Homework code: 1896
1. Descriptive, univariate analysis was performed to determine characteristics
of age, gender, laboratory results, FAB classification and Karnofsky score of
the two treatment arms. To determine similarity between treatment arms,
inferential testing was performed using chi-squared test for gender and
unpaired t-test for other variables, see Table 1.
Table 1: Characteristics of treatment arms
Treatment
type
Age
[range]
Gender
(%)
WBC
(SD)
Platelets
(SD)
Hgb
(SD)
FAB class
[95% CI]
Idarubicin
38
[17-58]
30 male
(46%)
29
(36.3)
66.6
(57.8)
9.2
(1.8)
2.95
[2.56-3.34]
Karnofsky
score
[95% CI]
79.5
[76.7 – 82.4]
Daunorubicin
39.8
[19-58]
35 male
(54%)
43.3
(55)
93.6
(92.4)
9.6
(1.5)
3.19
[2.8 – 3.6]
79.5
[76.4 – 82.6]
p-value
0.43
0.38
0.08
0.05
0.16
0.38
1.0
2. Using an intent to treat analysis, treatment arms were compared regarding
the primary endpoint, which was induction of complete remission. Risk
ratios were calculated for measure of association. Inferential testing was
performed using chi-squared test.
Among all 130 patients enrolled in the trial, the 75 patients who received
idarubicin were 1.33 times more likely to experience induction of complete
remission when compared to the 75 patients who received daunorubicin,
95% CI [1.05 – 1.7], p-value 0.014.
3. Decisions regarding adjustment for confounding should be made a priori
based on scientific knowledge and plausibility. Gender frequently influences
the effects of treatment and should be adjusted for if possible.
4. Logistic regression was performed to assess for any treatment benefit of
idarubicin over daunorubicin while adjusting for gender. Inferential testing
was based on likelihood ratio test statistic.
After adjustment for gender, the 75 patients who received idarubicin had an
odds of 2.5 of achieving complete remission when compared to the 75
patients who received daunorubicin, 95% CI [1.19-4.66], p-value 0.02.
5. To determine whether there was any treatment benefit for males receiving
idarubicin, the treatment groups were stratified by gender. The frequency of
primary endpoint of complete remission was compared between males who
received idarubicin to those who received daunorubicin. Inference testing
was performed using Pearson’s chi-squared test.
Among the 30 males who received idarubicin, 21 (70%) achieved primary
endpoint of induction of complete remission; in comparison, among the 35
males who received daunorubicin, 17 ( 48.5%) experienced complete
remission, p-value 0.08.
6. Similar stratified analysis was performed to determine whether there was
any difference in primary endpoint of complete remission among females
who received idarubicin compared to daunorubicin.
Among the 35 females who received idarubicin, 30 (85.7%) achieved
primary endpoint of induction of complete remission; in comparison, among
the 30 females who received daunorubicin, 21 (70%) experienced complete
remission, p-value 0.12.
7. To determine the extent to which treatment benefit differs by sex, stratified
risk ratios were calculated based upon gender classification. Inference
testing was performed using chi-squared test. Mantel-Haenszel test for
heterogeneity was used to determine if there was a significant difference of
treatment benefit between gender groups.
Stratified analysis revealed that the 35 women who received idarubicin were
1.22 times more likely to experience induction of complete remission when
compared to the 30 women receiving daunorubicin, 95% CI [0.93-1.6]. In
comparison, the 30 men who received idarubicin were 1.44 times more likely
to experience induction of complete remission compared to the 35 men who
received daunorubicin, 95% CI [0.95-2.18].
Though men appeared to have a greater treatment benefit with idarubicin,
the difference in treatment benefit between gender groups did not achieve
significance with inferential testing, p-value 0.5.
8. To account for any difference in treatment benefit by sex, a Mantel-Haenszel
estimate was calculated as a common measure of association across the
stratified categorical data.
After adjustment for gender, the 75 trial participants who received idarubicin
were 1.31 times as likely to experience primary endpoint of induction of
complete remission as the 75 participants who received daunorubicin, 95%
CI [1.04 – 1.66].
9. Estimates of probability of complete remission were calculated for all
combinations of treatment group and sex. Exact confidence intervals were
calculated for logistic regression estimates, while Wilson-based confidence
intervals were calculated for M-H estimates.
Logistic Regression (MLE)
Treatment
Probability Estimate
Female
Male
Idarubicin
.867
[0.69 - .96]
.714
[0.54 – 0.85]
Daunorubicin
.71
[0.54 – 0.85]
.5
[0.31 – 0.69]
OR
[95% CI]
2.57
[0.73 – 9.1]
2.47
[0.86 – 7.11]
Mental-Haenszel estimate (score)
Treatment
Probability Estimate
Female
Male
Idarubicin
.867
[0.7 – 0.94]
.714
[.55 - .84]
Daunorubicin
.71
[0.55 – 0.84]
.5
[.33 - .67]
RR
[95% CI]
1.22
[.93-1.6]
1.44
[.95 – 2.18]
There are some slight differences between the treatment arms. The numbers
of males and females differed, with 35 men and 30 women in the
daunorubicin arm, and 35 women and 30 men in the idarubicin arm. The
daunorubicin arm also had higher laboratory values, although this difference
did not quite reach significance. Overall the groups appear well-matched at
baseline.
When evaluating estimates of probability of achieving complete remission,
women in both treatment arms had a higher probability of achieving
remission. In fact the probability of women in the daunorubicin arm
achieving remission is similar to the probability of men in the idarubicin arm
achieving remission. Overall among both genders, the idarubicin treatment
arms had a higher probability of achieving remission.
10. It would be best to report the Mantel-Haenszel results, which are based upon
score test. Score test will improve mean-variance relationship, which is an
important consideration in categorical analysis of groups with relatively
small sample sizes. In addition, when possible, it is best to report cohort or
trial results in terms of relative risk or risk difference. Based on our stratified
analysis, we can say that patients who receive idarubicin are 31% more likely
to achieve complete remission that subjects who receive daunorubicin, 95%
CI [4% - 66%].
.do file
infile ptid str10 onstudy str1 tx str1 sex age fab karn wbc plt ///
hgb str1 eval str1 cr crchemo str10 crdate str10 fudate
///
str1 status str1 bmtx str10 bmtxdate str1 incl
///
using http://www.emersonstatistics.com/datasets/leukemia.txt
g onstJ= date(onstudy, "MD19Y")
g fudtJ= date(fudate, "MD19Y")
g obstime= fudtJ - onstJ
g male= .
replace male=1 if sex=="M"
replace male=0 if sex=="F"
g treat= .
replace treat=1 if tx=="I"
replace treat=0 if tx=="D"
g eva= .
replace eva=1 if eval=="Y"
replace eva=0 if eval=="N"
g creat= .
replace creat=1 if cr=="Y"
replace creat=0 if cr=="N"
g outcome = .
replace outcome=1 if status == "D"
replace outcome=0 if status == "A"
g transplant = .
replace transplant=1 if bmtx == "Y"
replace transplant=0 if bmtx == "N"
**"Question 1"
cs treat male
ttest age, by(treat)
ttest karn, by(treat)
ttest fab, by(treat)
**"Question 2"
cs creat treat
** "Question 3"
** "Question 4"
logistic creat treat male
estimates store full
quietly logistic creat treat
lrtest full
** "Question 5"
count if male == 1 & treat == 1 & creat == 1
count if male == 1 & treat == 1 & creat == 0
count if male == 1 & treat == 0 & creat == 1
count if male == 1 & treat == 0 & creat == 0
csi 21 17 9 18
** "Question 6"
count if male == 0 & treat == 1 & creat == 1
count if male == 0 & treat == 1 & creat == 0
count if male == 0 & treat == 0 & creat == 1
count if male == 0 & treat == 0 & creat == 0
csi 30 21 5 9
** "Question 7 & 8"
cs creat treat, by(male)
** "Question 9"
cii 35 .7, binomial
cii 30 .485, binomial
cii 30 .857, binomial
cii 35 .7, binomial
cii 35 .7, binomial wilson
cii 30 .485, binomial wilson
cii 30 .857, binomial wilson
cii 35 .7, binomial wilson