By default, NONMEM calls PRED with one data record after another, no data records being skipped, and no record being repeated. NONMEM can be requested to call PRED with a different sequence of records, using the Repetition feature of NONMEM. A NONMEM reserved data item RPT_ may be used, or reserved variables RPTO, RPTON, PRDFL may be set in PK or ERROR (see chapter III Section I.B). One use of this feature is for models involving convolution. The discussion below is unaffected by this use of feature.
With NONMEM each observation (i.e. each value of the dependent variable of the regression model) is placed on a separate data record called an observation record
and the value placed on the record is called the dependent variable (DV) data item.
There are situations where it is convenient to place all the values of the concomitant variables (i.e. independent variables) needed to compute the prediction of an observation on the same data record as that one containing the observation. Then all data records are observation records. There are other situations, though, where it is convenient to organize the concomitant variables across different data records. One situation arises when one needs to avoid the NONMEM restriction on the number of data items per data record. (The maximum number of data items per data record is given by constant PD in resource/SIZES.f90. The default value is 50. A larger value of PD may be specified using the NM-TRAN $SIZES record.) Another situation arises with the following example. Suppose there is a chronology to events: at time 4 days a dose of 5 mg is given, at time 5 days a dose of 10 mg is given, and at 7 days a plasma concentration is observed, etc. Then data may be originally recorded in the laboratory or clinic so that all the data items associated with an event occuring at given time are contained on one record, with different records for events at different times. Moreover, these records may be time ordered. When NONMEM data records must be constructed, it may be more convenient to construct records that look like these. NONMEM allows this. A NONMEM data record need not be an observation record, i.e. it need not contain an actual observation. If there are nonobservation data records, each record must contain a missing dependent variable (MDV) data item, whose value is 0 if the record is an observation record, and 1 if the record is not an observation record. The DV data item is ignored when the MDV data item is 1. It need simply be a null value which serves as a place holder for the position where an observation would be placed on an observation record. With NONMEM 7.3, values of MDV may be 100 or 101. MDV=100 is the same as MDV=0 but tells NONMEM to ignore the record during the Estimation and Covariance Steps. MDV=101 is the same as MDV=1 but tells NONMEM to ignore the record during the Estimation and Covariance Steps.
PREDPP is concerned with time ordered events: dose events, observation events, and other events such as the beginning of a urine collection and a measured change in kidney function. This routine expects that there will be different data records associated with each of the different events, and that these data records are time ordered. Data records are also organized into individual records
In the case of population data an individual record is a group of contiguous data records all associated with the same individual and having the same identification (ID) data item. In this case the time ordering of data records expected by PREDPP refers to the time ordering of data records within an individual record. In the case of data from a single subject individual records do not correspond on a one-to-one basis to either animal or human subjects. An individual record is simply any group of contiguous data records containing only one observation record and having the same ID data item. (Although, the presence of a multivariate observation means the individual record containing it has several observation records, each containing one element of the observation.) In this case the time ordering of data records expected by PREDPP refers to the time ordering of data records across all records.
To summarize the foregoing, a typical NONMEM data set for PREDPP (the first 9 records of which) can be illustrated thusly:
ID TIME C1 C2 ... CN DV MDV 1 1. .5 10 5 1 1 3. 9 10 1 1 6. 6 4.1 0 1 7. 7 10 1 1 10. 7 5.3 0 2 2. .9 8 8 1 2 2.5 6 1 2 6. 7 1 2 7. 8 6.4 0
where ID, MDV, and DV label the ID, MDV, and DV data items, where TIME labels the time data items, and where C1, C2, ..., CN label concomitant variable data items. NONMEM itself, and not PREDPP, recognizes and responds to ID, DV, and MDV data items. With PREDPP there must always be time data items giving the times of the events. (An exception occurs with ADVAN9, ADVAN15, or ADVAN17, when there are only algebraic equations to be solved, in which case the time data item is optional.) PREDPP, and not NONMEM, recognizes and responds to TIME data items. Note that some data items are left blank. In a NONMEM data set blanks are null values and will be read by the program as zeros, and displayed as such in tables and scatterplots. (In an NM-TRAN data set blanks serve as a place holder for a null value when a format is given. When a format is not given, a dot (’ . ’) or a zero (’ 0 ’) can be used as a place holder.) Neither NONMEM nor PREDPP supplies nonzero values for null values, or understands these null values as "missing values". In this illustration there are 2 individual records.
In addition to the time data item, each event record must have a data item indicating the type of event in question, and if, for example, the event is a dose-type event, the event record must have additional data items specifying certain dose type information. This is all discussed in Chapter V. Usually, all drug dosing information is specified in the event records, rather than in user-supplied code. In the illustration, the event identification data items are not shown, but the amount data items are labeled CN, though ordinarily, they might be labeled AMT; see section V.D.
In this illustration there are 9 data records corresponding to 5 different time ordered events within the first individual record and 4 different time ordered events within the second individual record. Three of these events are observation events: two within the first individual record, and one within the second individual record. At least four of these events are dose events (one cannot identify dose events simply by the presence of nonzero amount data items): three within the first individual record, and one within the second individual record. Typically, as in the illustration, with PREDPP each data record is also an event record
describing a separate event.
However, infrequently, several
data records may be needed to describe an event.†
----------
† The CONT data
item described in this paragraph cannot be used with NM-TRAN
abbreviated code.
----------
One can group a number of data records associated with a given event into a single event record. This is accomplished by including in each data record a data item indicating whether or not the record and the succeeding record are both members of the same event record. This data item is called the continuation data item (CONT data item). Its value is 1 if the data record and the succeeding record are both members of the same event record, and 0 otherwise (or if the data record is the last data record). It is PREDPP that recognizes and responds to CONT data items, and the concept of an event record is peculiar to PREDPP, not NONMEM. If continuation data items are not used, each data record constitutes a separate event record.
A slightly different example of a NONMEM data set is this:
ID TIME CONT C1 ... CN DV MDV 1 1 .5 1 1 1. 0 5 1 1 6. 0 4.1 0 1 7. 0 10 1 1 10. 0 5.3 0 2 1 .9 1 2 1 1 2 6. 0 8 1 2 7. 0 6.4 0
Here the CONT data items appear. There are 6 event records, 4 in the first individual record and 2 in the second individual record. Both the 1st and 3rd event records contain information needed to predict the second observation (at time 10 hr). At time 1 hr a dose of 5 mg is given, and at 7 hr a dose of 10 mg is given. A plasma concentration is measured at 6 hr. The time associated with each event must appear on the last data record of the corresponding event record; it need not appear elsewhere in the event record.
TOP
TABLE OF CONTENTS
NEXT CHAPTER ...