Academic

Question 1

What is parallel group comparison?

Answer 1

Different groups receive different interventions and are started at the same time. Different groups are then compared

Question 2

What is paired/matched comparison?

Answer 2

Subjects receive different treatments and are matched to mitigate confounding variables (age). Results are analysed by analysing differences between subject pairs (same age, sex etc)

Question 3

What is within subject comparison?

Answer 3

Comparison made between the SAME subject: - before specified intervention - after specified intervention

Question 4

What is single blind?> Double blind?

Answer 4

Single Blind: Patients dont know which intervention they are receiving Clinicians do know Double blind: Both patients and clinicians dont know

Question 5

What is a crossover study design? What is a washout period?

Answer 5

Crossover: each subject receives intervention and control treatments (often randomised into who receives control/intervention first) Washout period: period in between receiving control/intervention

Question 6

What is factorial design?

Answer 6

More than one independent variable given. (Placebo, aspirin, streptokinase or aspirin +streptokinase)

Question 7

Best trial design to test for efficacy of a drug?

Answer 7

RCT

Question 8

Best study design for assessing new diagnostic test?

Answer 8

Cross sectional survey (using both new and gold standard test)

Question 9

Best study design for assessment of screening method?

Answer 9

Cross sectional survey

Question 10

Best study design for disease prognosis?

Answer 10

Longitudinal cohort study

Question 11

Best study design for demonstrating causation?

Answer 11

Cohort Case Control (Case reports can be useful)

Question 12

What is: i) Incidence Ii) Prevalence

Answer 12

I) Incidence - is the number of new cases of a disease per year ii) prevalence - overall proportion of the population who suffer from the disease

Question 13

Compares two independent samples drawn from same population I) parametric Ii) non-parametric

Answer 13

I) Two sample (unpaired) t test Ii) Mann Whitney U test

Question 14

Compares two sets of observations on a single sample I) Parametric Ii) non-parametric

Answer 14

I) One sample (paired) t test Ii) Wilcoxton matched pairs test

Question 15

Compares three or more observations on a single sample i) parametric Ii) non parametric

Answer 15

I) One way analysis of variance using total sum of squares Ii) analysis of variance by ranks

Question 16

Tests influence of two variates on observation set i) parametric Ii) non-parametric

Answer 16

i) Two way analysis of variance Ii) Two way analysis of variance by ranks

Question 17

Tests null hypothesis that proportions of variables estimated from two + INDEPENDENT SAMPLES are the same

Answer 17

Only non- parametric - x2 test

Question 18

Test null hypothesis that proportions of variables estimated from a PAIRED sample are the same

Answer 18

Non-parametric test only - McNemar’s Test

Question 19

Assess the strength of straight line association between two continuous variables i) parametric Ii) non-parametric

Answer 19

i) parametric - Product moment correlation coefficient - PEARSON’S r Ii) non-parametric - Spearman’s rank coefficient gradient

Question 20

Describes the numerical relation between two quantitative variables, allowing one value to be predicted from another

Answer 20

i) parametric ONLY —> Regression by least squares method

Question 21

Describes the numerical relation between a dependent variable and several predictor variables

Answer 21

i) parametric - multiple regression by least squares method

Question 22

How do non-parametric tests work?

Answer 22

By using the rank order of the data points rather than the difference between them

Question 23

What do you do when faced with subgroup analysis

Answer 23

Consult a book such as Oxman and Guyatt - “a consumers guide to subgroup analysis”

Question 24

When to use paired test?

Answer 24

When for each subject something is measured twice

Question 25

When to use two tailed test?

Answer 25

Two tailed tests should generally be used as these look for extremely low/high readings Single tailed tests are only appropriate if very convincing evidence that data is going ONE way (i.e. Low and not high )

Question 26

What are the Bradford Hill Criteria?

Answer 26

For assessing causality: - Human Experiments - Strong association - consistent association between studies - temporal relationship (cause precede effect?) - Dose-response gradient - make epidemiological sense? - make biological sense? - Specific association? - Analgous to previously proven causation?

Question 27

Regression vs correlation?

Answer 27

Correlation: - Essentially determines whether a relationship between two/more sets of data is present - Pearson's rank / Spearman coefficient - Two normally distributed data sets, with two structurally independent variables (otherwise would be implicitly correlated) - with sufficiently strong confidence intervals Regression : - Mathematical term for one variable to be predicted from another - Multiple regression calculates by computers from inputting data swrs

Question 28

Relative Risk Calculation? Use: Control Outcome (CO) Experimental Outcome (EO)

Answer 28

First for Both CO and EO need to transform into ratios so COR and EOR: Do this by dividing the CO/population and EO by the population. Then Relative Risk Ratio: Is the ratio of EOR vs COR. So for something like death which wnats to be prevented want it to be \<1.0. So RR = EOR/ COR

Question 29

Relative Risk Reduction Calculation Using : Control Outcome ( CO ) Experimental Outcome (EO)

Answer 29

First for Both CO and EO need to transform into ratios so COR and EOR: Do this by dividing the CO/population and EO by the population. Looking at the reduction of risk in the intervention group versus control group. Then: COR-EOR/ COR

Question 30

Absolute Risk Reduction Calculation Using: Control Outcome (CO) Experimental Outcome (EO)

Answer 30

First for Both CO and EO need to transform into ratios so COR and EOR: Do this by dividing the CO/population and EO by the population. Now this risk calculation is NOT the relative risk. So simply need to find the difference in risk ratio between the control and experimental groups. COR-EOR = Absolute risk reduction ratio Absolute risk reduction ratio x 100 = Absolute risk (%)

Question 31

How do you calculate the number needed to treat?

Answer 31

Need the absolute risk reduction. Reciprocal of absolute reduction = NTT

Question 32

Calculate odds ratio. Using: Control Outcome (CO) Experimental Outcome (EO)

Answer 32

First for Both Control group and Experimental group need to transform into ratios against the alternative outcome! So Let's call the endpoint outcome (something like death) CO1 / EO1 and the other outcome (survival past 10 years) CO2 / EO2 Then need to get the odds of dying vs odds of surviving = CO1/CO2 and EO1/ EO2 respectively Let's call CO1/CO2 = COR EO1/EO2 = EOR COR/EOR. This in particular would show the odds of dying in the control group vs experimental group. If it was greater in the Control group \>1 .

Question 33

How to establish causation for rare adverse drug reaction?

Answer 33

Ideally by giving the same patient the same drug in controlled conditions

Question 34

What to look out for in comparitive drug studies for a new drug trial?

Answer 34

- Comparison against placebo! Not good! SHould be comparing against current gold standard treatment - Watch out for surrogate endpoints

Question 35

Three questions to ask yourself when making decisions about prescribing

Answer 35

i) Ultimate objective of treatment ii) Most appropriate treatment (based on research) iii) Treatment Target

Question 36

Surrogate endpoint?

Answer 36

A surrogate endpoint is a (easily measured) variable used in clinical trials where the actual outcome being measured is difficult/expensive/long. Surrogate endpoints can be - drug metabolite concentration laboratory measurements radiological findings biological markers of disease (troponin, CK, CRP etc) Macroscopic tissue appearance Problem is - Often they don't accurately predict harm/clinical benefit