Identifies one or more Super ID data items.

• item: the name of a data item listed on $INPUT. It defines an additional nested random level. and is referred to as a "super ID" data item. The first super ID item defines an additional random nesting level above that of subject ID. More than one super ID item may be listed on $LEVEL. Each subsequent super ID item defines an additional nesting level above that of the previous nesting level, i.e., above the previous super ID.

• nk[mk]: states that ETA(nk) is associated with this super ID item, and ETA(mk) is nested within ETA(nk). With NONMEM 7.4, a short-hand notation may be used to describe a series of values of nk. A sequence of values for nk can be described as

  1	start TO end BY interval

  TO is required. The character : may be used instead of TO. BY is optional. Default is 1. The value of BY may be negative. If the second value of BY differs, the same syntax may also be used for mk.

Nesting levels below the subject ID is modelled as with previous versions of NONMEM.

(See Interoccasion_variability example).

The order that super ID's are listed on $LEVEL defines their nesting level. The order that standard and super ID's are listed on $INPUT (i.e., the order in which they appear in each record of the data set) is immaterial.

When $LEVEL is used with FOCE ($ESTM METHOD=1), the SLOW option is required, and MATRIX=R is required with $COV.

In the example below, data item SID is the site ID, and CID is the country ID. There are several sites per country. The random effect ETA(1) for subject clearance CL is nested in ETA(5), the random effect for sites, which in turn is nested in ETA(9), the random effect for country. Similarly, the random effect ETA(2) for subject volume of distribution V1 is nested in ETA(6), the random effect for sites, which is further nested in ETA(10), the random effect for countries.

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11

$INPUT ... ID ... SID ... CID ...
 ...
$PK
MU_1=THETA(1)
MU_2=THETA(2)
CL=DEXP(MU_1+ETA(1)+ETA(5)+ETA(9))
V1=DEXP(MU_2+ETA(2)+ETA(6)+ETA(10))
 ...
$LEVEL
SID=(5[1],6[2])
CID=(9[5],10[6])

alternatively one can write the above nested structure as

1
2

 SID=(5 to 6[1])
 CID=(9 to 10[5])

For the above model NONMEM will impose appropriate constraints for the ETAs, e.g. ETA(5) will vary by site so that it changes with different SID values instead of every ID values.

The above method, using $LEVEL, is a linearized approximation at the super ID level, and takes advantage of a dual OBJ function call, freely allowing all ETAs to vary on the first call of OBJ, then averaging the SID ETAs, fixing them to these averages, and going through another OBJ call to allow the subject (ID) ETAs to be assessed. This approximation method works very well for the EM and Monte Carlo methods, and reasonably well for the FOCE/Laplace methods.

To perform an exact analysis, separate THETAs must be defined for each value pertaining to a super ID data item, so that theta is shared only by the subjects with the particular SID value. $LEVEL is not used. (See superid3_6).

If there are multiple $PROBLEM records, $LEVEL should be restated with each problem for which it is still relevant. For example, this is the case with $SUPER problems and $LEVEL.

See also LEVWT option (NM74) of the $ESTIMATION record. By default, LEVWT=0, and weights each level value equally, regardless of number of subjects per level value. If you wish to weight according to number of subjects for that value, set LEVWT=1 on the $EST record.

See also nested random effects modeling.