Descriptive Comparisons Of Groups

Question 1

What descriptive statistics would be used for continuous data that was normally distributed?

Answer 1

Mean and standard deviation

Question 2

What descriptive statistics would be used for continuous data that was not normally distributed?

Answer 2

Median and interquartile range

Question 3

How would you determine if the outcome variable was normally distributed?

Answer 3

Draw a histogram (either for the whole sample or in two groups separately)

Question 4

Our data shows men have a mean forced expiration volume (FEV) of 3.19 litres compared to that of 2.44 litres in women. What can we interpret from this data?

Answer 4

That women have a lower mean FEV compared to men

Question 5

Constructing a confidence interval around the mean can help us determine whether there is a true difference between the groups. How does this work?

Answer 5

If the confidence intervals for both means do not overlap then you can be fairly confident there is a true difference

Question 6

When the confidence intervals for your two means do overlap, what should you interpret instead?

Answer 6

The difference between the two means (mean difference) and the 95% CI for that difference

Question 7

Our mean difference is 0.64 and our CI is 0.45-0.85. What does this tell us?

Answer 7

On average group1 have a higher variable than group2

The CI tells gives us an idea of how good the estimate is compared to our true mean difference. Here, because it is above zero it indicates this is likely to be a true difference

Question 8

For normally distributed data we compute the difference in means. What do we compute for non-normally distributed data? What is the problem surrounding this?

Answer 8

We compute the difference in medians. However, the 95% CIs around medians or difference in medians cannot easily be computed

Question 9

How would you calculate the mean difference and CIs for a situation with more than two exposure groups?

Answer 9

• decide which is the unexposed (baseline/reference)

- calculate the stats in the same way, relative to the baseline

Question 10

Does a CI for the mean difference that spans zero always give a non-significant result?

Answer 10

Yes

Question 11

When we have more than two exposure groups, CIs cannot be used to assess whether there is a true overall association/difference or not. What would need to be carried out instead?

Answer 11

An appropriate hypothesis test

Question 12

Which statistics can be used for a descriptive comparison between two categorical groups?

Answer 12

• Risk statistics

- Odds of developing disease (odds ratio)

Question 13

What type of study are risk statistics limited to?

Answer 13

Cross-sectional studies

Question 14

1. How do you calculate the overall risk (prevalence) in the study?
2. How do you then calculate the risk ratio?

Answer 14

1. Number of people with disease divided by total number of people in study population
2. Risk of disease in exposed divided by the risk of disease in unexposed

Question 15

How would you interpret the following risk ratios: 1, >1?

Answer 15

1 = no change in risk
>1 = increase in risk

Question 16

The risk difference can also be quoted alongside the risk ratio. How is the risk difference calculated?

Answer 16

Risk of disease in exposed minus the risk of disease in unexposed

Question 17

Why are risk ratios inappropriate for case control studies?

Answer 17

Because when sample numbers increase the risk ratio increases whereas this should stay stable, which is what happens when using odds ratio

Question 18

Interpret the following risk ratios: 3, 1.5, 0.4, 0.9?

Answer 18

3: 3 times more likely
1. 5: 1.5-1 = 0.5 = 50% more likely
0. 4: 0.4-1 = -0.6 or -60% i.e. 60% less likely
0. 9: 0.9-1 = -0.1 or -10% i.e. 10% less likely

Question 19

1. How do you calculate the odds for your sample groups?

2. How do you then calculate the odds ratio?

Answer 19

1. Number of people with disease divided by number of people without disease (NOT the total number of people)
2. Odds of disease in exposed divided by the odds of disease in unexposed

Question 20

In what scenario are the odds ratio and risk ratio similar?

Answer 20

If the disease or outcome is fairly rare

Question 21

How do you calculate the odds ratio with greater than two exposure groups?

Answer 21

Choose which group is your reference/baseline group and then compare against this.

An OR of 1 is added to the reference group to indicate its use in this regard and no CIs are calculated

Question 22

What can you say about the following OR and CI: 2.33 (0.8-6.8)?

Answer 22

The CIs span one (could be 20% less or 7-fold more) therefore we can’t be confident that there is a true difference here

Question 23

Ratio measures (risk ratios or odds ratios) are more commonly used in what type of study? Why?

Answer 23

Epidemiological studies because they are better indicators of the aetiological strength of an association than the risk difference and therefore help us to decide whether an association seen is likely to be causal or not

Question 24

Why is the risk difference usually included alongside the risk ratio?

Answer 24

Because risk ratio alone can be misleading- the risk difference brings context.

For example, our risk ratio could be 5, meaning a five fold increase which looks significant. However, the actual values are 0.005% and 0.001%, therefore the risk difference is 0.004%, which is seemingly insignificant

Question 25

Other than study design, name a key factor in making you use odds ratio over risk ratio

Answer 25

The need to control for confounders

This requires more complicated statistical analyses and only ORs can be derived from these

Because of this, most studies will only present ORs even when RRs would also be appropriate

Question 26

What descriptive statistics would be used for categorical data?

Answer 26

Percentages or proportions