PBH 5107 Biostatistics for Public Health

Assignment 4
Due: October 24, 2023

Total grade: 10. One point per question.

Question 1 (6 marks)
End stage renal disease, also called kidney failure, is a very serious condition where a
patient’s kidneys are not functioning adequately to clean the blood. These patients require
regular dialysis, a highly unpleasant treatment where the blood is artificially filtered
independent of the kidneys. Their only hope at cure is a kidney transplant operation. Because
of the low number of available kidneys for transplant, patients typically have to wait years to
get a new kidney and many end up dying before one becomes available.

a) A normal hemodialysis treatment takes about four hours and patients have to do this three
times per week. An alternative method, called high-flux dialysis, is supposed to be quicker.
Suppose you observed 30 consecutive patients receiving high-flux dialysis and recorded the
time it took to complete treatment. How would you decide whether high flux dialysis is
really faster than normal hemodialysis?

1 sample t-test if mean of sample group is statistically different from known mean. CI on
difference to demonstrate the scale.

b) There are two fundamentally different methods for performing dialysis: hemodialysis and
peritoneal dialysis. Suppose you have a random sample of 300 patients with end stage renal
disease, 140 men and 160 women. Suppose further that 75 of the men and 79 of the
women were receiving peritoneal dialysis with the rest receiving hemodialysis. How would
you determine whether or not a patient’s gender is associated with the type of dialysis he
or she is likely to receive?

Chi-square test of independence or a fisher’s exact test

c) Patients receiving hemodialysis gain weight between treatments because they retain water.
You notice that female patients complain of this more often than men. You wonder if this is
because female patients retain more water, or if it’s just because they are more bothered
by the weight gain. You recruit 150 patients to help you answer this question. You weigh
each patient right after their hemodialysis treatment and then again right before their next
 treatment and also record their gender. How would you decide whether men and women
experience different amounts of weight gain?

I would favor doing an independent 2 sample t-test on the mean change between the two
treatments. Another option, we could do a paired t-test for men and another paired t-test for
women and then do a 2-sample independent t-test between men and women. Also, depending
on whether body weight is distributed normally or if its skewed, it is possible to do a Wilcoxon
signed rank test for women and a Wilcoxon signed rank test for men and then a non-parametric
Wilcoxon rank sum test between men and women. Due to the large sample size, I would
assume it is closer to a possible normal distribution than highly skewed.

d) Some people have accused the organ transplant system of being inequitable, claiming that
rich people manage to get transplanted more quickly. If this were true, it would suggest
that people are effectively buying kidneys. You retrospectively find 100 patients with an
annual income of over $200,000 and 300 patients with an annual income of less than
$40,000 who all started dialysis treatment four years ago. You determine that 4 of the
wealthy group and 8 of the not-wealthy group have now received a kidney transplant. How
do you decide whether wealthy people don’t wait as long?

Odds ratio here would be done because you wouldn’t have enough people or power to
run a t-test of proportions. Here it’s an odds ratio instead of a relative risk because
relative risk is forming the whole population whereas an odds ratio is from a specific
selected sample size. If Odds ratio > 1 then there is an increased risk, if the Odds ratio is
< 1 then there is a decreased risk. Then we could calculate the CI and if it includes 1 then
it would mean it’s not statistically significant.

Another option would be doing a Chi-square test or fisher’s exact test as appropriate
(more fisher’s exact test because in a cell there is under 5). If less than 0.05 (p-value)
then it is statistically significant and we reject the null hypothesis.

e) You suspect that kidney failure might be associated with depression. You conduct a study
of 200 patients who have been on dialysis for varying amounts of time. You enlist 200
healthy (no kidney disease) volunteers, matched for age, gender, and socio-economic
status. You administer a questionnaire that categorizes people as (a) not depressed, (b)
slightly depressed, (c) moderately depressed, or (d) severely depressed. What statistical
test would you use to decide whether patients on dialysis are more depressed than people
not on dialysis?

Wilcoxon rank sum test (2 independent sample) numerical values would be given to each
category and then we would determine whether the results are statistically significant or not.

f) You suspect that dialysis patients are more depressed on Monday than on Friday,
something you mentally refer to as the “I hate Mondays” effect. In order to test this out,
 you ask ten patients to tell you if they feel better on Monday or Friday. Of these patients, 7
say they feel better on Friday and 3 say they feel better on Monday. How would you decide
if this is convincing evidence that your hypothesis is correct?

Chi-square goodness of fit test if our null hypothesis is that Mondays and Fridays are not equal,
we get statistical significance and rejecting the null hypothesis therefore showing that Monday
and Friday are Equal. We could also do, sign test non-numeric comparison. We are comparing
whether a person is picking between Monday and Friday which is considered as a non-numeric
comparison.

Question 2 (4 marks)
There are many different instruments available to measure depression and anxiety. The Beck
Depression Inventory and the Beck Anxiety Inventory each involve 21 different questions, each
with 4 possible answers: 0/1/2/3. In each case 0 corresponds to ‘least depressed/anxious’ and
3 to ‘most depressed/anxious’. Thus, each scale produces a number from 0 to 63 with higher
scores indicating more depression or anxiety.

(a) Suppose that you have administered both scales to each of 200 individuals and you
want to test whether people who are more depressed tend to also be more anxious.
How would you do the test?

We could do a Spearmen test for a correlation in order to determine if there is a

(b) Suppose that you have sampled 100 people who rank as ‘moderately to seriously’
depressed (a score of 19 or higher), and 100 people who are in the ‘normal’ range (score
less than 10) and you’re interested in exploring whether there is an association between
being ‘moderately to severely’ anxious and being ‘moderately to severely’ depressed.
How would you decide whether or not there is an association?

Chi-square test of independence or fisher’s exact test as appropriate.

(c) Suppose that you have devised a new mindfulness-based, non-drug intervention for
treating depression. It takes twelve weeks to complete the intervention. You recruit 40
‘moderately to severely’ depressed individuals as measured by the Beck Depression
Inventory, administer the intervention, and then administer the Beck Depression
Inventory. How would you test to see whether your study participants were less
depressed after the intervention?
 We would get data after completing the intervention for the same sample. This will give
us paired data. We will then conduct either a paired t-test or a Wilcoxon signed-rank
test in order to know whether the intervention was efficient in otherwards whether it
was statistically significant or not. Because it is an ordinal scale, I would prefer the
Wilcoxon signed-rank test than a parametric test.

(d) Suppose you want to compare three different treatment regimes: medication, cognitive
behavioural therapy, and mindfulness therapy. You randomly allocate 15 ‘moderately
to severely depressed’ patients to each type of therapy and administer the Beck
Depression Inventory before and after 8 weeks of treatment. How could you test the
null hypothesis that all three treatment regimes are equally effective?

Kruskal-Wallis test: k independent samples. This would permit us to know whether there is a difference
when more than 2 variables are involved.