0% found this document useful (0 votes)
5 views17 pages

Nonlinear Regression Models Overview

The document discusses nonlinear modeling techniques, including polynomial and transformed variable models, as well as nonlinear regression models, particularly in the context of analyzing the effects of orlistat on fecal fat excretion. It outlines the structure of nonlinear regression equations, estimation methods, and the process of minimizing error sums of squares to derive coefficients. Additionally, it provides an example with patient data and the iterative process for estimating parameters, along with variance estimates and confidence intervals for the model coefficients.

Uploaded by

Sonali
Copyright
© All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
5 views17 pages

Nonlinear Regression Models Overview

The document discusses nonlinear modeling techniques, including polynomial and transformed variable models, as well as nonlinear regression models, particularly in the context of analyzing the effects of orlistat on fecal fat excretion. It outlines the structure of nonlinear regression equations, estimation methods, and the process of minimizing error sums of squares to derive coefficients. Additionally, it provides an example with patient data and the iterative process for estimating parameters, along with variance estimates and confidence intervals for the model coefficients.

Uploaded by

Sonali
Copyright
© All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd

Nonlinear Modeling

Nonlinear Relations wrt X – Linear wrt bs


1) Polynomial Models: E Yi  = b 0 + b1 X i + b 2 X i2
E Yi  
=  b 0 + b1 X i + b 2 X i2  = 0 + b1 + 2b 2 X i = h ( X i )
X i X i
 b0 
E Yi  E Yi  E Yi 
=1 = Xi = X i2 None are functions of β =  b1 
b 0 b1 b 2
 b 2 

 1 
2) Transformed Variable Models: E   Yi = b 0 + b1 ln ( X i1 ) + b 2  
 X i2 
E  Y=b  1  E  Y   1 
 = h1 ( X i1 )  = h2 ( X i 2 )
i i
1 = −b 2 
X i1  i1 
X X i 2  X2 
 i2 
 b0 
E Yi  E Yi  E Yi   1 
=1 = ln ( X i1 ) =  None are functions of β =  b1 
b 0 b1 b 2  X i2   b 2 

In each case: E Yi  = f ( Xi , β ) = X'i β


 1 
Case 1: X'i = 1 Xi X i2  Case 2: X'i = 1 ln ( X i1 ) 
 X i2 
Nonlinear Regression Models
Nonlinear Regression models often use γ as vector of coefficients to distinguish from linear models:
Exponential Regression Models (Often used for modeling growth, where rate of growth changes):
E Yi  E Yi 
E Yi  =  0 exp (  1 X i )  = exp ( 1 X i ) =  0 X i exp ( 1 X i ) functions of γ
 0  1
f ( Xi , γ ) =  0 exp (  1 X i )  X'i γ

More general exponential model (with errors independent and N 0,  2 ) : ( )


Yi =  0 +  1 exp ( 2 X i ) +  i Typically,  0  0,  1  0,  2  0

 Intercept: E (Yi | X i = 0 ) =  0 +  1 (1) =  0 +  1


 Asymptote: E (Yi | X i →  ) =  0 +  1 (1) =  0
 0.693    0.693    
 "Half-way" Point: E  Yi | X i =  =  0 +  1 exp   2    =  0 +  1 exp ( −0.693) =  0 +  1 
  2    2  2
   
Data Description - Orlistat
• 163 Patients assigned to one of the following doses
(mg/day) of orlistat: 0,
60,120,150,240,300,480,600,1200
• Response measured was fecal fat excretion (purpose is
to inhibit fat absorption, so higher levels of response
are considered favorable).
• Plot of raw data displays a generally increasing but
nonlinear pattern and large amount of variation across
subjects.
Fecal Fat Excretion vs Orlistat Daily Dose

60

50

40
Percent FFE

ffe
30
meanffe

20

10

0
0 200 400 600 800 1000 1200 1400
Dose
Nonlinear Regression Model - Example

1x
Y = 0 + +
2 + x

Simple Maximum Effect model:


 0 ≡ Mean Response at Dose 0
• 1 ≡ Maximal Effect of Orlistat (0+ 1 = Maximum Mean Response)
• 2 ≡ Dose providing 50% of maximal effect (ED50)
Nonlinear Least Squares
1 X i
f ( Xi , γ ) = f i ( γ ) = f ( 0 ,  1 ,  2 ) =  0 +
 2 + Xi
f ( Xi , γ ) f i ( γ )  Xi − 1 X i 
= Fi ( γ ) = = 1 
γ ' γ '  + (  2 + X i ) 
2
 2 X i

  1 X1 
 +
 f1 (  )    2 + X1 
0
 Y1 
     
Y=  f (γ) =  = 
Yn   f n (  )    X

 0 + 1 n 
  2 + X n 
 X1 − 1 X 1 
1 2 
 F1 (  )    2 + X 1 ( 2
 + X n) 
  
F(γ) =  =


 Fn (  )   Xn − 1 X n 
1
  +X 2 
 2 n ( 2
 + X n) 

F acts like the X matrix in linear regression (but depends on parameters)


Nonlinear Least Squares
^ ^ ^
Goal: Choose  0 ,  1 ,  2 that minimize error sum of squares:
2
n
 1X i  
Q = SSE ( γ ) =   Yi −  0 +   =
  + 
i =1   2 X i 

= ( Y - f ( γ )) ( Y - f ( γ ))
'

n 
Q  1X i  
= −2  Yi −  0 +  Fi (  j ) j = 0,1, 2
 j   2 + X i  
i =1  
Q set
= −2  Y - f ( γ )  F ( γ ) =  0 0 0
T

γ '
Estimated Variance-Covariance Matrix


-1
^   ^' ^
s 2
γ = s F F
2

 
T
   ^
 ^
Y-f  Y-f 
s2 =    
n− p


-1
^  
^' ^
s i = s F F
  ( i+1,i+1)
^ ^
Note: KNNL uses g for γ and D for F
Orlistat Example
• Reasonable Starting Values:
– 0: Mean of 0 Dose Group: 5
– 1: Difference between highest mean and dose 0
mean: 33-5=28
– 2: Dose with mean halfway between 5 and 33: 160
• Create Vectors Y and f (0)
• Generate matrix F(0)
• Obtain first “new” estimate of 
• Continue to Convergence
Orlistat Example – Iteration History (Tolerance = .0001)
iteration g0 g1 g2 SSE Delta(g)
0 5.0000 28.0000 160.0 13541.6
1 6.2379 28.5863 140.9 12945.5 365.5745418
2 6.1771 28.1281 133.7 12942.9 52.82513814
3 6.1507 27.9163 129.9 12942.2 14.44158887
4 6.1361 27.7967 127.8 12942.0 4.506448063
5 6.1277 27.7272 126.5 12941.9 1.510150161
6 6.1227 27.6861 125.8 12941.9 0.526393989
7 6.1197 27.6615 125.4 12941.9 0.187692352
8 6.1180 27.6467 125.1 12941.9 0.067822683
9 6.1169 27.6377 125.0 12941.9 0.024703325
10 6.1162 27.6323 124.9 12941.9 0.009040833
11 6.1158 27.6291 124.8 12941.9 0.003318268
12 6.1156 27.6271 124.8 12941.9 0.001220029
13 6.1155 27.6259 124.8 12941.9 0.000449042
14 6.1154 27.6251 124.7 12941.9 0.000165379
15 6.1153 27.6247 124.7 12941.9 6.09317E-05

^ 27.62 X
Y = 6.12 +
124.7 + X
Fitted Equation, Raw Data - FFE vs ODD

60

50

40
fecal fat excretion

ffe
30
f(odd,bhat)

20

10

0
0 200 400 600 800 1000 1200 1400
orlistat daily dose
Variance Estimates/Confidence Intervals
2
163
 ^
 
  i
Y − f i  γ 
 
s 2 = i =1  = 80.89
163 − 3
 1.1594 −0.7219 15.609 

' ^ -1
2  
s γ = s  F F  =  −0.7219 130.14 
^ ^
2
12.081
   15.609 130.14 2238.76 

Parameter Estimate Std. Error 95% CI

0 6.12 1.08 (3.96 , 8.28)


1 27.62 3.48 (20.66 , 34.58)
2 124.7 47.31 (30.08 , 219.32)
Notes on Nonlinear Least Squares
Large-Sample Theory:

When  i ~ N ( 0,  )
^
2
independent , for large n: γ is approximately normal

  
-1
^ ^ ^
^ ^
E γ γ Approximate  2
γ estimated by s γ = MSE  F 'F 
2

 

( )
^
 γ ~ N γ ,  2 ( F'F )
-1
(Approximately)

For small samples:


• When errors are normal, independent, with constant variance, we can often use the
t-distribution for tests and confidence intervals (software packages do this implicitly)
• When the extent of nonlinearity is extreme, or normality assumptions do not hold,
should use bootstrap to estimate standard errors of regression coefficients
Example
Many patients get concerned when a medical test involves an injection of
radioactive material. For example, to scan a gallbladder, a few drops of
Technetium-99m isotope are used.
Half of the Technetium-99m would be gone in about 6 hours. However, it takes
about 24 hours for the radiation levels to reach what we are exposed to in day-to-
day activities.
Below is given the relative intensity of radiation as a function of time.

If the level of the relative intensity of radiation is related to time via an


exponential formula

The value of the regression constants 𝐴 and λ,


Example contd…
Example contd…

You might also like