PBH 5107 Biostatistics for Public Health

Assignment 3
Due: October 17, 2023

Total grade: 14. One point per question.

Question 1. The compound in chili peppers that is responsible for the burning sensation is called
capsaicin. In a study published in 2016 [Food Chemistry 210 (2016) 606-612], a group of researchers at
Arizona State University measured the capsaicin content of one hundred samples of each of three
different kinds of chilis, the jalapeño, the habanero, and the ghost pepper. The results were given in
Scoville heat units as follows.
Jalapeño Habanero Ghost Pepper
Mean 21,700 253,000 536,000
SD 15,500 101,000 301,000
Median 19,300 266,000 519,000
Minimum 228 3,600 22,200
Maximum 62,900 502,000 1,250,000

The paper also found that the distribution of heat ratings was different for the three varieties: they
reported that it was roughly normal for the habanero, skewed for the ghost pepper, and highly skewed
for the jalapeño.

a) Suppose that Donald Trump and Ron De Santis both have a large patch of habaneros in their
gardens and Donald Trump claims his chilis are “much, much hotter than Ron’s.” In particular,
he says that, “the difference is huge: at least a hundred thousand Scoville units!” Ron calls
Trump a liar and challenges him to prove his assertion. Assuming that Trump was able to get a
random sample of habaneros from each garden (equal numbers from each), and assuming he
wanted to be at least 90% sure of proving his chilis to be at least 20,000 Scoville units hotter
than those of De Santis, how many chilis would he need?

Calculators: Inference for Means: Comparing Two Independent Samples

Mu1: 20’000

Sigma: 101’000

Power:90%

Sample size for 1 sided test is 437 habaneros (chillis)to prove Trump’s habanero chilies are hotter with
90% power.

Here we are using 1 sided because Trump wants to prove it is 20’000 Scoville units hotter therefore it is
in one direction not in 2 directions of a difference.
2 sample t-test ( 1 sided is fine)

b) Suppose that you have grown a large number of jalapeño peppers would like to do a one-sample
t-test to test whether or not yours are hotter than those reported on in Food Chemistry. You’ll
have to send some of yours to a lab to measure each one in terms of Scoville units. How many
should you send to ensure that your t-test is valid?

Jalapeño
Mean 21,700
SD 15,500
Median 19,300
Minimum 228
Maximu
62,900
m

(Highly skewed for Jalapeños)

With the Jalapeños pepper, the distribution is highly skewed, we have to have at least sample size of 30
Jalapeños (or 60 is more realistic according to the professor) in order to have a valid t-test for a highly
skewed distribution.

Distribution highly skewed for Jalapeños, at least 30 is good but at least 60 will do. We know highly
skewed distribution , if u do a t-test make sure it’s at least 30.

Wilcoxon ranked-test if less than 30. ( if parametric)

Question 2. Because chilis cause a sensation of pain, many people have assumed that a diet high in chili
must cause some damage. An article published in 1994 [American Journal of Epidemiology 139 (3) 263-
271] detailed the results of a case-control study conducted in Mexico to assess a possible link between
eating chili and gastric cancer. The study looked at 220 cases of gastric cancer and 752 controls without
gastric cancer, and asked each study participant was administered a food questionnaire. A total of 145
study participants reported not eating chili regularly, while the rest did eat chili regularly.

a) The study found that people with gastric cancer had an odds of 23.4 of eating chili regularly
while the people without gastric cancer had an odds of eating chili regularly of only 4.2. The
odds ratio between the two groups was highly statistically significant, indicating strong statistical
evidence that people with gastric cancer are more likely to eat chili regularly than people
without gastric cancer. Based on this evidence, what can we say about whether or not eating
chili regularly is a risk factor for gastric cancer? Make sure you look at the slides about case
control studies from lecture 5 before answering this question.

+ Gastric Cancer - Gastric Cancer Total

The data do not provide evidence of a difference between whether they eat chilli or not. In other words,
we cannot estimate the relative risk but we can only estimate the odds ratio.
We cannot conclude causation of eating chilli regularly for getting gastric cancer

We can use t-test and difference in proportions, if p-value is less than 0 it would mean that it is
statistically sign. Also, we can do Chi-square of independence or fisher’s exact test as appropriate
Or I would calculate CI for difference in proportions and if it does not include 0 , there is evidence that
it’s not equivalent
Or CI for RR and if it does not include 1, it’s not equivalent ( meaning there is a difference)
Or CI for OR and if it does not include or equivalent to 1, then it’s not equivalent (meaning there is a
difference)

Simplest way to anwer: it’s an odds ratio but case control study , so odds ratio for eating chilli regularly
comparing ppl gastric cancer or no gastric cancer is the same … data provide strong significance
evidence that eating chilli is associated with chilli is associated…

Statistically sign. Evidence of an association…

b) Assuming you had access to all the data collected in the study, could you estimate the
prevalence of gastric cancer in Mexico? Explain how you would estimate it, or why you couldn’t.
You may assume you have any data you like that could have been measured on the study
participants.

Sampling model of a case control study is not designed to represent the population. Therefore we
cannot estimate the prevalence of gastric cancer in Mexico with only data from a case control study. In
order to look at prevalence we either need a population or a representative random sample of the
population.

Cannot estimate prevelance from canse controle study bcz not representative sample

Question 3. Let’s suppose that you’re interested in confirming a study by Ahuja et al [Am J Clin Nutr.
2006;84(1):63-69] that found that a meal with lots of chili was associated with reduced postprandial
hyperinsulinemia (that’s a spike of insulin in the blood after a meal, which Ahuja et al found to be
reduced). You recruit 40 healthy people to help you compare the difference between eating a bland (no
chili) meal or a spicy meal (same as bland, but with lots of chili). You measure the outcome by plotting
blood insulin concentration against time for 120 minutes after the meal and calculating the area under
the curve. Higher values correspond to more hyperinsulinemia.

You consider two possible study designs. In the first (RCT) you randomize 20 to the bland meal and 20
to the spicy meal and measure the outcome. In the second (crossover), you repeat the meal on a
second day but now each person gets the opposite meal from the one they had the first day. You
measure the outcome again.
 a) In the RCT design, how would you decide whether a spicy meal was associated with a different
level of postprandial hyperglycemia compared to a bland meal?

Independent sample t-test if it’s statistically significant, also a CI which could permit us to describe the
range of the CI between bland meal and spicy meal (if it includes 0 it will be non-statistically significant)
We could do a relative risk calculation in order to understand implication of the effect of the spicy meal
in terms of proportional increase in relation to the bland meal.

No relative risk …
We can do wilcox ranked sum test
Confidence interval on difference in emans and exclude 0

b) In the crossover design, how would you decide whether a spicy meal was associated with a
different level of postprandial hyperglycemia compared to a bland meal?

Paired sample t-test in order to compare the difference in effect.

Can do paired t test , wilcoxon signed rank test

Question 4.
a) Based on the observation that eating chili seems to be more popular in hot countries than in
cold, it has been hypothesized that eating chili may help with temperature regulation (ie. it
might help cool you down). Suppose that you recruit a sample of 60 competitors in a marathon
held every April in the desert of Death Valley. Prior to the beginning of the race, you give half of
the runners a capsule containing pure capsaicin and the other half a capsule containing sugar.
Halfway through the marathon, you measure the skin temperature on the back of the right hand
of each of the competitors. What test could you use to investigate whether or not the capsaicin
was effective at cooling down the runners?

2 sample t-test for independent sample

Or Wilcoxon sign test , or CI

b) Suppose that you find that there is no effect of the capsaicin, but based on past experience you
believe that women seem to be complain less about heat than men. You would like to test
whether or not there is a difference in skin temperature between the sexes. What do you need
to do prior to conducting a statistical test?

- Need to make sure there are no confounding variables in order to make sure that the difference
is not due to something else than sin temperature and sex.
- Defining what to test, establishing a null hypothesis/alternative hypothesis and knowing how
much power we want (by using same data from question 4a
 How many men and how many women do I have , if equal the 30 men and 30 women, can use t
test … if it’s 55 men or 5 women cant use t test .. but can use wilcoxon

c) Regardless of whether or not there is an objective difference between skin temperature, it still
could be a fact that women complain less than men. Since you videotaped the skin temperature
test (a), you are able to determine who complained about the heat. How could you decide
whether there is statistical evidence of a difference between men and women with regards to
complaining?

We can use a chi-square of independence or fisher’s exact test as appropriate

Dichotomous outcome
Can do CI , relative risk , odds ratio , risk difference , t test proportions ,

d) Give a different way to answer part (c).

t-test on the difference of proportions of complaining versus not complaining.

e) Give yet a different way to answer part (c).

We could calculate the CI at 95% and see whether it contains a 0 or not, if it does not contain a 0
then we can assume that statistically sign that it was effective because there was a difference.
relative risk calculation and if it is not equal to 1 then there is a difference.
Odds ratio calculation

f) There are many documented reports of indigenous groups in Mesoamerica using chili topically
to prevent and treat skin infections, leading to a widespread belief that chili may have
antimicrobial properties. You are working in a clinic in a small village that has been affected by a
hurricane and you have 18 people who all have experienced small skin lacerations requiring
stitches but you do not have any antibiotics. You take this opportunity to test out the value of
chili by randomly applying fresh chili juice to half of the wounds prior to stitching them up. The
rest, you simply wash as well as you can with soap and water. Over the next week, you observe
that almost half of your patients experience wound infections. How could you decide whether
or not the chili helped?

Contingency table + Chi-square of independence or fisher’s exact test as appropriate

g) Some people have hypothesized that capsaicin may have value as an anti-obesity drug.
Proposed mechanisms for this include increased energy expenditure, lipid oxidation, and a
decrease in appetite. Some or all of this effect could be due to placebo: people who are trying
to lose weight may be more successful if they believe they are taking a drug that helps. You
recruit 150 people who have enrolled in a six-month weight loss program. After five months,
 you randomize (50 per group) the participants to (1) receive a daily capsule of capsaicin, (2)
receive a daily capsule of sugar, or (3) receive nothing. After one month, you record whether or
not each participant lost weight over the last month. Suppose you do a chi-square test of
independence comparing the three groups and it is not statistically significant. Explain in plain
language what you would conclude.

Therefore, there is not enough evidence on weight loss in relation to capsaicin or placebo affect in order
to infer causality.

h) Unfortunately, you fail to find even a suggestion that chili helped people lose weight, however it
did appear that maybe the placebo effect was real: people who took a pill might have been
more likely to lose weight in their last month (last of 6) than people who didn’t. Suppose you
wanted to do another randomized controlled trial comparing a placebo pill to no pill to help
people lose weight in month six. Suppose you wanted to have 80% power to detect a benefit
associated with the placebo pill. What information would you need to calculate a sample size?

We need to know the expected difference we want to detect, the standard deviation in order to
test an independent 2 sample t-test. For example, suppose I want to detect a difference in
weight of 14 pounds, assuming that the standardbred deviation is 30 pounds, and we want to
detect an 80% power we would need 57 individuals for each sample (sample size).

I need to also know wha the outcome is , is it the actual amount of weight they lost…