Introduction to the Scientific Basis of Medicine 2012-13

Course coordinator: Professor Kay-Tee Khaw kk101@medschl.cam.ac.uk

I

II

III

IV

Introduction to Epidemiology:

Inference and causation, inference (KT Khaw)

Topics:

Definition of epidemiology
The epidemiologic process
Assessment of associations
Concepts of confounding and bias
Criteria for causality
Translation to clinical practice

Epidemiology 2:

Methods (KT Khaw)

Topics:

Epidemiologic study designs:

Case control, longitudinal and intervention studies
Strengths and limitations of different designs

Epidemiology 3:

Risk (KT Khaw)

Topics:

Definitions
Measures of rates and risk and their use:
Prevalence
Incidence
Relative risk
Excess or attributable risk
Population excess or attributable risk

Epidemiology 4:

Evidence based medicine (M Holmes)

Topics:

Understanding of scientific method

Appreciation of the need for statistics and quantification of
uncertainty
Recognition of potential sources of bias; qualitative indicators of
uncertainty
Understanding the need for skepticism and objectivity

Epidemiology 5:

Measurement and Screening (KT Khaw)

Topics:

Principles of screening
Sensitivity, specificity and predictive value of a diagnostic test
Critical reading of the scientific literature

Introduction to Epidemiology:
Topics:

association and causation (KT Khaw)

Definition of epidemiology
The epidemiologic process
Assessment of associations
Concepts of confounding and bias
Criteria for causality
Translation to clinical practice

Definition:
Epidemiology is the quantitative study of the distribution, determinants and control of diseases in
populations.
Epidemiologic process:
1.
2.

3.
4

Counting:

Cases; populations; definitions of diseases

and exposures; measurement; rates
Comparisons: Measures of risk
Methods:
descriptive, analytic and experimental study
designs
Inference:
Association and causality
Generalisability
Action:
Clinical/health policy
Further research

Association and causation

Most study designs aim to examine associations between a given exposure (e.g. genetic profile,
medical or surgical intervention, behaviour such as diet or smoking, environment such as air
pollution, or service organisation) and a given outcome (e.g. disease occurrence, death,
recovery, functional health). When any associations are observed, we need to exclude
explanations such as chance, confounding or bias before we can reasonably assume causality.
A confounding factor is:
1.
A risk factor or predictive of outcome
2.
Associated with (but not consequence of) study exposure
3.
Extraneous to the association under investigation
4.
Unequally distributed between exposure groups
Exposure ----------- ? -----------> Outcome
\
/
\_
_/
Confounding factor
e.g we might find an association between increased coffee drinking and lung cancer. Cigarette
smoking might be a confounder of this association as cigarette smokers are more likely to have
lung cancer and heavy coffee drinkers are more likely to be cigarette smokers than non-coffee
drinkers.
Bias is systematic deviation from the truth e.g.
Persons with a disease may be more likely to recall past exposures (e.g. family history, exposure
to radiation) than persons without a disease even if there is no true difference in such exposures.
Good epidemiologic study design attempts to minimise confounding or bias or to consider these in
the analysis and interpretation of results.
When evaluating whether an association between exposure and outcome is believed to be causal,
criteria used to assess causality include:

1.
2.
3.
4.
5.
6.

Temporal relationship
Biological plausibility
Strength of association
Consistency of association
Dose response relationship
Reversibility

Suggested texts:
Basic text:
th
Epidemiology for the Uninitiated. Coggan D, Barker DJP, Rose G. 5 Edition BMJ 2003.
Other reading:
Basic Epidemiology.
Beaglehole R, Bonita R, Kjellstrom T.
Organisation: Geneva, 2005
How to read a paper. Greenhalgh T. BMA Books 2000

2nd Edition World Health

II

Epidemiology 2:

Methods (KT Khaw)

TYPES OF EPIDEMIOLOGIC STUDY

We usually design studies to enable us to make associations between specified exposures (e.g
disease risk factors, or treatments) and outcomes (e.g. health-related events such as heart attack
or survival). The major purpose of good study design is to enable us to make sensible inferences
about the nature of the associations; in particular, to disentangle how far any associations
demonstrated may be attributed to chance, to bias, confounding factors, or to causal
relationships.
OBSERVATIONAL STUDIES
These are studies in which certain features (past, present or future) are observed in groups of
individuals without any new intervention being introduced other than the gathering of information.
Observational studies can be used to investigate the diagnosis, causes and natural history of
disease, to assess the accuracy of diagnostic methods and to evaluate the process of care.
1.

Cross sectional studies

A defined population or sample of that population is surveyed and their disease and
exposure status determined at one point in time

2.

Case control studies

A group of people with a particular disease (cases) are compared with a group of people
without the disease (controls) to determine whether the cases have been exposed more
or less often to a specific factor than the controls. The case control study is usually
retrospective, that is, it looks back into the past.

3.

Cohort/longitudinal/prospective studies
Here, a group of people are identified and grouped on the basis of their exposure to a
specific factor and then are followed up over time to determine whether the subsequent
incidence of a particular disease is different in the different exposure groups.

Confounding occurs when two variables are associated but part or all the association is due to
independent associations with a third, confounding, variable - for example, falls may be
associated with diuretic drug use, suggesting a causal role for diuretics. Heart failure, however,
confounds this relationship as it is associated with use of diuretics and also is an independent
cause of falls. Good study design tries to minimise the effects of confounding but interpretation of
findings from observational studies is challenging since the role of confounding factors is difficult
to exclude completely. Observational studies may often be the only feasible means to study
associations.
INTERVENTION STUDIES
These are experimental studies in which the effects of interventions on pre-specified endpoints
are evaluated. Clinical trials usually examine efficacy and safety of interventions in persons who
either have a specific disease or who are at risk of developing a specific disease. Randomised
trials minimise confounding but are often difficult to conduct.
4.

Randomised controlled trials

Persons are randomly allocated to receive either the intervention or to be in a control
group (ie. to receive no treatment, or placebo, or an existing different intervention), and
then all groups are followed up to examine to determine differences in outcome. The
purpose of randomisation is to ensure as far as possible that groups are similar in all
respects except for the intervention received. The purpose of a control group is to enable
comparison of intervention with what might otherwise have occurred. Blinding is when it
is not known which intervention persons are receiving; double blinding refers to the case
when both the persons receiving the intervention and the investigator do not know which
intervention persons are receiving. The purpose of blinding is to ensure that biases in
assessment of the effects of intervention are minimised.

CROSS-SECTIONAL, CASE CONTROL AND PROSPECTIVE STUDIES

>------------------------------> TIME >-------------------------->
PAST
PRESENT
FUTURE
_________________________________________________________
CROSS-SECTIONAL
Survey a population,
identify cases
and measure exposures

Look for past <----exposure to

risk factor

CASE CONTROL STUDY

Select cases (disease)
and controls

PROSPECTIVE STUDY
Select cohort; ----> Follow and see
classify as to
who develops
exposure to risk
disease
factor
__________________________________________________________________
CASE CONTROL AND PROSPECTIVE STUDIES
Advantages
Disadvantages
___________________________________________________________________
Case control Relatively inexpensive
Selection problems for cases
study
Suitable for rare diseases
Selection problems for control
groups and matching variables
Smaller number of subjects

Biased recall

Relatively quick results

Sometimes unable to determine

whether independent variable
preceded dependent variable
Yields only an estimate of relative
risk (odds ratio)

Potential for selection and retrospective biases are large, but may often be the only feasible way
to address a question e.g. oral contraceptive pill and venous thrombosis
__________________________________________________________________
Prospective Lack of bias in exposure
Possible bias in ascertainment of
study
disease
Yields incidence rates as
well as true relative risk

Large numbers of subjects required

Long follow up period

Can yield associations with

additional diseases

Problem of attrition
Changes over time in criteria and
methods

Expensive, long term, used for "most important" questions

_________________________________________________________________

TRIALS/INTEVENTION STUDIES: ELEMENTS OF TRIAL DESIGN

1.
Randomization
Try to ensure groups similar in all respects except for intervention, that is, that possible
confounding factors including those we do not know about, are equally distributed in the different
intervention groups. To maintain advantages of randomization, the analyses of outcome have to
be by intention to treat.
2.
Controls
Compare what would have happened without intervention
3.
Blinding
Ensure any differences in outcome not due to bias in assessment
DESIGN OF CLINICAL TRIAL
REFERENCE POPULATION
|
External validity
Generalisability
STUDY SAMPLE
|
Internal validity
RANDOMIZE

INTERVENTION

IMPROVED NOT IMPROVED

NO INTERVENTION

IMPROVED NOT IMPROVED

STRENGTHS AND LIMITATIONS OF TRIALS

_______________________________________________________________
Advantages
Disadvantages
_______________________________________________________________
Minimise selection bias
Feasibility
- duration
Minimise confounding
- cost
Quantify benefits and risks
- ethics
- compliance
Efficacy and effectiveness
Generalisability
________________________________________________________________
INTERPRETATION OF TRIAL FINDINGS
Analyses by intention to treat crucial or scientific strengths of randomization are lost.
1.

Quality of study. How well was the trial conducted and analysed? Issues of
randomization, blinding, compliance, analyses by intention to treat.

2.

Relative risk comparison (e.g. percentage change in intervention versus control group and
statistical significance). Addresses the aetiologic question: Did intervention have
significant effect?

3.

Absolute risk comparison (e.g. magnitude of difference between intervention versus

control group - often expressed as numbers needed to treat). Addresses the policy
question: What is the absolute effect of the intervention?

3.

Generalisability - (e.g. how similar/different was the study population from the population to
whom you wish to apply the results?)

III

Epidemiology 3:

Measures of rates and risk (KT Khaw)

RATE
It is a general measure of frequency. It is the number of people with a disorder or the number of
events (e.g. deaths) in relation to the population at risk.
Numerator
=
Denominator

Number of cases
Related population

PREVALENCE
Proportion of defined group having a condition at any one time
=

number with condition at one point in time

population

Prevalence of disease is a function of the frequency with which new cases occur (i.e. incidence)
and the average duration of disease before termination by recovery or death. That is, prevalence
approximates incidence x duration.
INCIDENCE
Rate of occurrence of NEW cases of disease over a time period
=

new cases arising

population

over time period

Advantage is that it is not dependent on disease duration.

Comparison of incidence and prevalence
Rate
Incidence
Prevalence
___________________________________________________________________
Numerator
New cases occurring during
All cases counted on a single
the follow up period in a
survey or examination of a
group initially free of
group
the disease
Denominator
All susceptible subjects
All individuals examined
present at the beginning
including cases and non-cases
of the follow-up period
(population at risk)
Time
Duration of the follow-up
Single point in time
period
How measured
Cohort study
Prevalence or Cross-sectional
study
___________________________________________________________________

INCIDENCE

PREVALENCE

Recovery/death

RISK
It is the notion of the probability of some event. In a more restricted sense, it is the likelihood that
people who are without a disease, but exposed to a certain factors, will acquire the disease.
For example:
50 exposed -----> 5 cases
10 %
1000 persons {
in 1 year
950 non-exposed -----> 19 cases
2%
RELATIVE RISK
It is
incidence in exposed
=
10% = 5
incidence in non-exposed
2%
Relative risk indicates how many times more likely exposed persons are to become diseased,
relative to non-exposed and is more generalisable.
Thus, a relative risk greater than one associated with a given exposure suggests increased risk; a
relative risk less than one associated with a given exposure suggests reduced risk.
EXCESS RISK (Absolute risk difference)
It is incidence in exposed - incidence in non exposed
=
10% - 2%
= 8% or 8 cases per 100 per year
Excess risk indicates the excess incidence of disease associated with exposure, that is, for every
100 persons exposed, 8 extra cases occur.
Another way of looking at this would be that for every extra case, 100/8 or about 12-13 persons
have to be exposed. This approach is generally used in the context of clinical practice to derive
Numbers needed to treat (NNT), to examine the effects of treatments.
Numbers needed to treat (NNT) addresses the issue of how many people would have to be
exposed or treated to cause or prevent one case.
POPULATION EXCESS RISK
It is individual excess risk x prevalence of exposure
= 8/100 x 50/1000 = 4/1000 per year
Population excess risk indicates the incidence of disease in a population associated with the
occurrence of a risk factor i.e. the best that can happen if the factor is controlled. It depends on
the prevalence of exposure of the risk factor, which is more variable in different populations, and
is hence a less generalisable measure.
MEASURES OF RISK
RELATIVE
EXCESS
POPULATION
RISK
RISK
EXCESS RISK
____________________________________________________________
Estimates
Force of effect
Size of effect
Size of effect
for individual
for population
Relative rates

Absolute rates

Absolute rates
Prevalence of
risk factor

Generalisable

Yes

No

No (even less)

Use

Research

Practice
Practice
(Individual)
(Population)
____________________________________________________________

IV

Epidemiology 4:

Evidence based medicine (M Holmes)

Lecture handouts following the lecture

Epidemiology 5:

Screening (KT Khaw)

PRINCIPLES OF SCREENING
Wilson and Jungner, WHO 1968
Aim of Screening
Detect condition which would be otherwise unknown in apparently
asymptomatic individuals
Intervene
Improve outcome
Questions
-

Can we identify important potential problems?

What proportion are potentially remediable?
What resources are available to do so?

The condition
Should be an important health problem
Early treatment should be of more benefit than late treatment
Should be adequate facilities /resources for management of conditions
detected
The test
-

Should be acceptable: feasible, safe, cheap

Should be accurate: sensitive, specific, high predictive value
Chance of physical or psychological harm to those screened should be less
than the chance of benefit

Evaluation of screening: individual

Benefit:

for every individual in whom condition detected and improved

N.B. Survival time is often not good indicator as disease may be diagnosed
earlier (lead time bias)

Cost:

how many individuals had to undergo screening test?

how many individuals in whom condition detected but not improved?

Evaluation of screening: community

Benefit:

How many individuals in whom outcome improved?

Cost:

Resources used: time, personnel, materials: opportunity costs

Ideally, randomised controlled trials of screening vs no screening on outcomes but these are
hard to do.

SENSITIVITY, SPECIFICITY, AND PREDICTIVE VALUE OF A TEST

|
|
|
|
Prevalence% |
Healthy
FP
TN
|_____________________________________
Diseased
FN

Negative
Test criteria

TP

Positive

Sensitivity is the proportion of true positives correctly identified by the test = TP/TP+FN
Specificity is the proportion of true negatives correctly identified by the test = TN/TN+FP
Predictive value of a positive test is the likelihood that a person with a positive test has
the disease = TP/TP+FP
Predictive value of a negative test is the likelihood that a person with a negative test does
not have the disease = TN/TN+FN

Test

DISEASE
yes
no
______________________________________
yes
TP
FP
TP+FP test positive
______________________________________
no
FN
TN
FN+TN test negative
______________________________________
TP+FN FP+TN
true positive true negative

Sensitivity and specificity are independent of the prevalence.

Predictive value depends on the prevalence of the disease.
EXAMPLE: ECG abnormality to predict angiographic disease.
Assuming a test has 90% sensitivity and 90% specificity:
For a disease of 50% prevalence, predictive value would be 100 x 90/100 = 90%, that is out
of every 100 persons with a positive test, 10 would NOT have the disease.
Disease
yes
no
FP
TN
Prevalence%
test
yes
900
100 1000
e.g. cardiology
no
100
900 1000
TP
FN
clinic
1000 1000 2000
For a disease of 1% prevalence, predictive value would be 100 x 18/216 = 8%, that is, out of
every 100 persons with a positive test, 92 would NOT have the disease.
Disease
yes
no
FP
TN
Prevalence%
test
yes
18
198
216
e.g general no
2
1782 1784
population
20
1980 2000
FN

TP

Reading the literature

The exact criteria for critical reading of the literature depend on the type of study. The
following, derived from various teaching sources, may provide a helpful guide. In most
studies, the aim is usually to make an association between a given exposure and a given
outcome, then to interprete the nature of any association found.
1.

Aim of the study. Is there a clear statement of the aim?

2.

What was the study design? cross sectional, case control, prospective, or trial?
Is the description of methods full and clear?
Do you know exactly what was done?

3.

Who were the study population?

What selective forces were there?
Is there a rationale for the choice of this particular population?

What was the main exposure under investigation?

How was it ascertained, defined and measured in this study?
Method of measurement - variability? validity? appropriate?

5.

What was the main outcome under investigation?

How was it ascertained, defined and measured in this study?
Method of measurement - variability? validity? appropriate?
e.g. case definition; representativeness of all cases?

6.

What possible confounding factors or sources of bias might there be?

Were these measured/documented?

7.

Data presentation and analysis how clear and complete?

descriptive characteristics of study population
complete presentation of numbers
measures of association and risk
statistical significance
control for confounders and bias

8.

Interpretation of results: could the association observed be due to:

chance
artefact
bias - measurement
- selection
confounding
cause
How well have the authors considered the various explanations?

9.

Conclusion:
How do the findings relate to the aims?
How generalisable do you think the findings are?
Are the interpretation of results, and hence, the conclusions justified?

10.

Summary/ Abstract: does this clearly and fairly summarise what is shown in the
paper?

11.

Was the design appropriate to the aim of the study? How would you do this study
better?
What action or recommendations do you think are justified based on the study?

12.