Overview:�Study design and analysis� in epidemiology

Reshma Kassanjee (PhD)

University of Cape Town

SACEMA�

MMED 2024

�

000

2

Outline

• Motivation: Why are we talking about “studies”?
• A classical epidemiological study
• Study classifications
• Why study design matters
• Randomised controlled trials
• Measures of disease occurrence (incidence and prevalence) and effect (example)
• Close

​

​

​

000

3

Outline

• Motivation: Why are we talking about “studies”?
• A classical epidemiological study
• Study classifications
• Why study design matters
• Randomised controlled trials
• Measures of disease occurrence (incidence and prevalence) and effect (example)
• Close

​

​

​

000

M

M

E

D �

4

Why are we talking about “studies”?

​

modelling

M

M

E

D �

5

Meaningful

modelling of

epidemiological

data

Why are we talking about “studies”?

M

M

E

D �

6

Meaningful

modelling of

epidemiological

data

Come from studies!

• There is some underlying real-world process�(signal with some noise)
• Through observation or data collection (i.e. studies) we obtain data

Why are we talking about “studies”?

7

Questions

Insights

Modelling exercise

Why are we talking about “studies”?

8

Questions

Insights

Model

Data

Why are we talking about “studies”?

9

Why are we talking about “studies”?

10

Why are we talking about “studies”?

11

Why are we talking about “studies”?

Data on disease occurrence were obtained from studies

12

Why are we talking about “studies”?

13

Why are we talking about “studies”?

14

Why are we talking about “studies”?

• Sometimes you want to know something about one group of people

​

• Sometimes you want to know about an ‘effect’ - difference between two groups

� (e.g. how much this mortality would change if we introduced treatment)

15

Outline

• Motivation: Why are we talking about “studies”?
• A classical epidemiological study
• Study classifications
• Why study design matters
• Randomised controlled trials
• Measures of disease occurrence (incidence and prevalence) and effect (example)
• Close

​

​

​

000

16

1. Define study question
2. Select study type
3. Collect data
4. Analyse data
5. Interpret results
6. Report study and findings

Classical epidemiological study

17

1. Define study question
2. Select study type
3. Collect data
4. Analyse data
5. Interpret results
6. Report study and findings

All of this must be done ethically

Classical epidemiological study

18

1. Define study question
2. Select study type
3. Collect data
4. Analyse data
5. Interpret results
6. Report study and findings

All of this must be done ethically

Classical epidemiological study

Aim to measure disease occurrence and ‘effect’

19

1. Define study question
2. Select study type
3. Collect data
4. Analyse data
5. Interpret results
6. Report study and findings

All of this must be done ethically

Classical epidemiological study

Aim to measure disease occurrence and ‘effect’

20

Outline

• Motivation: Why are we talking about “studies”?
• A classical epidemiological study
• Study classifications
• Why study design matters
• Randomised controlled trials
• Measures of disease occurrence (incidence and prevalence) and effect (example)
• Close

​

​

​

000

21

Some common distinctions between study designs:

• Observational versus Experimental
• Longitudinal versus Cross-sectional
• Prospective versus Retrospective
• Descriptive versus Analytic
• Case-control versus Cohort
• Qualitative versus Quantitative: we will discuss only quantitative studies

Study classifications

22

Some common distinctions between study designs:

• Observational versus Experimental
• Longitudinal versus Cross-sectional
• Prospective versus Retrospective
• Descriptive versus Analytic
• Case-control versus Cohort
• Qualitative versus Quantitative: we will discuss only quantitative studies

Study classifications

E.g. COVID-19 vaccination, and/or�its effect on�severe disease

23

Some common distinctions between study designs:

• Observational versus Experimental
• Longitudinal versus Cross-sectional
• Prospective versus Retrospective
• Descriptive versus Analytic
• Case-control versus Cohort
• Qualitative versus Quantitative: we will discuss only quantitative studies

Study classifications

E.g. COVID-19 vaccination, and/or�its effect on�severe disease

Exposure

Outcome

24

Observational versus Experimental

25

• Observational
• Observe the association between some exposure and outcome without interfering with what level/category of the exposure a person receives
• Experimental
• You set the level/category of the exposure the person receives, then you observe

​

Observational versus Experimental

26

Longitudinal versus cross-sectional

27

• Longitudinal (cohort, follow-up)
• Participants are observed over time
• Cross-sectional (survey, prevalence study)
• Participants are observed at a single point in time
• Data on the exposure and outcome (and other variables) are collected

Longitudinal versus cross-sectional

28

Prospective versus Retrospective

29

• Prospective
• Follow participants forward in time
• Data collection continues after your study starts
• Retrospective
• Looks backward in time
• Typically using existing data

Prospective versus Retrospective

30

Descriptive versus Analytic

31

• Descriptive
• Describe characteristics/exposure/outcome
• Analytical
• Investigate relationships
• Test hypotheses

Descriptive versus Analytic

32

Case-control versus Cohort

33

• Case-control
• Select participants based on outcome
• Then study the association with the exposure
• Carefully choose controls to compare to cases
• Specific methods of analysis / measures that are valid
• Cohort
• Select participants based on exposure
• Then study the association with the outcome

Case-control versus Cohort

34

Some common distinctions between study designs:

• Observational versus Experimental
• Longitudinal versus Cross-sectional
• Prospective versus Retrospective
• Descriptive versus Analytic
• Case-control versus Cohort

Study classifications

35

Combining design elements

36

Outline

• Motivation: Why are we talking about “studies”?
• A classical epidemiological study
• Study classifications
• Why study design matters
• Randomised controlled trials
• Measures of disease occurrence (incidence and prevalence) and effect (example)
• Close

​

​

​

000

estimate =

37

truth

bias

random error

+

+

Why does study design even matter?

38

• Suppose we want to know the impact of a new experimental ‘immunity boosting’ tablet on a person’s risk of having any severe illness
• At a workplace, we offer free courses of the tablet, and then for six months we record any illness in ‘treated’ versus ‘untreated’

​

Example study 1

39

• Suppose we want to know the impact of a new experimental ‘immunity boosting’ tablet on a person’s risk of having any severe illness
• At a workplace, we offer free courses of the tablet, and then for six months we record any illness in ‘treated’ versus ‘untreated’

​

Observational versus Experimental

Longitudinal versus Cross-sectional

Prospective versus Retrospective

Descriptive versus Analytic

Case-control versus Cohort

Example study 1

40

• Suppose we want to know the impact of a new experimental ‘immunity boosting’ tablet on a person’s risk of having any severe illness
• At a workplace, we offer free courses of the tablet, and then for six months we record any illness in ‘treated’ versus ‘untreated’
• We find a 30% reduction in illness when ‘treated’…
• Would you use the tablet? Do you have concerns about the estimate?

​

​

Example study 1

Exposure:�Treatment

Outcome:

Illness

Attitude -�wellbeing

Exercise

Example study 1

Exposure:�Treatment

Outcome:

Illness

Attitude -�wellbeing

Exercise

Example study 1

Confounding

• A concept relevant to causal inference
• In a setting and analysis: the relationship or association between the exposure and outcome ≠ the causal effect
• There is some non-causal backdoor path connecting the exposure and outcome

Example study 1

Exposure:�Treatment

Outcome:

Illness

Attitude -�wellbeing

Exercise

44

Outline

• Motivation: Why are we talking about “studies”?
• A classical epidemiological study
• Study classifications
• Why study design matters
• Randomised controlled trials
• Measures of disease occurrence (incidence and prevalence) and effect (example)
• Close

​

​

​

000

45

• Gold standard for experimental studies that aim to estimate causal effects
• The exposure is randomly assigned to participants
• Randomisation ensure there is no systematic difference between the people with different exposures/treatment��🡪 difference in outcomes can be attributed to the exposure

​

Randomised controlled trials

Exposure:�Treatment

Outcome:

Illness

Attitude -�wellbeing

Exercise

Example study 1

47

• Controlled: appropriate comparison group/arm
• Best practice is to include masking (blinding)…
• …by using a placebo: a ‘treatment’ that has no therapeutic effect, that resembles and is administered in the same way as the treatment being tested

​

Randomised controlled trials

48

Hierarchy of evidence

https://www.nature.com/articles/s43016-021-00388-5

49

Outline

• Motivation: Why are we talking about “studies”?
• A classical epidemiological study
• Study classifications
• Why study design matters
• Randomised controlled trials
• Measures of disease occurrence (incidence and prevalence) and effect (example)
• Close

​

​

​

000

50

1. Define study question
2. Select study type
3. Collect data
4. Analyse data
5. Interpret results
6. Report study and findings

All of this must be done ethically

Classical epidemiological study

Aim to measure disease occurrence and ‘effect’

51

• Common measures of disease occurrence are:�prevalence and incidence

​

Measures of disease occurrence

52

• Common measures of disease occurrence are:�prevalence and incidence
• Prevalence: the proportion of the population with the condition at a point in time
• Incidence: how quickly new cases of the condition are occurring (incidence rate, or cumulative incidence)

​

Measures of disease occurrence

53

• Common measures of disease occurrence are:�prevalence and incidence
• Prevalence: the proportion of the population with the condition at a point in time
• Incidence: how quickly new cases of the condition are occurring (incidence rate, or cumulative incidence)

​

Measures of disease occurrence

Longitudinal versus Cross-sectional

54

• Common measures of disease occurrence are:�prevalence and incidence
• Prevalence: the proportion of the population with the condition at a point in time
• Incidence: how quickly new cases of the condition are occurring (incidence rate, or cumulative incidence)
• Sometimes prevalence is quite informative of incidence �(e.g., flu), sometimes not (e.g., HIV)

​

Measures of disease occurrence

55

Measures of disease effect

56

Measures of disease effect�Risk ratio example

• Population: People living with HIV. Outcome: Becoming virally suppressed within 1 year of treatment start. Exposure: Drug A versus Drug B

​

​

​

​

​

​

​

57

Measures of disease effect�Risk ratio example

• Population: People living with HIV. Outcome: Becoming virally suppressed within 1 year of treatment start. Exposure: Drug A versus Drug B

​

​

​

​

​

​

​

58

Measures of disease effect�Risk ratio example

• Population: People living with HIV. Outcome: Becoming virally suppressed within 1 year of treatment start. Exposure: Drug A versus Drug B

​

​

​

​

​

​

​

59

Measures of disease effect�Risk ratio example

• Population: People living with HIV. Outcome: Becoming virally suppressed within 1 year of treatment start. Exposure: Drug A versus Drug B

​

​

​

​

​

​

​

60

Measures of disease effect�Risk ratio example

• Population: People living with HIV. Outcome: Becoming virally suppressed within 1 year of treatment start. Exposure: Drug A versus Drug B

​

​

​

​

​

• Probability of becoming virally suppressed is 1.16�(95% CI: 1.00, 1.34) times greater when using Drug A�compared to B (i.e. 16% higher)

​

​

61

Outline

• Motivation: Why are we talking about “studies”?
• A classical epidemiological study
• Study classifications
• Why study design matters
• Randomised controlled trials
• Measures of disease occurrence (incidence and prevalence) and effect (example)
• Close

​

​

​

000

62

• There are different ways to design a study
• There are different ways to measure something of interest
• Study design matters for accuracy and precision

​

​

What I hope stays with you…

63

• There are different ways to design a study
• There are different ways to measure something of interest
• Study design matters for accuracy and precision

​

• When you use data/estimates in your model (or collect data) be sure to engage with (all) details of the study
• Modelling studies go from processes at a small scale (relating to individuals) to a large scale (populations, across time and place); you need to understand your ‘small scale’ data well to interpret your ‘large scale’ results

​

What I hope stays with you…

This presentation is made available through a Creative Commons Attribution-Noncommercial license. Details of the license and permitted uses are available at� http://creativecommons.org/licenses/by-nc/3.0/

Overview: Study Design and Analysis�Attribution: R. Kassanjee�Clinic on the Meaningful Modeling of Epidemiological Data

© 2014-2024 International Clinics on Infectious Disease Dynamics and Data

000