Epidemiology summary weeks 1 to 6

Definition: Study of distribution &

determinants of health-related events
in specific populations, & applying it to
prevent & control health problems
Risk factor: Factors that are positively
associated with risk of disease
development but are not sufficient to
cause the disease (e.g. smoking is a
risk factor for CHD)
Prevalence: Frequency of existing
cases in a predefined population at a
given point of time (Period prevalence
= over a longer time)
Incidence: Number of new cases of
disease during a specified time period/
number of people exposed
Cumulative incidence: Number of
people who get a disease during a
specified period/number of people free
of disease in the population at risk at
beginning of the period.
Population attributable risk: % of
disease in whole population that can be
attributed to a particular exposure
Risk difference: Difference in rates of
occurrence between exposed &
unexposed groups in population
Confounding: When another exposure
exists in the study population. Links
both disease & exposure being studied
Mortality: Death rates.
Morbidity: Illness rather than
mortality
Health determinants: Underlying
factors (e.g. env, social, econ) that are
responsible for health & disease

Sufficient
A cause is
sufficient when it
inevitably
produces or
initiates an
outcome
Smoking is not
sufficient to cause
lung cancer
(Some smoke
>50 years and
dont get it)

Necessary
A cause is
necessary if an
outcome cannot
develop in its
absence.
You must have
the
mycobacterium
tuberculosis in
order to get
tuberculosis

Memorise this table!

Disease (lung
Risk
Yes
factor
Yes
A
Or
No
C
exposur
Tot
A+C
e
al

cancer)
No
Total
B
A+B
D
C+D
B+D

Odds ratio: Association of exposure to

disease (Used in case control & cross
sectional studies for when disease is
rare)
Formula: (Proves hypothesis)/(disproves
it)

Validity: See if results correspond

to truth
Internal
validity (e.g.
Exposure withdisease
Exposure
without
disease
=

blood Hb must be accurately

Noexposure with disease
No exposure
, no disease
distinguished bet. Anaemia vs.
non) and External Validity (e.g.
A
B
AD

=
results must apply to people that
C
D BC
are not in study)

(
( )( )

)(

Relative risk: For cohort & RCTs

Re = A/(A+B), Ru = C/(C+D)
RR = Re/Ru = (A X (C+D))/(C X (A+B))
Risk difference/attributable risk =
Re-Ru
Attributable fraction: % of cases
attributed to exposure: ((ReRu)/Re)x100%
If RR/OR includes 1: No diff, includes=diff

DALY: Years of lost life + years lost to

disability

Population attributable risk: [A/

(A+C)]X100%

STUDIES (See the triangle)

Ecological: Observation (O) made on
groups (not individuals). E.g. Liver
disease and alcohol consumption
X-sectional: Assess exposure &
outcome simultaneously. No evaluation
of temporality. E.g. cannabis and
depression
Case-control: Backwards in time to look
for exposure. Start with diseased state.
Rare diseases, odds ratio
Prospective Cohort: Forwards in time.
Start with healthy people + exposure
and without exposure and see who gets
the disease.
Retrospective Cohort: Forwards in time.
Exposure is defined at time in past. No
knowledge of possible outcomes.

Standard Error: Distance from sample

statistic to population parameter. As n
increases, SE decreases. As SD
increases, SE increases as well.

Of course, RCT and Meta-Analyses

Establishing causation:
1. Temporal relationship (**)

Cause must precede event

2. Plausibility

Clinical tests
3. Consistency

Several studies showing

similar results
4. Strength of association

Degree of association between

cause and effect
5. Dose-response r/s

Change in level of cause are

associated with changes in
incidence of effect
6. Reversibility

If removal of cause results

in decreased disease risk
Errors
Random error (Biological variation,
sampling error, measurement error
(tools not good enough)), Sample
size, Systematic error, Selection
bias, Measurement bias (nv
measure correctly) (recall bias,
biochemical bias, observer bias)

SE=

SD
n

Confidence interval: Interval around

sample statistic that captures population
parameter =

xx T multiplier SE( x )
(You should get a range of values)
CI between 2 means:

( x 1 x 2 ) T multiplier SE( x 1x 2)

If CI includes zero: No diff, if includes =

diff.
T-test 3 kinds: 1 sample (mean vs.
hypo), 2 sample (2 sep. means) paired

Categorical data: Values that

can be sorted into groups. Sex/eye
colour. Bar charts & pie graphs
used
Numerical data: Values that can
be measured. Placed in order.
(Height, arm span, weights).
Scatter plots/line graphs used.
Regression (3 types)
Linear(numerical data
continuous) Logistic regression
(binary data), survival
analyses/cox hazard model
p-value: Likelihood a value for
mean is so rare its difficult to
believe. Errors: Type 1: No
difference but test says there is
Type 2: Null hypothesis is wrong
but test says it is correct.
Null hypo (Ho): Default
statement. No relationship
between 2 ideas
Alt hypo (H1): Hoped/expected
that relationship is present.

(bef & aft)

Chi-squared test:

(OE)2
E