EBM semester 1

If X……. THEN Y!?!?

Sasha Hall
 • Do at least 1 VIA, discuss answers as a group,
they repeat questions/ topics and have niche
buzzwords that might come up
• Link to equella:
https://equella.monash.edu/more/hierarchy.do?topi
c=ce942a21-0771-420f-93bc-9b1ac9456902&page=1

• Do VIAs as a group and discuss answers

Tips • If the question sucked for enough people,
contest it ☺

• If x then y is a lie

• Also d- dimers kept coming up and we were never taught

them ☹ (Haem topic)
• Very sensitive blood test, so if result is negative rule out a DVT/ PE
• If test is positive, it is meaningless- could be positive due to many
things
 • Tute 1: types of variables, types of bias, time place person, case
definitions
• Tute 2: prevalence, incidence, mortality and morbidity calculations,
“abnormality”, sufficient vs necessary causes, Bradford Hill criteria
• Tute 3: Cohort studies and RR, RRR ARR absolute and attributable
Overview of risk, Case control studies and OR, HR Kaplan Meier survival curves
• Tute 4: RCTs, NNT NNH, Intention to treat analysis, hierarchy of
Sem 1 evidence

content • Tute 5: outbreak investigation, attack rate, different types of

outbreaks etc
• Tute 6: third variable effects: difference between bias, confounding
and effect modification/ interaction
• Tute 7: Hypothesis testing, p values, types of analyses
Feedback from last • Third variable effects, p values, all the calculations
years mid year exam-
Year A content though
 Types of data and
variables, choosing
statistical tests
 Types of data and variables
Data
• Numerical
• Discrete: whole numerical values, shoe sizes, age in years*
• Continuous: values on a scale, can have multiple decimals, height, age*
• Interval: numerical values within equally spaced ranges/ brackets – brackets of age*
• Categorical
• Ordinal: each category is ordered, 1st 2nd 3rd; BMI ranges
• Nominal: no order, just categories, favourite colours, blood type
• Binary/ dichotomous, only 2 categories/ responses, yes/ no, marital status, dead/ alive

Variables
• Dependent: variable you expect to change (outcome- disease or no disease)
• Independent: variable you alter (exposure- smoking or no smoking)

Choosing statistical tests for hypothesis testing with a scary table🡪 just need to decide
which variable is which and the type of data for each variable
MCQ 53 (1 mark)
In Table 1 what type of
variable most accurately
describes age?
A. Categorical.
B. Dichotomous.
C. Discrete.
D. Interval
MCQ 53 (1 mark)
In Table 1 what type of
variable most accurately
describes age?
A. Categorical.
B. Dichotomous.
C. Discrete.
D. Interval
Exposure/ independent:
Age- continuous

Outcome/ dependent:
UTI Y/ N – binary

M- logistic regression
Exposure/ independent:
Blood loss in ml-
continuous

Outcome/ dependent:
AKI Y/ N – binary

D- logistic regression
Exposure/ independent:
category of AKI aetiology
– categorical

Outcome/ dependent:
mortality at 30 days as a
number of participants
or a percentage –
continuous

N- analysis of variance
Bias, error, confounding,
effect modification
 Types of bias and error (results differ from true values)
Systematic error: same amount of error each time Random error: error due to chance alone- will be a
(precise) due to a flaw in the process/ system/ tools different amount of error each time (not precise)
CANNOT be fixed by taking repeated measurements- will FIX it by taking repeated measurements ie. Multiple
always give you the same amount of error BP readings
Small sample size🡪 large random error
• Selection bias: when there is a systematic difference between the sample and the population or a systematic
difference between control and intervention group, due to systematic error in inclusion/ exclusion criteria and study
selection processes and allocation to one group or the other
• Sample and population: exclusion, inclusion criteria, control and intervention group: Loss to follow up

• Measurement bias: results of test are not the true value due to systematic error in the tools used, any bias to do with
measurement of outcomes is a type of measurement bias
• Many types of measurement bias: detection bias, recall bias etc

• Misclassification: when you incorrectly assign a subject to the control or case group
• Differential misclassification: the error (misclassification) occurs at a different level in cases vs controls ie. Person who measured control group
overestimated the exposure
• Biases results towards statistically significant results that are an over or under estimation of the results

• Non-differential misclassification: the error (misclassification) occurs to the same degree in both cases and control groups
• Biases results towards results being not statistically significant- accept the null hypothesis (no difference between the groups)

• FIX IT by ensuring same objective standard for classifying and measuring exposures
Selection bias
• Things that cause selection bias
Systematic difference between population of interest and
• Controlling for selection bias
sample o Case control studies:
• Inclusion/ exclusion criteria excluding certain diseases or o Have a strong objective case
variables definition
• Bias from method of recruitment/ advertising the study- ie. o Appropriate/ matched control
Advertising in magazines only excludes people that are selection
illiterate/ don’t interact with magazines, advertising via text o Having a high participation rate
excludes those without a phone
• Healthy worker effect: study recruitment often occurs in in both groups
working people/ workplaces, which fundamentally excludes o Cohort studies:
recruitment in those unable to work due to disability o Maximise retention of the cohort
• Volunteers: certain types of people volunteer for studies- (both exposed and unexposed
normally more health conscious etc, inherently biased people)
population o Where possible, follow up with
those LTFU
Systematic difference between control and intervention
groups
• Loss to follow up (LTFU): means you are only analysing the
people that stayed vs all people recruited
• FIX with intention to treat analysis cf. only analysing those that
completed the whole study to the end
Types of measurement bias that you might see
• Recall/ responder bias: misclassification of data from the truth due to interviewers/ researchers,
instruments or methods or the subjects themselves ie. Bias from self-reported data and recalling past
information inaccurately. Can over or under estimate true effects
• FIX it with objective measures of outcome assessment like biological data instead of self-
assessments or self-reports where possible
• Reporting bias: bias from a participant is being likely to only include or disclose positive information/ a
study skews how information or results are presented
• Detection/ surveillance/ ascertainment bias: bias from outcome assessment process ie. If disease status
is known when recording exposure retrospectively may be more likely to inflate amount of exposure
• FIX it with blinding of outcome assessors
• Publication bias: when studies with negative or non statistically significant results are less likely to be
published than studies with positive findings or that confirm expectations (selection bias when
publishing studies)
 Bias: systematic deviations of the results from the truth, due to how
Third variable effects data was collected or measured

Confounding: Effect modifier:

• third variable that is not on causal pathway (not caused
by E or O), but has separate association with the
exposure and the outcome
• Confuses the true association- BAD, want to get rid of it
• STRATIFY IT OUT🡪 both stratified groups will be the
same RR (move in same direction) and move in
opposite direction to crude RR
• E: alcohol O: lung cancer confounder: smoking
• third variable that alters the effect of the exposure on the
disease; the magnitude of E affecting O will differ
depending on the third variable (old age vs young age)
• A finding of note! GOOD, want to report it
• STRATIFY IT OUT🡪 stratified groups RR with move in
opposite directions from each other, and are different from
crude RR
• E: surgery O: death effect modifier: age
How to control for confounding variables
• In study design stage (before data collection) and in data analysis
stage (after data collection)
This is the question stem
Measurement bias
- Includes recall bias
Measurement bias again
- Detection bias is a form of measurement bias
Random error is inversely proportional to sample size.

With decreasing sample size, random error increases.

Selection bias
Inherent bias in the cases vs controls due to the
setting from which they were selected (where they
lived)
Incidence prevalence,
rate calculations
 Incidence v prevalence
Incidence
• How many NEW cases or events
occurring over a period of time
• New cases over period of time
• Rate- has a temporal component
How do prevalence and incidence relate?
• Prevalence of lung cancer= incidence rate of new
cases x average duration of having lung cancer
Prevalence
• How many EXISTING cases or
events at that time point
• Snapshot in time of disease
present in the population
• Number/ value
Incidence calculations
• Numerator always number of new cases
• Incidence proportion= cumulative incidence
• measures the proportion of a specific group of people who develop the disease during a specified
period
• Intuitive, just a normal fraction. No temporal component, just to do with a time period
• Number of new cases in a specified period/ number people at risk at the start of that period
• Incidence rate= incidence density
• measures how quickly people are developing a disease by incorporating time into it
• Not intuitive ☹, denominator needs person time (temporal component)
• Number of new cases in a specified period/ person years that the population at risk spent being at
risk of getting the disease
• Person time/ person years= how many cumulative years the “population at risk” were “exposed/ at
risk”
• number of people at risk X average no years those people were at risk for
Lifestyle factors are considered important components for the development of cardiovascular disease. A
(hypothetical) study was conducted in hypertensive individuals to investigate if there was a reduction in the
incidence of cardiovascular disease. The study followed twelve people with hypertension over a period of two years.
It was noted that one person left the study at the beginning of the second year of the study. Three people
developed cardiovascular disease by the end of the first year and two people developed cardiovascular disease by
the end of the second year.

What is the incidence rate for the development of cardiovascular disease in this study population?

A. 41.7 per 100 per year.

B. 5/12 person years.

C. 5/20 person years.

D. 5/24 person years.

Lifestyle factors are considered important components for the development of cardiovascular disease. A
(hypothetical) study was conducted in hypertensive individuals to investigate if there was a reduction in the
incidence of cardiovascular disease. The study followed twelve people with hypertension over a period of two years.
It was noted that one person left the study at the beginning of the second year of the study. Three people
developed cardiovascular disease by the end of the first year and two people developed cardiovascular disease by
the end of the second year.

What is the incidence rate for the development of cardiovascular disease in this study population?

A. 41.7 per 100 per year. New cases: 5

B. 5/12 person years. Person time: 12 people for 2 years BUT

4 x 1 year: 3 got CVD, 1 dropped out
C. 5/20 person years. 8 x 2 years
4+ 16= 20
D. 5/24 person years.
5/ 20 person years
Prevalence calculations
• Point prevalence: existing disease at snapshot in time
• Number of existing cases at specific time point/ population at same specific
time point
• Intuitive
• Period prevalence: prevalence during a specified time period (longer
snapshot)
• Number of cases / mid-interval population over a specified time period
• Stupid, has it over a time period using the population at the middle of the
period
Point prevalence
- Just normal prevalence
 Summary
Measure
Attack Rate/ incidence (measure of
risk)
Case-fatality
Cause-specific death rate
Crude death rate
Incidence proportion
(in this population, how many people
are getting the disease?)
Incidence rate
(speed of disease occurrence- in this
amount of time, how many people are
getting the disease?)
Infant mortality rate
Period prevalence
Point prevalence
Relative survival rate
Numerator
Number of new cases of disease during specified
time interval
Number of cause-specific deaths among the
incidence cases
Number of cause-specific deaths among the
incident cases
Total number of deaths from all causes during a
given time
Number of new cases of disease during specified
time interval
Number of new cases of disease during specified
time interval
Number of deaths among children <1 year age
during a given time
Number of current cases at specified period of time
Number of current cases at specified point in time
Observed survival in people with the disease
Denominator
Population at start of specified time interval
Number of incidence cases
Mid-interval population
Mid-interval population
Population at start of specified time interval
Summed person-years of observation
Number of live births during the same interval
Mid-interval population
Population at the same specified point in time
Expected survival if disease were absent
 Burden of disease terms
Average number of years that an individual of a given age is expected to live Person born in 1980 in Melbourne is expected to live until age
• based on age and projected rates of death from current mortality data 90, as of 2003 data
Life expectancy • Hypothetical

Sum of the differences between a predetermined minimally accepted age • Average LE is 90

QALY:
(often life expectancy) and the age at death for everyone who died from that quality adjusted
• Average life years
age of death from CVD is 65
Potential years cause earlier than the minimally accepted age. • PYLL= 90- 65 = 25
of life lost • Uses population data HALY: health adjusted
• On average, life with
people yearsCVD lost 25 years of life due to
CVD
(PYLL)

Weights each year of life by the perceived quality of that life • Cystic fibrosis patient is 20 years old
Quality- − 0 = dead, 0.1- terrible health • Subjectively weight each year at 0.7
adjusted life − 1 = year of perfect health • QALY= 20 x0.7= 14 years
• 20 years of life with CF equivalent to 14 years in perfect
years (QALY) health
Combination of QALY and life expectancy • LE of 90
Health- Represents the equivalent number of years an individual can expect to live • T2DM at 65 (0.7), gets dementia at 85 (0.4)
adjusted life in full health • HALE= 65 x1, 20 x 0.7, 5 x 0.4= 65 + 14 + 2= 81
expectancy − 0 = dead, 0.1- terrible health • HALE is 81 due to disease/ disability states
(HALE) − 1 = year of perfect health
Years of life lost due to premature death or disability. Weights each year of • 30 year old has LE of 90, becomes quadriplegic at 30 (0.8),
life by disability lives for another 40 years
Disability- Scale is in opposite direction, 0 is full health • DALY= PYLL + YLL due to disability
adjusted life − 0 = full year of health • PYLL= 90- 70= 20
years (DALY) − 1 = year lost to death • YLL due to disability= 40 x 0.8= 32
• DALY= 20 + 32= 52 years of life lost
Having another stroke- new events- incidence

Rate of new cases🡪 cumulative incidence over 5 years

Doesn’t include person years in there🡪 not incidence
density
Risk and odds
Risk
• Risk of an outcome occurring in the population
• E.g. risk of food poisoning at a party
• Absolute risk= number of events/ number of Menopaus No
people at risk e menopause
• = attack rate= event rate= incidence proportion Fracture 54 a
(pre)
90 b 144 a+b
• Just a fraction No 46 c 210 d 256 c+d
fracture
• Measure of probability or chance, comes out with a 100 a+c 300 b+d
proportion
• Absolute risk in beta blocker group= a/a+c= 36/
100= 0.36 Beta Placebo
blocker
• Absolute risk in placebo group= b/ b+d= 60/100= Death 36 a 60 b 96 a+b
0.6 No 64 c 40 d 104 c+d
death
100 a+c 100 b+d
 Relative risk (same as risk ratio)
• Relative risk= proportion of the baseline risk (unexposed group) that existed in the intervention group
(exposed group)
• Absolute risk of disease in exposed/ absolute risk of disease in unexposed
• Cheat way using the table= (a/ a+c)/ (b/b+d)= a(b+d)/ b(a+c)
• RR is used for cohorts, RCTs, outbreaks because the direction of inquiry is prospective/ looking
forward in time
• RR=1 means same risk of disease in both groups

• RR < 1 means exposure decreases risk of
disease
• E.g. RR= 0.36/ 0.6= 0.6
• “risk of mortality in those with HF taking
beta blockers was 60% of mortality risk for
those not taking beta blockers”
• RRR: “beta blockers decrease risk of
mortality by 40%”
• RR> 1 means exposure increases risk of
disease
• E.g. RR= 0.54/ 0.3= 1.8
• “Risk of fractures in menopause is 180% of
the risk of fractures in pre-menopausal
women”
• RRR: “menopause increases risk of
fractures by 80%”
 Relative risk reduction (and relative risk increase)

ARR, RRR ▪ Uses RELATIVE RISK

Absolute risk reduction= attributable risk= risk difference ▪ how much the intervention increases or decreases the baseline risk
Uses ABSOLUTE RISK (in terms of proportions/ percentages)
• the absolute difference in rates in the control group compared to the
treated (experimental) group ▪ Reduction- where RR is less than 1🡪 1-RR
• Quantifies the amount of risk that is due to the exposure only by ▪ Increase- where RR is greater than 1🡪 RR-1
subtracting the baseline risk (risk in control group)
• Ie. the risk that is above the baseline risk, thus telling you how ▪ 2 formulas, same answer
many more events occurred specifically due to the exposure or
how many events were prevented by the exposure
▪ RRR= 100-(RR as a percentage)= (1-(RR as a decimal)) x100
• ARR= absolute risk in exposed- absolute risk in unexposed
• = incidence exposed- incidence unexposed ▪ RRR= CER-EER/CER
• = event rate exposed- event rate unexposed
• = EER- CER ∙ CER= control event rate
• E.g. beta blockers and mortality ∙ EER= exposed event rate
• EER= 0.36 CER= 0.6
• = ARR= 0.36- 0.6= -0.24 RR Reduction- where RR is less RR Increase- where RR is
than 1 greater than 1
• There was a 24% reduction in risk due to beta blockers
Formula 🡪 1-RR Formula 🡪 RR-1

• Atrributable risk is used for outbreak investigations ie. Where the

EER> CER (same as absolute risk increase) Beta blockers for HF and Fractures and menopause: RR
• E.g. fractures and menopause mortality: RR is 0.6 is 1.8
RRR= (1-0.6) x 100= 40% RRI= (1.8-1) x100= 80%
• EER= 0.54 CER= 0.3
reduction increase
• Attributable risk= 0.54- 0.3= 0.24 “beta blockers reduce risk of “menopause increases risk of
• “24% of the risk of fractures occurred because of being post- mortality by 40%” fractures by 80%”
menopausal/ positive menopausal status”
 ARR vs RRR

• https://www.youtube.com/watch?v=QPXXTE8N4PY
• Use the following options to answer EMQs 35–38
A. 150 / 75
Non–smoker Smoker
B. 150 – 75
C. 60 – 15 Non–diabetic 15 60

D. 60 / 15 Diabetic 75 150
E. 150
F. 75 / 15
G. 150 / 60
Smoking and diabetes are considered risk factors for myocardial infarction. One of the studies Skye found
examined the association between smoking, diabetes and MI. The 15 year incidence of MI per 1000 people is
presented in the table below.

1. Amongst diabetics, what is the relative risk of MI for smokers compared to non–smokers?

2. Amongst non–diabetics, what is the risk attributable to smoking?

3. What is the absolute risk of MI in diabetics who smoke?

4. What is the risk difference among diabetics between those who smoke and don’t smoke?
• Use the following options to answer EMQs 35–38
A. 150 / 75
Non–smoker Smoker
B. 150 – 75
Non–diabetic 15 60
C. 60 – 15
Diabetic 75 150
D. 60 / 15
E. 150
F. 75 / 15
G. 150 / 60
Smoking and diabetes are considered risk factors for myocardial infarction. One of the studies Skye found
examined the association between smoking, diabetes and MI. The 15 year incidence of MI per 1000 people is
presented in the table below. Incidence= absolute risk ……. Not your standard 2 x2 table

1. Amongst diabetics, what is the relative risk of MI for smokers compared to non–smokers?
abR smokers/ abR non smokers= 150/75
2. Amongst non–diabetics, what is the risk attributable to smoking?
Attributable risk= 60-15
3. What is the absolute risk of MI in diabetics who smoke?
150
4. What is the risk difference among diabetics between those who smoke and don’t smoke?
Risk difference= attributable risk= 150-75
 Odds ratio
• OR= Odds of event in exposed group / odds of event in unexposed
group , given the outcome has occurred (disease or no disease)
• E.g. odds of being a smoker in those with lung cancer vs odds of being
a smoker in those without lung cancer Odds=/= risk (probability or
chance)
• Used for case controls and cross sectional studied because the
direction of inquiry is retrospective Odds do not divide between the
• Cannot calculate incidence retrospectively total number
• OR= odds of outcome in exposed/ odds of outcome in unexposed= Odds of rolling 6= 1/5
(a/c) / (b/d)= ad/ bc Pr of rolling 6= 1/6
• OR= 1 means same odds of outcome in both exposure groups= no
association
• OR> 1 means greater odds of disease in exposed group vs control
• OR is 2.5- for every non smoker that develops cancer, 2.5 smokers will
develop cancer
• Cancer is 2.5 times more likely in smokers than no smokers
• OR< 1 means lower odds of disease in exposed group vs control
• OR is 0.7- exercise is associated with a 30% lower odds of CVD
 Study designs
hierarchy of evidence
Hierarchy of evidence
and types of studies
 • Have to be reproducible (ie. Documented search criteria)
• Often include a meta-analysis (statistical output in the form
Systematic reviews of a forest plot)
• Pooled result is the big diamond at the bottom
Study designs- cohort
vs case control
•Cohort
• Pick a bunch of people (some will or won’t have the exposure),
follow them up and see who gets disease
• Direction of inquiry is always forward
• Prospective (exposure status unknown at time of
commencement) or retrospective (exposure status collected
retrospectively, participants followed prospectively for outcome
• RR= risk of disease given that were exposed
• $$$ take ages, less bias
•Case control
• Pick diseased and healthy people, quiz them on their past to see
who was exposed and who wasn’t
• Always retrospective direction of inquiry- have to know disease
status
• Survey of people with and without lung cancer about their
previous smoking history (did you smoke Y/N, for how long?
• Rare disease
• OR= odds of disease given their exposure
• Retrospective so can’t use incidence
• $ quick, more bias
Other study designs
• Systematic reviews + meta-analyses
• Randomised controlled trial
• Cross-sectional: take data from one point in time
• Case report- one patient’s case history
• Case series- the case histories of multiple patients
• Ecological study- measures associations on population
levels
• Ecological fallacy- incorrectly assuming that the
association that exists at population level is also
true for an individual ie. The class average IQ is high
so Bob has a high IQ
• Use the following options to answer EMQs 32–34. Choose the appropriate study design as described in the questions.
A. Randomized controlled trial
B. Meta–analysis
C. Ecological study
D. Case–control study
E. Case report
F. Case series
G. Cross–sectional study
H. Prospective cohort study
I. Retrospective cohort study
1. A group of researchers wish to test for a possible association between levels of C–reactive protein (CRP) and coronary heart
disease. The 388 participants of the study were aged 55 to 60. On the same day that they received a blood test to check CRP
levels, they received an ECG and completed a questionnaire relating to possible risk factors for coronary heart disease.
• Use the following options to answer EMQs 32–34. Choose the appropriate study design as described in the questions.
A. Randomized controlled trial
B. Meta–analysis
C. Ecological study
D. Case–control study
E. Case report
F. Case series
G. Cross–sectional study
H. Prospective cohort study
I. Retrospective cohort study
1. A group of researchers wish to test for a possible association between levels of C–reactive protein (CRP) and coronary heart
disease. The 388 participants of the study were aged 55 to 60. On the same day that they received a blood test to check CRP
levels, they received an ECG and completed a questionnaire relating to possible risk factors for coronary heart disease.

Cross- sectional
- Testing an association with data taken at one time point
• Use the following options to answer EMQs 32–34. Choose the appropriate study design as described in the questions.
A. Randomized controlled trial
B. Meta–analysis
C. Ecological study
D. Case–control study
E. Case report
F. Case series
G. Cross–sectional study
H. Prospective cohort study
I. Retrospective cohort study
1. A group of researchers wish to test for a possible association between diabetes and coronary heart disease. Participants were
grouped based on if they had diabetes or not and followed for 15 years. The end–point was an episode of myocardial
infarction.
• Use the following options to answer EMQs 32–34. Choose the appropriate study design as described in the questions.
A. Randomized controlled trial
B. Meta–analysis
C. Ecological study
D. Case–control study
E. Case report
F. Case series
G. Cross–sectional study
H. Prospective cohort study
I. Retrospective cohort study
1. A group of researchers wish to test for a possible association between diabetes and coronary heart disease. Participants were
grouped based on if they had diabetes or not and followed for 15 years. The end–point was an episode of myocardial
infarction.

Cohort study
Grouped on exposure and followed prospectively
• Use the following options to answer EMQs 32–34. Choose the appropriate study design as described in the questions.
A. Randomized controlled trial
B. Meta–analysis
C. Ecological study
D. Case–control study
E. Case report
F. Case series
G. Cross–sectional study
H. Prospective cohort study
I. Retrospective cohort study
1. A group of researchers describe 10 morbidly obese male patients who presented with symptoms of coronary heart disease at
a cardiovascular specialist clinic. Tests results for dyslipidaemia, hypertension and high cholesterol were recorded as well as
symptoms of cardiovascular disease like heart failure.
• Use the following options to answer EMQs 32–34. Choose the appropriate study design as described in the questions.
A. Randomized controlled trial
B. Meta–analysis
C. Ecological study
D. Case–control study
E. Case report
F. Case series
G. Cross–sectional study
H. Prospective cohort study
I. Retrospective cohort study
1. A group of researchers describe 10 morbidly obese male patients who presented with symptoms of coronary heart disease at
a cardiovascular specialist clinic. Tests results for dyslipidaemia, hypertension and high cholesterol were recorded as well as
symptoms of cardiovascular disease like heart failure.

Case- series
There is no association being tested, it is 10 individual cases as a collection of data
If it was 1 case, it would be a singular case report
• Which of the following is true?
• A) case control studies are more appropriate for rare outcomes
• B) case control studies are more appropriate for rare exposures
• C) Cohort studies are faster to conduct than case controls
• D) Cohort studies are cheaper to conduct than case controls
• Which of the following is true?
• A) case control studies are more appropriate for rare outcomes
• B) case control studies are more appropriate for rare exposures
• C) Cohort studies are faster to conduct than case controls
• D) Cohort studies are cheaper to conduct than case controls
• Which of the following is the best type of publication to inform clinical
practice?
• A) Cochrane systematic review
• B) case study
• C) expert opinion
• D) Meta analysis of RCTs
• E) RCT
• Which of the following is the best type of publication to inform clinical
practice?
• A) Cochrane systematic review
• B) case study
• C) expert opinion
• D) Meta analysis of RCTs
• E) RCT
RCTs, ITT, NNT NNH
RCTs

• Need to be randomised to prevent bias and minimise the effect of

confounders
• Need to be blinded
• Single blind- patients
• Double blind- outcome assessors

• Hawthorne effect: the alteration of behaviour by the subjects of a study due to

their awareness of being observed
NNT, NNH
Number needed to treat🡪 how good is the intervention at
preventing mortality or disease outcomes?
• NNT= the number of patients that need to be given the
intervention in order to prevent one negative outcome (e.g.
death) from occurring, in a specified period.
o NNT = 1/ (control event rate – treatment event rate)
o NNT = 1/ absolute risk reduction
• NNT= 5
• 5 people need to be treated with frusemide to prevent one
person from dying
• Shows treatment benefit, want it to be a low number
(reflects large benefit of the intervention)
Intervention= frusemide for HF
Outcome= mortality
Adverse effect= allergy
Number needed to harm🡪 how safe is the intervention/ how
likely is it that the intervention will cause a harmful adverse
event?
• NNH= the number of patients that need to receive the
intervention before 1 additional patient is harmed by taking
the intervention, during a specified period.
o NNH = 1/ (treatment event rate - control event rate)
o NNH = 1/ absolute risk increase
• NNH= 20
• 20 people need to be treated with frusemide before one
patient experiences an allergic reaction to it
• Shows treatment risk, want it to be a high number (reflects
low adverse events)
ITT analysis vs per protocol
• Intention to treat analysis:
analyse the data from everyone
who was randomised, includes
LTFU patient data before they
became LTFU
• Minimises bias from LTFU
(dropouts)
• Per protocol analysis: only
analyse the data from those who
completed the whole protocol/
study
• Does not include LFTU due to
death, leaving study etc
Outbreaks and
case definitions
Case definition
A case-definition is a set of diagnostic criteria that must be fulfilled in order to
identify a person as a case of a particular disease.

Four criteria for a case definition: symptoms, person, place, time

1. well-defined clinical symptoms (with our without laboratory confirmation)

2. information relating to the time (time of onset of symptoms)
3. persons affected
4. the place or location where the cases are occurring
MCQ 27 (1 mark)

When undertaking prevalence studies, the World Health Organization defines osteoarthritis on the basis of
symptoms such as joint pain, the finding from the physical examination often accompanied by radiographic imaging.
What is this collection of clinical criteria used for?

To:

A. determine the health outcomes of the patient after the osteoarthritis.

B. establish whether the diagnostic tests are accurate and reliable.

C. provide a case‐definition for the diagnosis of osteoarthritis.

D. report the statistical information about osteoarthritis to health authorities.

MCQ 27 (1 mark)

When undertaking prevalence studies, the World Health Organization defines osteoarthritis on the basis of
symptoms such as joint pain, the finding from the physical examination often accompanied by radiographic imaging.
What is this collection of clinical criteria used for?

To:

A. determine the health outcomes of the patient after the osteoarthritis.

B. establish whether the diagnostic tests are accurate and reliable.

C. provide a case‐definition for the diagnosis of osteoarthritis.

D. report the statistical information about osteoarthritis to health authorities.

 Attack rate

Outbreaks - Just a measure of incidence in

an outbreak
- New cases/ population
Typically describes infectious agents
• Epidemic: a local outbreak- disease cases above normal/ expected
incidence levels in a local community or region
o Local outbreak of measles

• Endemic: constant baseline level of disease specific to that region or

population; disease always present in a low amount
o Malaria is endemic to Nigeria, endemic levels of influenza in Melbourne
• Pandemic: An epidemic that effects a large number of people across
international geographical boundaries- a “worldwide epidemic”
• COVID-19
Typically describes environmental non infectious agents ie. Cancer due to asbestos
or pesticide poisoning in workers
• Cluster: group of specific presentations or rare forms of disease in a specific time
period that occur at levels greater than would be expected due to chance
 Types of outbreak curves- “epidemic time curves”
Point source: one peak, triangle
shape
Curve has a steep upslope and a
gradual downslope
Continuous extended source: one
peak, longer, plateau shape
Curve has a sharp upswing followed
by a plateau and then a right tail
downward
Propagated/ contagious source:
many peaks, get bigger each time,
periods of minimal disease in
between
Serial transmission leads to an
epidemic curve with progressively
taller peaks
 Causation- sufficient,
necessary, component
causes, Bradford Hill criteria
Sufficient and necessary causes
• Sufficient- the exposure/ presence of factor is always enough to cause the disease ▪ If it is a chronic non-communicable
▪ Something will either be both necessary and sufficient or it is just a risk disease🡪 answer is likely a risk factor
factor- Sufficient causes of disease ALONE are rare because most diseases
are multifactorial or component cause
• A necessary and sufficient cause of Ebola Fever is the Ebola virus
• A necessary and sufficient cause of Huntington’s is the genetic mutation ▪ If an infectious/ communicable
• Necessary: the exposure is always needed to get the disease/ outcome; disease is disease🡪 necessary cause (presence
never present when this factor is absent of virus in serum)/ sufficient cause
▪ Getting infected by HIV is necessary to have AIDS
(being around sick people)
▪ Exposure to mycobacterium tuberculosis is necessary to get symptomatic
TB, but not sufficient (latent infection, need weakened immune system to
become symptomatic)
• Component cause
• Smaller factors that together with other factors may make up a sufficient cause
of disease but are NOT sufficient to cause disease on their own
• Each of those factors contribute to the disease a little bit on their own but you need
the whole pie for it to be a sufficient cause
Completely separate from:
• Risk factor
▪ These are different from necessary and sufficient, they are about risk
▪ Risk factors are exposures that increase your risk (or likelihood) of
getting the disease/ outcome
▪ Risk factors can also be sufficient or necessary causes but they don’t have
to be
The presence of pre‐existing hypertension in patients is linked to the occurrence of cardiovascular disease. Which
ONE of the following describes the causal relationship between hypertension and cardiovascular disease?

A. Component cause.

B. Contributory cause.

C. Necessary cause.

D. Sufficient cause.
The presence of pre‐existing hypertension in patients is linked to the occurrence of cardiovascular disease. Which
ONE of the following describes the causal relationship between hypertension and cardiovascular disease?

A. Component cause.

B. Contributory cause. NOT A THING

C. Necessary cause.

D. Sufficient cause.

If risk factor was an option, pick risk factor

 Criteria to establish causation/ Bradford Hill
criteria
• Low yield
• Does not prove causation, but presence of these factors can be
used to argue/ support that there is causation

TY
o Temporal relationship- exposure occurs before the

ILI
outcome/ disease- the only essential factor

IB
TEMPORALITY CONSISTENCY

RS
o Plausibility – is the causative relationship consistent with

VE
other knowledge? Does it make sense with other scientific
and biological knowledge? Eg. Animal studies?

RE
o Consistency- many studies give same finding
o Strength of association- what is the correlation co-

AT H
efficient? (r)

N
CI GT
SE
SP E

IO
o Dose response relationship- if you change the exposure

RE DOS

SO N
ON
dose = change in outcome (e.g. increase one, increases PLAUSIBILITY

AS TRE
other)

S
o Reversibility- if you remove the exposure, does this result
in no outcome?
o Study design- not a classical one, shouldn’t be here
 Hypothesis testing, p
values, type I and type II
errors
Hypothesis testing
1. Form your hypotheses

Null hypothesis: H0- that there is no difference between Alternative hypothesis: Ha, H1- there is a difference
the groups; that the results are due to chance between the groups, results are not due to chance.
∙ Baseline or default assumption ∙ What you are trying to prove
2. Choose a statistical test (hypothesis test) depending on the types of
data

3. Interpret your p value

• Level of significance/ threshold of significance= 0.05, will get a p value

between 0 and 1.
o If p value is below 0.05 (p< 0.05): there was a statistically
significant difference between the variables. Ie the results are
likely not due to chance. Reject the null hypothesis ( inferred that
you are therefore accepting the alternate hypothesis)
o If p value is above 0.05 (p> 0.05): there was no statistically
significant difference between the variables. Ie the results are
likely due to chance. Accept the null hypothesis.
 Type 1 and type 2 errors
• Type 1- false positive rate= alpha
• Reject null (there is a difference) when null is true (there is no difference)
• Type 2 error- false negative rate= beta
• Accept null (there is no difference) when you should have rejected it (there is
a difference)
Other random
stuff
Validity- internal and external
Validity: the test is capable of measuring what it is intended to
measure; No systematic error, random error small.
We talk about the validity of a study
• Internal validity: validity of conclusions drawn within the study itself- degree
to which the study results are free from bias and confounding
• Are results correct for the group being studied?
• External validity: generalisability- the degree to which the observations are
true in other settings
• How do results generalise/apply to those not in it? To what wider population can we
generalise the results?
• Internal validity ≠ external validity
What is a nested case control?
• = a case control study NESTED WITHIN a wider cohort study (prospective
or retrospective cohort)
- Wider cohort study:
o No one has disease yet, varying levels of exposure eg. NSAID use
o Looking at osteoarthritis as an end outcome for the cohort study
o 1000 people
- Smaller case control study: taken at a snapshot in time, some time after
study onset and take A SMALLER SAMPLE OF PEOPLE WITHIN THE
WIDER COHORT STUDY, can look at different outcome
o Cases: people who developed the disease within the cohort study
ie. heart failure
o Controls: people who haven’t developed the disease yet but are
in the cohort study ie. no heart failure yet
▪ You sample a subgroup of the healthy people to match your
cases
o Ie. take 200 people with HF and 200 people without (matched to
cases), varying levels of NSAID use throughout
Good luck ☺