Introduction to Biostatistics (171:161)

Breheny

Final (Practice)
Real final: Monday, May 12, 12:30-2:30

1. (17 points) A study of ozone exposure and asthma conducted at the University of
Southern California states that “3535 children with no history of asthma were recruited
from schools in 12 communities in southern California and were followed up for 5 years.”

(a) Is this a case-control, cohort, or cross-sectional study?

(b) Suppose the researchers want to compare the likelihood of developing asthma for
children who live in high-ozone areas vs. low-ozone areas. Can they estimate
relative risks from this study, or is this type of study incompatible with the
estimation of relative risks?

(c) Can they estimate odds ratios, or are odds ratios incompatible with this type of
study?

(d) In the study, the authors find that children who play three or more sports in high-
ozone communities are 3.3 times more likely to develop asthma than children who
play no sports. If the investigators wanted to test whether this observed increase
could be due to chance, name two tests they might they perform.

(e) Of the two tests in part (d), which test would you perform in this situation? Why?

(f) The investigators do not report a p-value for the comparison in part (d), but they
do report a 95% confidence interval of (1.9-5.8). If they were to carry out a test,
what can you say about the p-value they would obtain?

1
2. (12 points) In an effort to determine whether two drugs used in treatment of thyroid
disorders differed in terms of increasing the risk of cancer, researchers at Cambridge
performed a study in which rats were randomly assigned to receive one of the drugs.
The rats were then exposed to a known carcinogen, and the time until each rat died
of cancer was recorded. The first few outcomes for one of the groups of rats is given
below. There were 21 rats in this group.

Day Cancer Death

142 Yes
156 Yes
163 Yes
198 Yes
204 No
205 Yes
...

(a) On day 205, how many rats were at risk for dying of cancer?

(b) What was the observed probability of dying from cancer on day 205?

(c) Estimate the probability of surviving until day 160. Show your work.

(d) Below are the Kaplan-Meier curves for this study:

1.0
0.8
Probability of survival

0.6
0.4
0.2
0.0

0 50 100 150 200 250 300 350

Time (days)

What was the median survival time for group with the solid black line?

2
3. (16 points) Nausea and vomiting are frequent side effects of cancer chemotherapy, which
can contribute to the decreased ability of patients to undergo long-term chemother-
apy schedules. To investigate the capacity of marijuana to reduce these side effects,
researchers at the National Cancer Institute performed a double-blind, randomized,
cross-over trial. Fifteen cancer patients on chemotherapy schedules were randomly as-
signed to receive either a marijuana treatment or a placebo treatment after their first
three chemotherapy sessions, and then crossed over to the opposite treatment after
their next three sessions. The treatments, which involved both cigarettes and pills,
were made to appear the same whether in active or placebo form.
The outcome of interest was the number of episodes of vomiting and wretching that the
subjects experienced. The patients averaged 6.2 episodes while receiving marijuana,
compared with 38.1 episodes on the placebo. The mean of the difference between the
number of episodes the patients suffered on placebo and on marijuana was 31.9 and
the standard deviation of that difference is 35.9.

(a) What is the probability that the difference in the average number of episodes of
nausea and vomiting would be as large or larger than what was observed in the
study, just by random chance?

(b) By how much does the treatment reduce the number of episodes? Provide a 95%
confidence interval.

3
 (c) Below is a histogram of the differences in the number of episodes between the
treatment and placebo for each patient.

8
6
Frequency

4
2
0

0 20 40 60 80 100 120

Difference in episodes (placebo−marijuana)

Taking into account this histogram and the description of the study, do you have
any concerns about how you analyzed the data in (a) and (b)? Why?

(d) What might be a better way to analyze these data?

4. (21 points) The H1N1 virus was the first pandemic of the 21st century. Development
of effective vaccines for this virus was a top priority in the field of public health in
2009. In one study, researchers randomized adults 18-50 years of age into four groups,
with each group receiving vaccines for the H1N1 virus in a different way:

• A: One dose
• B: Two doses at the same time
• C: One dose, then a second dose 1 week later
• D: One dose, then a second dose 2 weeks later

One hundred adults were involved in the study, with 25 subjects randomized to each
group. To measure the immune response provoked by the vaccine, the researchers
performed an assay which measured the level of H1N1 antibodies in each subject’s
blood. The outcome of interest was the ratio of this antibody level three weeks after
their first dose compared with baseline (before receiving any vaccine).

4
(a) Instead of analyzing the ratio directly, the researchers took the log of the ratio
before conducting any hypothesis tests. Why would they have done this?

(b) Suppose the investigators wanted to test whether the mean levels of the log-ratio
from (a) were identical in the four groups. How could they have tested this?

(c) The investigators performed all possible pairwise comparisons between the groups
above at a nominal significance level of .05. They found that groups B, C, and
D all had significantly higher immune responses than group A. What is the false
discovery rate of this procedure?

(d) For the series of hypothesis tests described in (c), suppose that the investigators
wanted to limit the rate of making a single type I error to 5%. What nominal
significance level would they have to use for each individual hypothesis test?

(e) For the entire sample (all four groups combined), 42 people experienced mus-
cle aches as a side-effect of the vaccination. If you had to construct by hand a
confidence interval for the percent of the population that would experience mus-
cle aches after receiving the H1N1 vaccine, what distribution would you use to
construct the interval?

(f) With a computer, you could construct an exact confidence interval for the param-
eter of interest in part (e). What distribution would that interval be based on?

5
 (g) Would you expect the intervals from parts (e) and (f) to be close to each other?
Why or why not?

5. (3 points) One study of 22 patients who received gastric bypass surgery found that
the average systolic blood pressure of patients before the surgery was 124 mm Hg, and
114 mm Hg after. The investigators tested whether or not this reduction was due to
chance; what test do you suppose they used?

6. (5 points) The number of live births by weight and prenatal care in Harlan County,
Texas, 1986 were as follows:

Prenatal care
Birth weight Yes No
≤ 5.5 lb 3,307 234
> 5.5 lb 46,471 1461

(a) What is the probability of a baby weighing less than 5.5 pounds?

(b) What is the probability of a baby weighing less than 5.5 pounds, given that the
child received no prenatal care?

7. (8 points) Consider a study which examined the association between sleep apnea, as
measured by the apnea-hypopnea index (AHI), and diabetes, as measured by fasting
insulin levels in the blood.

(a) Is the above study an observational study or a controlled experiment?

(b) What statistical method might the researchers use to analyze their data?

6
 (c) Obesity is positively correlated with both sleep apnea and diabetes. If the above
study makes no effort to control for the effect of obesity, will the effect of sleep
apnea on diabetes be overestimated or underestimated?

8. (15 points) A 1980 study compared the indoor air quality of offices where smoking was
permitted with offices where it was forbidden. In the offices where smoking was per-
mitted, the average carbon monoxide concentration was 11.6 parts per million (ppm),
with standard deviation 7.3 ppm. In the no-smoking offices, the carbon monoxide
concentration was 6.9 ppm, with standard deviation 2.7 ppm.

(a) Suppose that carbon monoxide levels follow a normal distribution; what two tests
would be appropriate for determining whether the difference in carbon monoxide
levels is larger than you would expect by chance alone?

(b) Of the two tests in (a), which test seems more appropriate in this case? Why?

(c) If large outliers are present in the carbon monoxide levels, what test would be
better than the tests in part (a)?

(d) What kind of plot might you draw in order to inspect whether the data seemed
to follow a normal distribution or not?

7
9. (6 points) The two questions below refer to this graph, which plots the power of a
one-sample t-test versus effect size:

1.0

0.8

0.6

Power
0.4

0.2

0.0

0.0 0.2 0.4 0.6 0.8 1.0

Effect size

(a) The three lines correspond to different sample sizes of 50, 100, and 200. Which
sample size corresponds to the solid black line?

(b) For the most powerful curve, what is the smallest effect size that can be detected
with 80% power?