Transcript PPT

Model-based characterization of dose-response relationship in exploratory clinical
development – extracting a small signal from noisy response data
1
Wilkins ,
1
Renard ,
1
Lemarechal ,
Michael
Per Olsson
Justin
Didier
Marie-Odile
Steve
(1) Novartis Pharma, AG, Basel, Switzerland; (2) Exprimo NV, Berenlaan, 4, Beerse, B-2340, Belgium.
5. Model Evaluation
where the blue shaded area represents the 95% prediction interval
Goodness-of-fits Plots
Observed FEV1 [L]
Exploratory clinical development trials often include biomarkers or clinical readout (safety or efficacy) that
exhibit significant within-subject variability in their time courses. This variability is due in part to systemic
diurnal patterns as well as apparent random changes. Typical examples include heart rate, blood pressure,
cortisol and histamine. Applying mixed-effects modeling to analyze data from the entire time-course is
appealing because it allows simultaneous quantification of fixed and random effects.
This model-based approach is used here to extract a small drug effect signal from very noisy lung function
data; previously, a primary analysis based on single measurement at 24 hours post dose had failed to
establish a dose-response relationship.
Visual Predictive Checks
4.0
Dose= 0
3.5
2.5
4
2.0
1.0
0.5
0.5
Observed FEV1 [L]
Period washout Period washout Period
1
2
3
4
1.0
1.5
2.0
2.5
3.0
3.5
4.0
4.0
3
2
3.0
2.5
1
1
0
0
0
0
2.0
1.0
1.5
2.0
2.5
3.0
3.5
4.0
Time [hr]
40
0
10
20
30
Time [hr]
40
0
Dose= 260
IOV (% RSE)
A: rhythm adjusted 24-hour mean FEV1 [L]
B1: 24-hour period amplitude
C1: 24-hour period phase shift [hr in clock time]
1.88 (5)
0.114 (10.8)
14.6 (1)
32.4 (18)
39.3 (31.7)
5.4 (36.4)
5.2 (19)
NE
NE
B2: 12-hour period amplitude
0.04 (13.2)
NE
NT
9 (1.4)
NE
NT
KA [hr-1]
KE [hr-1]
0.313 (41.2)
0.094 (16.9)
NE
NE
NT
53.8 (28.1)
EDK50 [μg.hr-1]
4.05 (23.7)
NE
75.8 (43.1)
Emax [L]
0.466 (10.3)
NE
41.5 (42.6)
10
20
30
Time [hr]
40
Dose= 400
4
FEV1 [L]
FEV1 [L]
analysis using PsN[2], the model failed to
converge only once, suggesting good
stability. Bootstrap estimates of the
mean for all model parameters indicated
very small bias compared with original
estimates: smaller than 10% for all fixed
effects parameters, smaller than 18% for
all random effects parameters.
 Case-deletion diagnostics using PsN
did not identify any influential subjects.
4
3
2
4
3
2
3
2
1
1
1
0
0
0
0
10
20
30
Time [hr]
40
0
10
20
30
Time [hr]
40
0
10
20
30
Time [hr]
0.5
6. Simulation of dose-response with FEV1@24hr post dose as end point
% of 400 μg
response
Predicted doses (95% CI)
[μg]
90
316 (196 – NA)
75
222 (138 – 305)
50
117 (73 – 161)
10
NA: not estimable
18 (11 -24)
model-based
dose selection
for next dosefinding trial
0.1
4. Model Parameter Estimates
NE: not estimated; NT: not tested
30
Dose= 160
0.0
Log-transformed FEV1 measurements (excluding the positive control group) were analyzed simultaneously to
develop a population kinetic-pharmacodynamic (K-PD) model [1] using NONMEM, Version 1.1. Model
components included:
 Two cosine functions with periods of 24 hours and 12 hours to account for circadian variation at baseline
 A 2-compartment model with first order absorption and elimination to account for the kinetics of drug
amount in the virtual effect compartment
 An Emax function to relate the longitudinal drug input function with the response
 Exponential error models to allow for inter-individual variability (IIV) and inter-occasion variability (IOV);
and an additive error model for residual variability.
6.2 (11.2)
20
1.0
24hr FEV1 [L]
0.2
0.3
0.4
3. Model Structure
Residual variability [CV%]
10
1.5
 From a 200-iteration bootstrap
Part A and B: 34 patients; Part A only: 4 patients; Part B only: 6 patients
C1: 12-hour period phase shift [hr in clock time]
2
Individual Predictions [L]
Period washout Period
1
2
Estimate (% RSE) IIV (% RSE)
2
3
3.5
0.5
Parameter
3
1
0.5
Part B: One of four selected dose levels of Drug X or
placebo, in randomized order
4
1.5
2. Data
Part A: 400 μg of Drug X, placebo, or positive control
in randomized order
Dose= 80
3.0
Population Predictions [L]
Forced expiratory volume in one second (FEV1) was measured in patients at pre-dose baseline and a range of
time points up to 48 hours post dose in a Phase I trial:
Dose= 20
FEV1 [L]
1. Introduction
1
Pascoe
FEV1 [L]
2
Gisleskog ,
FEV1 [L]
1
Looby ,
FEV1 [L]
Kai
1
Wu ,
0
100
200
Doses (ug)
300
400
7. Conclusions
The large between- and within-patient variability typically seen in FEV1 data can confound a small treatment signal.
Standard statistical approaches, therefore, often fail to characterize dose-response relationships, particularly in small
exploratory trials. A model-based approach which accounts for systematic and random sources of variability appears to
improve the signal-to-noise ratio of the efficacy signal sufficiently to enable characterization of the dose-response
relationship.
References
[1] Jacqmin P. et al, Modeling response time profiles in the absence of drug concentrations; definition and performance evaluation of the K-PD model. J Pharmacokinet Pharmacodyn 2006 34: 57-85
[2] Lindbom L. et al, PsN-Toolkit-A collection of computer intensive statistical methods for non-linear mixed effect modeling using NONMEM. Comput Methods Programs Biomed 2005 79: 241-257
40