TABLE 3.

Main steps in population model building^a

Step	Model description	OBFVb	Comments
1	One-compartment model, CL = θ1CLCR	1,476	ςɛ2 = 16.8
2	Two-compartment model, CL = θ1CLCR	1,325	ςɛ2 = 0.441; much better than step 1
3	Infusion duration implemented as a random variable	1,325	Not better than step 2
4	= θ1CLCR + θ2	1,290	Better than step 2
5	= θ1 × 6 × [θ3 − (age/100)]/(bw/SCR) + θ4, with i = 1 for males and 2 for females	1,281	Better than step 4c
6	Similar to step 5, but θ1 = θ2	1,285	Sex is a significant covariatec
7 to 10	Typical values of either CL, V1, CLD, or Vt are allowed to differ according to o.d. or b.i.d. regimen	1,282 to 1,287	No influence of dosing regimen
11 to 14	Similar to steps 7 to 10 but with separated η’s according to o.d. or b.i.d. regimen	1,282 to 1,289	No influence of dosing regimen
15	= θ7 + [(θ8 − θ7) × time]/(θ9 + time)	1,281	θ8 tends to θ7; no influence of time on Vt
16	Similar to step 15, but for V1	1,281	No influence of length of therapy
17	= θ5 × (bw/65) + θ6	1,280	Not significantc
18	= θ7 (bw/65) + θ8	1,281	Not significantc
19	Similar to step 5, but with FOCE η-ɛ interaction method	1,248	ςɛ2 = 0.189
20	Similar to step 19, but C = Ĉ + ɛ1Ĉb + ɛ2 with var (ɛ2) fixed to 0.25	1,258	Not better than step 19

^aAfter step 2 the two-compartment model was always used. After step 6 the clearance model described in step 5 was always used. 

^bOBFV, objective function value. 

^cAdditional criteria (see Materials and Methods section) were also considered for the decision.