29.07.2014 Views

ST4241: Design and Analysis of Clinical Trials - The Department of ...

ST4241: Design and Analysis of Clinical Trials - The Department of ...

ST4241: Design and Analysis of Clinical Trials - The Department of ...

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

<strong>ST4241</strong>: <strong>Design</strong> <strong>and</strong> <strong>Analysis</strong> <strong>of</strong> <strong>Clinical</strong> <strong>Trials</strong><br />

2012/2013: Semester I<br />

Tutorial 8<br />

1. <strong>The</strong> linear model for the balanced incomplete block design with design parameters g, k, r<br />

<strong>and</strong> λ is as follows:<br />

X ij = µ + s i + α j + ɛ ij ,<br />

where ∑ j α j = 0, s i ’s are i.i.d. r<strong>and</strong>om variables with mean zero <strong>and</strong> variance σ 2 s, ɛ ij ’s are<br />

i.i.d. r<strong>and</strong>om errors with mean zero <strong>and</strong> variance σ 2 e <strong>and</strong> are independent <strong>of</strong> s i ’s. Let a j be the<br />

unbiased estimate <strong>of</strong> α j as defined in the lecture notes.<br />

(i) Show that the a j ’s are least squares estimates <strong>of</strong> α j ’s; that is they are minimizers <strong>of</strong><br />

∑ ∑<br />

(X ij − µ − s i − α j ) 2 , subject to ∑<br />

i j<br />

j<br />

α j = 0.<br />

By Lagrange method, taking derivatives w.r.t. the unknowns <strong>of</strong> the function <strong>and</strong> setting<br />

them to zero<br />

∑ ∑<br />

(X ij − µ − s i − α j ) 2 + κ ∑ α j<br />

i j<br />

j<br />

yields the following equations:<br />

(1)<br />

(2)<br />

(3)<br />

(4)<br />

X·· = grµ + k ∑ s i<br />

X i· = k(µ + s i ) + ∑ α j , i = 1, . . . , n<br />

j∈T i<br />

X·j = r(µ + α j ) + ∑ s i + κ, j = 1, . . . , g<br />

i∈S j<br />

∑<br />

α j = 0<br />

j<br />

where T i is the set <strong>of</strong> j’s such that Treatment j is applied in Block i, S j is the set <strong>of</strong> i’s<br />

such that Block i receives Treatment j.<br />

1


Sum up (3) over j yields<br />

(5) X·· = grµ + k ∑ s i + gκ<br />

(5)-(1) yields that κ = 0. From (2) <strong>and</strong> (3), we have<br />

(6)<br />

(7)<br />

µ + s i = ¯X i· − 1 k<br />

∑<br />

j ′ ∈T i<br />

α j<br />

′<br />

X·j = rα j + ∑ i∈S j<br />

(µ + s i )<br />

Substituting (6) into (7) yields<br />

It yields<br />

(ii) Show that<br />

X·j = rα j + ∑ ( ¯X i· − 1 ∑<br />

α<br />

k j<br />

′)<br />

i∈S j j ′ ∈T i<br />

= rα j + rM j − 1 ∑ ∑<br />

k<br />

α j<br />

′<br />

i∈S j j ′ ∈T i<br />

= rα j + rM j − r − λ<br />

k α j.<br />

(1 − r − λ<br />

kr )α j = ¯X·j − M j , i.e., α j = 1<br />

eff ( ¯X·j − M j ).<br />

Var(a j ) =<br />

where eff = g(k−1)<br />

k(g−1) .<br />

σ2 e<br />

reff<br />

g − 1<br />

, Cov(a j , a k ) = − σ2 e<br />

g<br />

1<br />

reff g ,<br />

Write<br />

¯X·j − M j = 1 ∑<br />

(X ij −<br />

r<br />

¯X i·)<br />

i∈S j<br />

= 1 ∑<br />

(α j − 1 ∑<br />

α<br />

r k j<br />

′ + ɛ ij − 1 ∑<br />

ɛ<br />

k ij<br />

′).<br />

i∈S j j ′ ∈T i j ′ ∈T i<br />

2


Thus<br />

Var( ¯X·j − M j ) = 1 ∑<br />

Var(ɛ<br />

r 2 ij − 1 ∑<br />

ɛ<br />

k ij<br />

′)<br />

i∈S j j ′ ∈T i<br />

= 1 ∑<br />

[(1 − 1 r 2 k )2 + k − 1 ]σ<br />

k 2 e<br />

2<br />

i∈S j<br />

= k − 1<br />

rk σ2 e.<br />

Hence<br />

Var(a j ) = k − 1<br />

rkeff 2 σ2 e =<br />

σ2 e k − 1<br />

reff keff =<br />

σ2 e<br />

reff<br />

g − 1<br />

.<br />

g<br />

Similarly,<br />

Cov( ¯X·j − M j , ¯X·l − M l )<br />

= 1 ∑ ∑<br />

Cov(ɛ<br />

r 2 ij − 1 ∑<br />

ɛ<br />

k ij<br />

′, ɛ i ′ l − 1 ∑<br />

ɛ ′<br />

k i j<br />

′)<br />

i∈S j i ′ ∈S l j ′ ∈T i j ′ ∈T ′ i<br />

= 1 r λCov(ɛ 2 ij − 1 ∑<br />

ɛ<br />

k ij<br />

′, ɛ il − 1 ∑<br />

ɛ<br />

k ij<br />

′)<br />

j ′ ∈T i j ′ ∈T i<br />

= λ r 2 (−σ2 e<br />

k ).<br />

Hence<br />

Cov(a j , a l ) =<br />

λ<br />

r 2 eff 2 (−σ2 e<br />

k ) = − 1 σe<br />

2<br />

reff g .<br />

(iii) Show that, for any contrast ∑ g<br />

j=1 c ja j ,<br />

Var(<br />

g∑<br />

c j a j ) =<br />

j=1<br />

σ2 e<br />

reff<br />

∑<br />

c 2 j.<br />

j<br />

3


g∑<br />

Var( c j a j )<br />

j=1<br />

= ∑ j<br />

= ∑ j<br />

= ∑ j<br />

∑<br />

c j c l Cov(a j , a l )<br />

c 2 j<br />

c 2 j<br />

l<br />

σ 2<br />

reff<br />

g − 1<br />

g<br />

− ∑ j≠l<br />

σ 2<br />

reff [g − 1 + 1 g g ] = ∑ j<br />

1 σe<br />

2 c j c l<br />

reff g<br />

c 2 j<br />

σ 2<br />

reff .<br />

Note that<br />

0 = ( ∑ j<br />

c j ) 2 = ∑ j<br />

c 2 j + ∑ j≠l<br />

c j c l .<br />

2. <strong>The</strong> data from the interexaminer reliability study considered in the lecture notes is given on<br />

the next page.<br />

(i) Using the linear model approach, derive the F ratio for testing the significance <strong>of</strong> the<br />

differences among the examiners.<br />

Examiner<br />

Patient 1 2 3 4 5 6 Mean<br />

1 10 14 10 11.33<br />

2 3 3 1 2.33<br />

3 7 12 9 9.33<br />

4 3 8 5 5.33<br />

5 20 26 20 22.00<br />

6 20 14 20 18.00<br />

7 5 8 14 9.00<br />

8 14 18 15 15.67<br />

9 12 17 12 13.67<br />

10 18 19 13 16.67<br />

Mean 8.6 11.2 13.2 10.6 16.2 14.2 12.33<br />

<strong>The</strong> data is fitted to the following model<br />

∑10<br />

6∑<br />

X = µ 0 + s i b i + a j t j + ɛ,<br />

i=2 j=2<br />

4


where the dummy variables b i ’s <strong>and</strong> t j ’s are defined conventionally.<br />

<strong>The</strong> R code<br />

lm.fit1=lm(x~s+a)<br />

produces the following ANOVA table:<br />

Df Sum Sq Mean Sq F value Pr(>F)<br />

s 9 982.00 109.11 11.7558 2.615e-05 ***<br />

a 5 35.44 7.09 0.7638 0.5898<br />

Residuals 15 139.22 9.28<br />

<strong>The</strong> F ratio for the examiner effect is 0.7638 with a p-value 0.5898. <strong>The</strong>re is no significant<br />

effect <strong>of</strong> the examiners.<br />

(ii) By looking at the data, it seems that Examiner 5 <strong>and</strong> 6 have higher mean scores than the<br />

other examiners. One would like to see whether the following contrasts are significant:<br />

α 1 + α 2 + α 3 + α 4<br />

4<br />

α 1 + α 2 + α 3 + α 4<br />

4<br />

α 1 + α 2 + α 3 + α 4<br />

4<br />

− α 5 + α 6<br />

,<br />

2<br />

− α 5 ,<br />

− α 6 .<br />

Using an appropriate multiple comparison criterion, test the significance <strong>of</strong> the above<br />

contrasts by controlling the overall error rate at 0.05.<br />

In terms <strong>of</strong> the parameters <strong>of</strong> the linear model, the contrasts are equivalent to the linear<br />

forms:<br />

a 2 + a 3 + a 4<br />

4<br />

a 2 + a 3 + a 4<br />

4<br />

a 2 + a 3 + a 4<br />

4<br />

− a 5 + a 6<br />

,<br />

2<br />

− a 5 ,<br />

− a 6 .<br />

<strong>The</strong> estimated a j ’s <strong>and</strong> their variance matrix are as follows:<br />

5


a2 a3 a4 a5 a6<br />

1.750000 1.083333 3.333333 3.416667 1.416667<br />

a2 a3 a4 a5 a6<br />

a2 4.640741 2.320370 2.320370 2.320370 2.320370<br />

a3 2.320370 4.640741 2.320370 2.320370 2.320370<br />

a4 2.320370 2.320370 4.640741 2.320370 2.320370<br />

a5 2.320370 2.320370 2.320370 4.640741 2.320370<br />

a6 2.320370 2.320370 2.320370 2.320370 4.640741<br />

By using the above information, the test statistics for the three contrasts are computed as<br />

L 1 = −0.6633, L 2 = −1.1010, L 3 = 0.0734.<br />

<strong>The</strong> Schiffe criterion should be used. <strong>The</strong> critical value is given by √ 5F 5,15,0.05 = 3.8087.<br />

None <strong>of</strong> the contrasts is significant.<br />

3. <strong>The</strong> data form the study comparing four formulations considered in the lecture notes is given<br />

at the end <strong>of</strong> this question. Using R, fit a linear model to the data. Using the information in<br />

the fitted object, do the following:<br />

(i) By computing the means for each formulation, it is obvious form that the estimated effect<br />

<strong>of</strong> Formulation 3 is different from all the others. Check that the value <strong>of</strong> the contrast<br />

C = a 3 − (a 1 + a 2 + a 4 )/3 is −0.6489 with an estimated st<strong>and</strong>ard error 0.0883. Test the<br />

significance <strong>of</strong> the contrast at level 0.05 using Scheffe’s criterion. (Why should Scheffe’s<br />

criterion be used?)<br />

<strong>The</strong> following code is used to extract the information from the fitted object lm.fit <strong>and</strong><br />

compute the required quantities:<br />

b=lm.fit$coef[14:16]<br />

v=vcov(lm.fit)[14:16,14:16]<br />

c1 = c(-1/3,1,-1/3)<br />

t(c1)%*%b<br />

sqrt( t(c1)%*%v%*%c1 )<br />

L = t(c1)%*%b/sqrt( t(c1)%*%v%*%c1 )<br />

<strong>The</strong> computation yields the given values for C <strong>and</strong> its st<strong>and</strong>ard error. <strong>The</strong> test statistic<br />

equals −7.35599. Its absolute value, compared with √ 3F 3,8,0.05 = 3.492641, is not<br />

significant.<br />

6


(ii) Check that, <strong>of</strong> the six pairwise differences a j − a j<br />

′, only the three differences involving<br />

Formulation 3 are statistically significant by the Tukey criterion.<br />

<strong>The</strong> following code is used for the computation:<br />

c12 = c(1,0,0)<br />

c13 = c(0,1,0)<br />

c14 = c(0,0,1)<br />

c23 = c(1,-1,0)<br />

c24 = c(1,0,-1)<br />

c34 = c(0,1,-1)<br />

C = rbind(c12,c13,c14,c23,c24,c34)<br />

l = C%*%b<br />

ss = sqrt( diag( C%*%v%*%t(C)) )<br />

L = l/ss<br />

<strong>The</strong> computed test statistics corresponding the six pairs are:<br />

c12 0.76096856<br />

c13 -5.73428051<br />

c14 0.05461109<br />

c23 6.49524907<br />

c24 0.70635747<br />

c34 -5.78889161<br />

<strong>The</strong> critical value for the statistics using Tukey’s criterion is q 4,8,0.05 / √ 2 = 4.529/ √ 2 =<br />

3.2025. It is seen that c13, c23, c34 are significant.<br />

7


Formulation<br />

Patient 1 2 3 4<br />

1 -1.0894 (A) -1.3200 (B)<br />

2 -1.7577 (B) -0.9817 (A)<br />

3 -1.0771 (B) -1.7531 (A)<br />

4 -0.9381 (A) -1.6769 (B)<br />

5 -1.2044 (B) -0.7795 (A)<br />

6 -1.0395 (B) -1.0426 (A)<br />

7 -1.0991 (B) -0.8092 (A)<br />

8 -2.0245 (A) -1.3374 (B)<br />

9 -0.9846 (A) -1.4712 (B)<br />

10 -1.1395 (B) -1.6683 (A)<br />

11 -0.8069 (A) -1.1913 (B)<br />

12 -0.7789 (A) -1.1694 (B)<br />

8

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

Saved successfully!

Ooh no, something went wrong!