Biost 536 HW1
10/2/14
1.
Methods: Descriptive statistics are presented both by treatment group
(idarubicin and daunorubicin) and by the overall study population. Statistics
for continuous variables include the mean, standard deviation, minimum, and
maximum. Percentages are shown for binary and categorical variables.
Results: Data is available on 130 study participants, with 65 in both idarubicin
and daunorubicin treatment arms. Those on daunorubicin were more likely to
be male, have higher baseline measures of white blood cells and platelets,
and be alive at last follow-up. The two treatment groups were comparable in
terms of age, Karnofsky score, baseline hemoglobin, and evaluability. 12
participants were missing data on FAB classification, and one was missing for
baseline white blood cell, platelet, and hemoglobin measures.
2.
Methods: The proportion of those in each arm having complete remission
was calculated. Differences in probability of complete remission were
calculated using Pearson’s chi-squared test for independence. 95% confidence
interval was calculated using Wald statistics.
Results: 51/65 (78%) participants in the Idarubicin arm had complete
remission, while 38/65 (58%) in the Daunorubicin arm had complete
remission. Based on the 95% confidence interval, the difference in complete
remission of 20% between the two groups would not be unusual if the true
difference were between 4.4% and 35.6%. The one-sided p-value of 0.007
suggests that we can with high confidence reject the null hypothesis that
subjects taking idarubicin do not have better primary clinical outcomes than
those taking daunorubicin.
3.
Because this study is a randomized controlled trial, we might expect there to
be no confounding (including by sex). However, there are more males in the
daunorubicin arm (Table from Question 1), and there does appear to be a
relationship between sex and complete remission (table below), so sex may be
confounding the association between treatment and remission.
Female
Male
No complete remission
14
27
Complete remission
51
38
4.
Methods: The odds of having a complete remission was compared between
the two treatment arms using logistic regression. Statistical inference was
done using the Wald statistic from the regression slope parameter and
standard error, with one sided p-value and 95% confidence interval.
Results: For females, 30/35 (86%) on Idarubicin had complete remission,
compared to 21/30 (70%) on Daunorubicin (odds ratio of 2.57). For males,
21/30 (70%) on Idarubicin and 17/35 (49%) on Daunorubicin had a complete
recovery (odds ratio of 2.47). The odds of complete remission for those on
Idarubicin was 2.51 times the odds of those on Daunorubicin of the same sex.
This is beyond what might be expected by chance if there was no true
association (one sided p-value 0.011). Such an observed association would not
be unexpected if the true odds ratio was between 1.14 and 5.53.
5.
Methods: The odds of having a complete remission was compared between
the two treatment arms among males using logistic regression. Statistical
inference was done using the Wald statistic from the regression slope
parameter and standard error, with one-sided p-value and 95% confidence
interval.
Results: Among males, the odds of complete remission for those on
Idarubicin were 2.47 times the odds of those on Daunorubicin. This is beyond
what might be expected by chance if there was no true association (one sided
p-value 0.042). Such an observed association would not be unexpected if the
true odds ratio was between 0.89 and 6.88.
6.
Methods: The odds of having a complete remission was compared between
the two treatment arms among females using logistic regression. Statistical
inference was done using the Wald statistic from the regression slope
parameter and standard error, with one-sided p-value and 95% confidence
interval.
Results: Among females, the odds of complete remission for those on
Idarubicin were 2.57 times the odds of those on Daunorubicin. This is not
beyond what might be expected by chance if there was no true association
(one sided p-value 0.065). Such an observed association would not be
unexpected if the true odds ratio was between 0.75 and 8.77.
7.
Methods: The odds of having a complete remission was compared between
the two treatment arms using logistic regression. In order to assess effect
modification by sex, sex and a sex-treatment interaction were added to the
model. Statistical inference was done using the Wald statistic from the
regression slope parameter and standard error, with one-sided p-value and
95% confidence interval.
Results: For females, 30/35 (86%) on Idarubicin had complete remission,
compared to 21/30 (70%) on Daunorubicin (odds ratio of 2.57). For males,
21/30 (70%) on Idarubicin and 17/35 (49%) on Daunorubicin had a complete
recovery (odds ratio of 2.47). There is a 4.0% decrease in the odds ratio of
treatment effect on complete remission for males compared to females. This
is not beyond what might be expected by chance if there was no true
difference in the effect of the treatments by sex (p-value = 0.96).
8.
Based on the results of problem 7, idarubicin use seems to be associated with
more frequent induction of remission. However, sex does not seem to modify
this association.
9.
Descriptive statistics for probability of complete remission by treatment group
and sex are below:
Idarubicin
Daunorubicin
Female
30/35 (86%)
21/30 (70%)
Male
21/30 (70%)
17/35 (49%)
For problem 5, the model just included males, so it would be the same values
as the descriptive statistics for males (70% and 49%). For problem 6, it is just
females, so it would be the same as the descriptive values for females (86%
and 70%). For problems 4 and 7, it would be the same four values as the
descriptives table.
10.
Although there may be some confounding by sex, the adjusted and
unadjusted ORs were fairly similar. Adding the interaction term into our
regression model also suggested that sex did not modify the effect of
treatment. Thus, the unadjusted model should be used to decide whether
idarubicin should be approved for the indication of AML. Looking at males or
females only reduces the sample size of the dataset being analyzed.
Unnecessarily adding variables such as an interaction term or confounders can
hurt the model as well.