control stream and data file
Record 3- names of data items (in the order) on each data record
Quick Notes:
Names of the data items in the $INPUT record are the ones supplied to NMTRAN
Remember what IGNORE option does?
October 09, 2008
Hands on session

9. Building a control stream and data file
ID- Identification number
TIME- Time associated with the event described in the data record
AMT- Dose Amount (AMT) data item for PREDPP
Dose given at the specified TIME
CONCNGML- Dependent Variable (DV)
observed concentration of drug in ng/mL
EVID- Event Identification data item for PREDPP (takes values 0, 1, 2, 3, 4)
MDV- Missing Dependent Variable data items for NONMEM (takes values 0, 1)
October 09, 2008
Hands on session

10. Building a control stream and data file
What if an additional of 50 mg was given at 12 hr?
Method 1:
Explicitly specify dosing at the specified TIME
Need additional record if dosing and observation occurred at identical times
Observed
Concentrations
October 09, 2008
Hands on session

11. Building a control stream and data file
Method 2:
Use of ADDL (Additional Dose data item for PREDPP) and II (Interdose Interval data item for PREDPP)
The units of II should be consistent with those of TIME
If II value contains a colon (:), it is assumed to be a clock time (hh:min)
October 09, 2008
Hands on session

12. Building a control stream and data file
Record 4: the pharmacokinetic model to be used (PREDPP)
PREDPP subroutines
ADVAN1: One compartment model with intravenous administration
ADVAN2: One compartment model with first order administration
ADVAN3: Two compartment model with intravenous administration
ADAVN4: Two compartment model with first order administration
ADVAN5 & 7: General linear models
ADVAN6, 8 & 9: General nonlinear models (Differential equations)
ADVAn10: One Compartment Model with Michaelis-Menten Elimination
ADVAN11: Three compartment model with intravenous administration
ADVAN12: Three compartment model with first order administration
October 09, 2008
Hands on session

13. Building a control stream and data file
Record 5: Specification of the PK parameters.
3 lines specifying CL, V and S1 are so called “abbreviated code”
The entire block is called $PK block
The amount A in the observation compartment at the time of observation, divided by the value of a parameter S, is used as the prediction.
In this example A (mg) in CENTRAL compartment will be divided by S1 to match units for observations (ng/mL) and units for predictions (ng/mL). With out scaling factor, the predictions will be passed to $ERROR as mg/L or ug/mL
October 09, 2008
Hands on session

14. Building a control stream and data file
If we did not use TRANS2 in this example
The abbreviated code would need code for K, the rate constant of elimination (K=CL/V)
PREDPP needs the values of microconstants, rather than physiological-based pharmacokinetic parameters such as clearance
See ..\help\trans(1-6).ppp for the choice of translator routines
TRANS2 Used with ADVAN1 and ADVAN2.
TRANS3 Used with ADVAN3 and ADVAN4.
TRANS4 Used with ADVAN3, ADVAN4, ADVAN11, ADVAN12
TRANS5 Used with ADVAN3 and ADVAN4.
TRANS6 Used with ADVAN3, ADVAN4, ADVAN11, ADVAN12
October 09, 2008
Hands on session

15. The (user) choice of the model in the NONMEM ouput
October 09, 2008
Hands on session

16. Building a control stream and data file
Record 6: user’s specification of the (statistical) model for the lack of fit of the pharmacokinetic model to the data
The model is specified in the abbreviated code
the $ERROR record, along with this line of abbreviated code is called the $ERROR block
October 09, 2008
Hands on session

17. Building a control stream and data file
Record 7: Information on possible values of THETA (Format: lower bound, initial estimate, upper bound)
Record 8: Information on initial estimate of variance of ETA
Record 9: Information on initial estimate of variance of ERR
In the current
specification, upper bound
is understood to be
unlimited
October 09, 2008
Hands on session

18. Building a control stream and data file
Record 10: Instructions for the NONMEM Estimation Step
METHOD
For example; 0 (FIRST ORDER (FO) method; requires POSTHOC option to get individual estimates); 1 (FIRST ORDER CONDITIONAL ESTIMATION (FOCE)); HYBRID (Use conditional estimates for the etas during the computation of the objective function, with the exception of those etas listed in the ZERO option)
Review $ESTIMATION from NONMEM help folder ($estimat.ctl)
October 09, 2008
Hands on session

19. Building a control stream and data file
Record 11: Instructions for NONMEM Covariance Step
Record 12: Requests that NONMEM generate a table
October 09, 2008
Hands on session

20. NONMEM output
Documentation
October 09, 2008
Hands on session

21. NONMEM output Search Summary Scaled Transformed Parameters (STP)
Gradient vector of the objective function with respect to the STP
Objective function values
Note that gradient vector at the last iteration is smaller than that at the 0th iteration
Gradient equal zero or unexpectedly large gradient usually means problem with NONMEM estimation- For example, over parameterization or numerical constraints etc.
October 09, 2008
Hands on session

22. NONMEM output
Minimum objective function value: a goodness of fit statistic
For example sum of squares (and as with a sum of squares, the lower the value, the better the fit)
October 09, 2008
Hands on session

23. Parameter estimates and standard error estimates
NONMEM output
Parameter estimates and standard error estimates
CL and V estimates: compare that to your homework mean CLi and Vi (1.c)
Inter individual variability (IIV) in CL and V: compare square root of that to homework estimate of SD. (1st level random effect) (1.c)
Residual variance estimates (2nd level random effect)
Note that all estimates (including IIV and residual variance) have standard error estimated.
“Variability versus Uncertainty” Slide 9 of 57
October 09, 2008
Hands on session

24. NONMEM output Review covarian.out Review correlat.out
Basic output of NONMEM's Covariance Step. Under asymptotic theory, describes the variability under the assumed model of the parameter estimates across (imagined) replicated data
sets, using the design of the real data set.
Review covarian.out
The correlation matrix is the
variance-covariance matrix in correlation form
Correlation between TH1 and TH2
= / (sqrt( )*sqrt(0.386))
Review correlat.out
October 09, 2008
Hands on session

25. Model diagnostics
We will use CENSUS ( for this exercise but learning some programming language (for example, R/S-Plus/SAS etc.) will widen your skills sets and offer flexibility
CENSUS Snapshot
Double click the chicken!
October 09, 2008
Hands on session

26. Model diagnostics Alternatively, click Create a working database
File > New >
Browse to intended working folder and click “OK”
Import the run into CENSUS environment
Runs > Import Run…
Browse to NONMEM output and click “Open”
Click each tab and explore
October 09, 2008
Hands on session

27. Model diagnostics Select the Run (only one here) and click
Here are graphic options available
October 09, 2008
Hands on session

28. Model diagnostics
Compare these plots to your homework plots- 1.h.i and 1.h.ii
October 09, 2008
Hands on session

29. Model diagnostics Compare these plots to your homework plots- 1.d.i
October 09, 2008
Hands on session

30. Model diagnostics
Compare these plots to your homework plots- 1.h.iii and 1.h.iv
October 09, 2008
Hands on session

31. Model diagnostics
Compare these plots to your homework plots- 1.h.iii and 1.h.iv
October 09, 2008
Hands on session

32. Model diagnostics Only BW plots are shown here
Compare these plots to your homework plots- 1.e.i
Only BW plots are shown here
October 09, 2008
Hands on session

33. Will be available by Tuesday (October 14th)
For the next session
One compartment PK model for oral administration (HW#3) and two compartment PK model for IV administration (HW#4)
Submit control records and NONMEM output
Summarize PK parameters
Review covariate model (for HW#2)
Will be available by Tuesday (October 14th)
My notes:
Single precision vs double precision
NONMEM limitations
Estimation methods: Random effects must enter linearly
Only one level of nesting
October 09, 2008
Hands on session