Data Analysis and Interpretation

Question 1

What are the statistical components of a research paper?

Answer 1

• Design of study
• Outcomes (primary and secondary)
• Who are the participants
• Statement of sample size required
• Analysis methods

Question 2

What are the examples of the design of studies used? Which is the most powerful evidence and what type of study is it?

Answer 2

• Interventional / Observational
• Prospective / Retrospective
• Cross-sectional / Longitudinal

The most powerful evidence is provided by a RCT (randomised controlled trial) which is a prospective interventional study

Question 3

What are descriptive statistics?

Answer 3

Summarise the profile of the participants. Usually presented as a table of numbers.

Question 4

What are inferential statistics?

Answer 4

Infer some relationship between variables.

For example: Does an ‘intervention’ change the primary outcome? The ‘intervention’ may be any therapy (drug treatment, physical therapy, mental health therapy)

Question 5

What is the chi-square test? How to know if there is a relationship between treatment and outcome?

Answer 5

Chi-square test is a way to assess whether improvement in QOL is the same or differs between intervention and control (standard care)

• If p < 0.05 then conclude that there is some relationship between treatment and outcome

Question 6

What are examples of continuous variables?

Answer 6

• Age
• Systolic BP
• Weight
• Height
• BMI

Question 7

How to get the p-values for comparison of continuous variables?

Answer 7

Work out the chance of observing the difference we actually observed (Student’s t-test)

• t = (mean₁ – mean₂ ) / SD_p

Question 8

How to calculate relative risk?

Answer 8

Question 9

How is relative risk related to p-value?

Answer 9

When doing a single 2 by 2 table, you will get the same p value for comparing relative risk as you would if doing a chi-squared test.

• If p<0.05, we conclude that the ‘risk’ of the outcome differs between control and intervention group.

Question 10

For relative risk;

A) What does a RR > 1 mean?

B) What does a RR < 1 mean?

C) What does a RR of 1 mean?

Answer 10

A)

• RR > 1 means greater chance of outcome for the intervention

B)

• RR < 1 means lower chance of outcome for intervention

C)

• RR = 1 means no association – same chance of outcome for both groups

Question 11

How to calculate odds ratio?

Answer 11

Question 12

For odds ratio;

A) What does a OR > 1 mean?

B) What does a OR < 1 mean?

C) What does a OR = 1 mean?

Answer 12

The interpretation of the OR is exactly the same as the RR

• OR > 1 Outcome more likely for Intervention compared to control
• OR < 1 Outcome less likely for Intervention compared to control
• OR = 1 Outcome is equally likely for Intervention and control

Question 13

For both RR and OR, what are we interested in why?

Answer 13

We are interested in the p-value to determine the statistical significance of any possible association

Question 14

What is the 95% confidence interval? What does it assume?

Answer 14

the range of values within which we would ‘expect’ to find the true population RR or OR

• We are 95% confident that the true RR or OR will lie within these limits, based on our particular study
• This assumes that our study participants are a representative sample of the population about which we are making this statement

Question 15

What type of studies is the odds ratio used for? What is the another name for it?

Answer 15

The Odds Ratio is used for case-control studies, and for ‘multivariate’ analyses – when we are interested in the study outcome with respect to some particular variable. In this case the name of the analysis is a “Logistic Regression” analysis

• Important thing to note is that the results of the Logistic regression are expressed as Odds Ratios, their p-values and 95% Confidence intervals
• For each variable being tested, the reference group is our choice. If there are 3 groups, we can use any one to be reference for the remaining 2 groups

Question 16

For odds ratio and relative risks;

A) When p<0.05 the 95% CI does not include?

B) If p is not quoted, can a conclusion about association be drawn from the 95% CI alone?

Answer 16

A)

• When p<0.05, the 95% CI does NOT include ‘1’

B)

• Yes

Question 17

What is the hazard ratio?

> The OR and RR relate to an outcome which is Yes/No

Answer 17

When there is a long follow-up time, we may assume that the RR for the intervention is constant over time, or that the ‘hazard’ associated with the intervention maintains the same proportion relative to the control group, across time.

> The risk is called the hazard ratio and obtained from the ‘cox proportional hazards model’

Question 18

How is the interpretation of the hazard ratio done?

Answer 18

Interpretation of the Hazard Ratio is exactly the same as for the RR and OR, and it is reported with 95% CI and p-values

Question 19

When is logistic regression used?

Answer 19

Logistic regression is used when the main outcome of interest is binary

• Yes/No, or Success/Failure, for example

Question 20

What is a univariate analysis? What are the methods used?

Answer 20

‘Univariate’ analysis means examining the data for an association between one variable and the outcome variable

• Methods used for this include Chi-square and t-test

Question 21

When is multivariate analysis used?

Answer 21

Multivariate analysis is used when we have a set of variables which may be associated with the outcome, and we want to see which of them make a separate or independent contribution to help explain the outcome

• if a variable does appear to be independently associated, we sometimes say that it is associated with the outcome ‘after adjustment’ for various other variables (which may be other clinical or demographic factors, for example)

Question 22

When is logistic regression helpful?

Answer 22

Logistic Regression is a tool by which we can identify which of a set of variables are independently associated with a binary outcome

> use it for studies with binary outcomes