___________________________________________________________________
| |
| NMPRD5 |
|_________________________________________________________________|
MEANING: NONMEM-PRED common
CONTEXT: PRED and ERROR routines
USAGE:
COMMON /NMPRD5/ C(NO,LVR-1)
(If double precision is to be used):
DOUBLE PRECISION C
(If single precision is to be used):
REAL C
The values of NO and LVR must be the same as that found in file SIZES
(See sizes).
DISCUSSION:
An individual record is divided into L2 records. An L2 record may
contain one or more observations (on one or more separate data records
respectively), in which case it is called an observation-L2 record.
The values of epsilons used in the intraindividual model may be corre-
lated across the observations contained in the L2 record, and thus the
L2 record may define a multivariate observation - the L2 observation.
(When all L2 observations in the data set are univariate, L2 data
items need not appear, and when L2 data items do not appear, NONMEM
assumes that each data record is a distinct L2 record.)
By default, the values of a given epsilon are statistically indepen-
dent across L2 observations within an individual record. Using
NMPRD5, however, these values may be correlated. More precisely, the
values of the epsilons associated with the mth diagonal block of SIGMA
may be correlated across L2 observations, and it will be understood
that for two different epsilons (eps1 and eps 2, say) associated with
the mth block, the correlation between the values of eps1 for L2
observations A and B will be taken to be the same as the correlation
between the values of eps2 for these same two L2 observations.
To use NMPRD5, proceed as follows. With the first data record of the
nth observation-L2 record, and with respect to the values of the epsi-
lons associated with the mth diagonal block of SIGMA, the PRED routine
should set C(k,m), for k=1,...,n, to the correlation between the
values for the kth L2 observation and the nth L2 observation. (C(n,m)
should be set to 1.0; in particular, with the first data record of the
1st observation-L2 record, C(1,m) should be set to 1.0.)
E.g. Suppose that the L2 observations are univariate and chronologi-
cally ordered by a TIME data item, and suppose that the intraindivi-
dual model has only one epsilon. Then the values of this epsilon may
be autocorrelated across the L2 observations, as specified by:
DIMENSION T(50)
IF (NEWIND.NE.2) I=0
IF (NEWL2.EQ.1.AND.MV.EQ.0) THEN
I=I+1
T(I)=TIME
DO 5 J=1,I
5 C(J,1)=EXP(-THETA(4)*(TIME-T(J)))
ENDIF
Here MV is a data item that is assumed to exist in the data set and
which is analogous to the usual Missing Dependent Variable (MDV) data
item, only MV signals the presence/absence of a L2 observation within
an L2 record, on the first data record of the L2 record. If the
-- --- ----- ---- ------ -- --- -- ------
first data record of an observation-L2 record is always an observation
record, then the MV item may be the MDV item itself. Such is the case
when there are no L2 data items (in which case, since NEWL2 is always
1, NEWL2 need not be tested).
E.g. Suppose that the L2 observations are bivariate and chronologi-
cally ordered by a TIME data item, and suppose that the intraindivi-
dual model has two epsilons (one for each element of the bivariate L2
observation), each associated with the same diagonal block of SIGMA,
Then the values of these epsilons may be autocorrelated across the L2
observations, as specified by the above code. But if the two epsilons
are associated with two different diagonal blocks, then one might use
this code:
DIMENSION T(50)
IF (NEWIND.NE.2) THEN
I1=0
I2=0
IF (NEWL2.EQ.1.AND.MV1.EQ.0) THEN
I1=I1+1
T1(I1)=TIME
DO 5 J1=1,I1
5 C(J1,1)=EXP(-THETA(4)*(TIME-T1(J1)))
ENDIF
IF (NEWL2.EQ.1.AND.MV2.EQ.0) THEN
I2=I2+1
T2(I2)=TIME
DO 5 J2=1,I2
5 C(J2,2)=EXP(-THETA(5)*(TIME-T2(J2)))
ENDIF
Here MV1 and MV2 are two data items that are assumed to exist in the
data set, and MDV1 (MDV2) signals the presence/absence of the first
(second) element of the L2 observation within an L2 record, on the
first data record of the L2 record.
RESTRICTIONS:
Because C is not recognized by NM-TRAN, and because of other restric-
tions regarding abbreviated code, a specification of C, as above,
within a block of abbreviated code, must be done using verbatim code.
"Single-subject data" with correlated residual error, can be simulated
and analyzed. To do this, though, a technique is needed which can
always be used with such data: the data are handled as data from a
population sample with a single individual, and OMEGA is constrained
to be 0.
If data are simulated, the correlations must be stored in NMPRD5
before PRED calls the NONMEM utility routine SIMEPS. Since NM-TRAN
generated or Library code has a call to SIMEPS in its second section
(see Guide IV), this means that correlations must be computed and
stored using verbatim code in the FIRST block.
REFERENCES: None.
Go to main index.
Created by nmhelp2html v. 1.0 written by Niclas Jonsson (Modified by AJB 5/2006,11/2007)