DESCRIPTIVE COMPARISON OF GROUPS - LEARNING OUTCOMES

Question 1

What are the descriptive statistics for categorical data?

Answer 1

Percentages and proportions

Question 2

What are the descriptive statistics for normally distributed continuous data?

Answer 2

Mean and standard deviation

Question 3

What are the descriptive statistics for non-normally distributed continuous data?

Answer 3

Median and Interquartile range

Question 4

What is meant by the ‘mean difference’?

Answer 4

Mean difference = Mean of exposed group - Mean of unexposed group

If the confidence interval for the mean difference spans 0 then we cannot confirm a real difference between the two groups because it means that no difference is a plausible value.

Question 5

How do we examine the mean difference when we have more than 2 exposure groups?

Answer 5

We first decide which is the unexposed (baseline) group. For an ordered variable this is usually the first or last group. We avoid groups with very small numbers. The mean difference and CI is then calculated as normal, relative to the baseline.

Question 6

For what kind of data is the mean and mean difference an appropriate statistic?

Answer 6

Continuous data

Question 7

What are the only kind of studies that risk statistics are applicable for?

Answer 7

Risk statistics are applicable for cross-sectional studies only.

Question 8

How do we calculate the overall risk of disease?

Answer 8

Overall risk (prevalence) = Number of people with disease / total number of people in the study population

Question 9

How do we calculate the risk of disease for a particular group, e.g. smokers?

Answer 9

Risk = number of smokers with disease / total number of smokers

Question 10

How do you calculate the risk ratio?

Answer 10

Risk ratio = Risk of disease in exposed / risk of disease in unexposed

Question 11

What is the risk difference?

Answer 11

Risk difference = Risk of disease in exposed - Risk of disease in unexposed

For example:

If the risk in diseased is 19% and the risk in unexposed is 4% then we would calculate 19%-4%=15% and this would tell us that the exposed group have 15% more disease than the unexposed group.

Question 12

What is the benefit of looking at both the risk ratio and the risk difference rather than than the risk ratio alone?

Answer 12

The risk ratio may tell us we have say a 4-fold increase but this may actually only equate to a fraction of a percent difference in reality, which can be seen if we also calculate the risk difference.

Question 13

What does a risk ratio of 1 mean?

Answer 13

No change in risk

Question 14

What does a risk ratio less than 1 mean?

Answer 14

Decrease in risk

Question 15

What does a risk ratio of greater than 1 mean?

Answer 15

Increase in risk

Question 16

What is the easiest way to carry out an interpretation of risk ratio?

Answer 16

Look at the deviation from 1, which represents no change in risk - e.g. 1.5 - 1 = 0.5 which means 50% more likely
0.4 - 1 = -0.6 which means 60% less likely

Question 17

What is the Odds ratio?

Answer 17

Odds ratio = odds of disease in exposed / odds of disease in unexposed

Looks at odds rather than absolute risk.

Here we are not using group totals, rather we are comparing disease in exposed to disease in unexposed.

Question 18

How do we calculate odds?

Answer 18

Odds = number of people with disease / number of people without disease

Question 19

What measure of risk would we give for a case-control study?

Answer 19

The odds ratio - we can only use risk ratios for cross-sectional studies

Question 20

For rare diseases the risk ratio and odds ratio should give you ________ values.

Answer 20

For rare diseases the risk ratio and odds ratio should give you similar values.

(rare means

Question 21

How do we calculate risk ratios or odds ratios if we have more than 2 groups?

Answer 21

We choose a baseline group and calculate statistics relative to that group.

Question 22

If our 95% CI for risk ratio or odds ratio spans 1 what does this mean?

Answer 22

It means that we can’t say that there is any real difference between the groups. Because our value of 1 is in the plausible range we can’t be confident that there is a real difference there.

Question 23

Risk ratio is only valid for _________ studies. For case-control studies we must use the __________.

Answer 23

Risk ratio is only valid for cross-sectional studies. For case-control studies we must use the odds ratio.

Question 24

If we are correcting for other factors is it easier to use the risk ratio or the odds ratio?

Answer 24

If you are correcting for other factors it is easier to use the odds ratio.

Question 25

What descriptive statistics are appropriate for continuous variables that are normally distributed?

Answer 25

The mean difference an confidence intervals

Question 26

What descriptive statistics are appropriate for categorical variables?

Answer 26

Risk difference, risk ratio and odds ratio

Question 27

For continuous variables that are not normally distributed what descriptive statistic / measure of difference would be appropriate?

Answer 27

The median difference, but it is not as easy to calculate confidence intervals for this.

Question 28

What differences might we see in the risk ratio and odds ratio for more common diseases?

Answer 28

The risk ratio and odds ratio for common diseases will vary more for common diseases than rare diseases and the odds ratio will tend to give us an overestimation of the true risk.

Question 29

Why is the odds ratio appropriate for case control studies, but the risk ratio is not?

Answer 29

In case-control studies the risk ratio outcome will change greatly depending on the number of cases and controls recruited. Risk ratio depends on the number of cases / controls in the study, which is fixed by the researchers. The odds ratio depends on proportion of cases / controls with exposure. The risk ratio is only appropriate in studies where participants are recruited independently of disease status. The odds ratio is also more commonly used because it is much easier to adjust in a multivariate analysis.