11.06.2014 Views

Uncertainty Analysis of Dose-Response Data with Threshold Modeling

Uncertainty Analysis of Dose-Response Data with Threshold Modeling

Uncertainty Analysis of Dose-Response Data with Threshold Modeling

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

<strong>Uncertainty</strong> <strong>Analysis</strong> <strong>of</strong><br />

<strong>Dose</strong>-<strong>Response</strong> <strong>Data</strong> <strong>with</strong><br />

<strong>Threshold</strong> <strong>Modeling</strong><br />

Jeff Swartout, U.S. EPA, ORD, NCEA, Cincinnati, OH<br />

Office <strong>of</strong> Research and Development<br />

Full Name <strong>of</strong> Lab, Center, Office, Division or Staff goes here. <br />

October 30, 2007


Purpose and Motivation<br />

• Provide alternative to a non-zero BMR<br />

• Consistency in risk estimates<br />

• Compare threshold vs. non-threshold approaches<br />

1


Population <strong>Threshold</strong> Concept<br />

• Considering only adverse (toxic) effects, such as<br />

functional damage to an organ system or death, in the<br />

extreme, there must be some level <strong>of</strong> exposure, below<br />

which the effect does not occur in any individual (Cox,<br />

1987).<br />

• One molecule may destroy an enzyme or disrupt a<br />

membrane but cannot, by itself, result in functional<br />

damage unless the effect is fixed and heritable.<br />

2


<strong>Threshold</strong> vs. Non-<strong>Threshold</strong><br />

cumulative response<br />

0.0001 0.0010 0.0100 0.1000 1.0000<br />

Rat data<br />

HED extrapolation<br />

Non-threshold fit<br />

<strong>Threshold</strong> fit<br />

RfD<br />

+<br />

+<br />

0.1 1.0 10.0 100.0<br />

dose<br />

3


<strong>Threshold</strong> <strong>Dose</strong>-<strong>Response</strong> Models<br />

• Individual tolerance distributions<br />

– Lognormal, Weibull, log-logistic, etc.<br />

– No population threshold<br />

• Population threshold models<br />

– Tolerance distribution <strong>with</strong> D - T term<br />

– Tukey-lambda family (Cox, 1987)<br />

4


<strong>Threshold</strong> Models<br />

Hill (log-logistic)<br />

Pareto<br />

ED<br />

( D −T<br />

)<br />

N<br />

N<br />

50<br />

+ ( D −<br />

T<br />

)<br />

N<br />

⎛<br />

1− ⎜<br />

T<br />

⎝<br />

α<br />

⎞<br />

⎟<br />

⎠<br />

−α<br />

Weibull<br />

1<br />

−<br />

exp<br />

⎡<br />

⎢−<br />

⎢<br />

⎣<br />

⎛<br />

⎜<br />

⎝<br />

D<br />

−<br />

b<br />

T<br />

⎞<br />

⎟<br />

⎠<br />

c<br />

⎤<br />

⎥<br />

⎥<br />

⎦<br />

D = administered dose<br />

T = threshold dose parameter<br />

N = Hill exponent<br />

C = Weibull power<br />

5


Bootstrap Procedure<br />

• Fit threshold models to raw data<br />

• Select best-fitting model<br />

• Compute “true” response for each dose<br />

– Use non-threshold model fit for zero-response dose<br />

groups<br />

6


Bootstrap Procedure<br />

7<br />

• Generate random binomial response for each dose<br />

group (parametric bootstrap)<br />

– Simulates re-running the experiment at fixed doses<br />

<strong>with</strong> random draws from the same population, given<br />

the true probability <strong>of</strong> response at each dose = p d<br />

– Generates a new response vector (number <strong>of</strong><br />

responders)<br />

• rbinom(n d , p d )<br />

– n d is the number <strong>of</strong> individuals in dose group d<br />

– p d is fitted response to raw data<br />

• Fit all models to bootstrapped response<br />

• Save threshold estimates at each iteration from bestfitting<br />

model


Assumptions and Limitations<br />

• True animal response represented by initial model fit<br />

• <strong>Response</strong> at zero-observed response doses equivalent<br />

to fitted non-threshold response (divided by 2)<br />

• Assumed response distribution valid near threshold<br />

• Binomial uncertainty only<br />

• Constraints on parameter space are ignored<br />

8


Sample Bootstrap Output<br />

(Frambozadrine)<br />

cumulative response<br />

0.001 0.005 0.050 0.500<br />

Hill<br />

Weibull<br />

gamma<br />

lognormal<br />

Pareto<br />

1 5 10 50 100<br />

dose (mg/kg-day)<br />

BMDL 10<br />

<strong>Threshold</strong> fits shown in relation to the BMDL<br />

9<br />

Colors indicate best-fitting model at each iteration (100 shown)


Sample Bootstrap Output<br />

(Mordorine)<br />

cumulative response<br />

0.05 0.10 0.50 1.00<br />

0.001 0.010 0.100 1.000 10.000<br />

BMDL 10<br />

dose<br />

<strong>Threshold</strong> fits shown in relation to the BMDL<br />

10<br />

Colors indicate best-fitting model at each iteration (100 shown)


Sample Bootstrap <strong>Threshold</strong><br />

Distributions<br />

0 1 2 3 4 5 6<br />

0.0 0.5 1.0 1.5 2.0<br />

0.0 0.5 1.0 1.5 2.0<br />

<strong>Threshold</strong> (log10 mg/kg-day)<br />

-2.0 -1.5 -1.0 -0.5 0.0 0.5 1.0<br />

<strong>Threshold</strong> (log10 mg/kg-day)<br />

0.0 0.2 0.4 0.6 0.8 1.0<br />

-4 -3 -2 -1 0 1<br />

<strong>Threshold</strong> (log10 mg/kg-day)<br />

0.0 0.2 0.4 0.6 0.8 1.0 1.2<br />

-3 -2 -1 0 1<br />

<strong>Threshold</strong> (log10 mg/kg-day)<br />

11


Frambozadrine<br />

cumulative response<br />

0.0 0.2 0.4 0.6 0.8 1.0<br />

Weibull (BMD)<br />

Weibull threhsold<br />

BMDL---BMD<br />

T05---Tml<br />

5 10 50 100<br />

dose (mg/kg-day)<br />

12


Frobozinate<br />

cumulative response<br />

0.0 0.2 0.4 0.6 0.8 1.0<br />

log-logistic (BMD)<br />

Pareto<br />

BMDL---BMD<br />

T05---Tml<br />

0.1 0.5 1.0 5.0 10.0 50.0 100.0<br />

dose (mg/kg-day)<br />

13


Gruesite<br />

cumulative response<br />

0.0 0.2 0.4 0.6 0.8 1.0<br />

log-logistic (BMD)<br />

Pareto<br />

BMDL---BMD<br />

T05---Tml<br />

10^-4 10^-3 10^-2 10^-1 10^0 10^1 10^2 10^3<br />

dose (mg/kg-day)<br />

14


Phluginium<br />

cumulative response<br />

0.0 0.2 0.4 0.6 0.8 1.0<br />

log-normal (BMD)<br />

log-normal threhsold<br />

BMDL---BMD<br />

T05---Tml<br />

0.5 1.0 5.0 10.0 50.0<br />

dose (mg/kg-day)<br />

15


Mordorene<br />

cumulative response<br />

0.0 0.2 0.4 0.6 0.8 1.0<br />

Weibull (BMD)<br />

Pareto<br />

BMDL---BMD<br />

T05---Tml<br />

0.01 0.10 1.00 10.00<br />

dose (mg/kg-day)<br />

16


Neelixir<br />

cumulative response<br />

0.0 0.2 0.4 0.6 0.8 1.0<br />

Weibull (BMD)<br />

Weibull threhsold<br />

BMDL---BMD<br />

Tml<br />

0.1 1.0 10.0 100.0<br />

dose (mg/kg-day)<br />

17


Summary <strong>of</strong> Results<br />

Compound<br />

Weibull<br />

Power<br />

BMDL 10<br />

a<br />

TL b<br />

BMDLr c<br />

TLr d<br />

TL: BMDL<br />

Frambozadrine<br />

1.4<br />

24.6<br />

4.45<br />

1.7<br />

4.8<br />

0.18<br />

Frobozinate<br />

0.34<br />

0.10<br />

0.091<br />

3.0<br />

11<br />

0.91<br />

Gruesite<br />

0.35<br />

5.7 x 10 -5<br />

0.27<br />

2940<br />

3.7<br />

4780<br />

Phluginium<br />

0.83<br />

1.89<br />

0.19<br />

1.2<br />

5.1<br />

0.10<br />

Mordorene<br />

0.86<br />

0.0027<br />

0.030<br />

217<br />

33<br />

11<br />

Neelixir<br />

1.5<br />

18.1<br />

0<br />

0.25<br />

–<br />

0.22 e<br />

a<br />

95% lower confidence bound on BMD 10 (BMDS)<br />

b<br />

95% lower confidence bound on threshold (bootstrap)<br />

c<br />

Ratio <strong>of</strong> BMDMLE to BMDL<br />

d<br />

Ratio <strong>of</strong> TMLE to TL<br />

e<br />

TMLE : BMDMLE<br />

18


19<br />

That’s All<br />

Weibull (BMD)<br />

Weibull threhsold<br />

BMDL---BMD<br />

T05---Tml<br />

Hill<br />

Weibull<br />

gamma<br />

lognormal<br />

Pareto<br />

5 10 50 100<br />

dose (mg/kg-day)<br />

dose (mg/kg-day)<br />

0.0 0.5 1.0 1.5 2.0<br />

cumulative response<br />

0.0 0.2 0.4 0.6 0.8 1.0<br />

Probability density<br />

0 1 2 3 4 5 6<br />

cumulative response<br />

0.001 0.005 0.050 0.500<br />

1 5 10 50 100<br />

<strong>Threshold</strong> (log10 mg/kg-day)

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!