Module 1: Distribution and Determinants of PopHlth Flashcards
What is epidemiology
The study of frequency/occurrence of dis-ease (N) in populations (D) over a period of time (T)
ALWAYS starts by describing a population, then count the number of cases of dis-ease that occur in the population
Dis-ease frequency aka…
Dis-ease occurrence
Dis-ease risk
Dis-ease distribution
Why measure the frequency of dis-ease in different populations?
If frequency (distribution) of dis-ease is different between 2 populations, this can help identify the causes (determinants)
How to calculate frequency of dis-ease
[Number of cases of disease] / [Number of people in population]
Definition of a population/group
A group of people who share one or more common features
Definition of dis-ease
Narrow: the absence of death, disease, or disability
Broad: the capacity to do what matters most to you
Numerator and denominator
Numerator: cases of disease
Denominator: population
Epidemiology = (N/D) / T
Define GATE
Graphical Approach To Epidemiology
A map of all epidemiological studies
What is PECOT
The 5 parts of every epidemiological study Participants/Population Exposure group Comparison group Outcomes Time
EG and CG
Exposure groups and Comparison groups are denominators for calculating dis-ease occurrence
Which ‘exposure’ can be chosen as EG or CG?
Any, as long as you are clear and state which group is which
PECOT: Outcomes
a & b (or c & d) are numerators
Mostly use those with dis-ease (a & b)
Goal of epidemiological studies is…
To calculate:
Exposure Group Occurrence (EGO) and
Comparison Group Occurrence (CGO)
Effects (RR and RD) in the whole population
Time arrows
Down: Over a period of time
Across: At one point in time
Incidence
Where the number of dis-ease events that occur are counted forward from a starting point over a period of time
EGO and CGO: incidence measures of occurrence
Prevalence
Where the number of people with dis-ease are counted at one point in time
EGO and CGO called prevalence measures of occurrence
Less ‘perfect’ measurement - can lose some incidence rain through cure cloud and death drops
When to use incidence or prevalence
Incidence: if easy to observe when dis-ease occurs
Prevalence: if hard to observe when dis-ease occurs, we measure IF it has occurred
Incidence rain drops
Each raindrop is a person having a dis-ease event
Prevalence pool
If too difficult to measure each raindrop in the drizzle as it falls, measure how much ‘water’ there is in the prevalence pool after the drizzle has fallen
Incidence and prevalence; categorical or numerical measurement?
Incidence:
Always involves counting categorical (yes/no) dis-ease EVENTS
Prevalence:
Involves counting categorical dis-ease EVENTS
OR
Measuring numerical dis-ease STATES
Cohort vs cross-sectional studies
Cohort - follow over time; can also measure prevalence - either at beginning or any point during the study (e.g. video and taking snapshot)
Cross-sectional - relevant dis-ease events counted at the same time - can only measure prevalence
Video questionnaire
Show video of symptom/dis-ease as some diseases may translate differently in different languages
Prevalence - time measured ‘backwards’?
Measure a period of time, e.g. if someone has had the symptom within the last 3 months
EG and CG - different dis-ease frequencies?
Could be due to medicines/drugs
If prevalence pool:
Some people may have been cured or may have died (cured cloud and death drips)
Randomised control trials (RCTs)
Like cohort studies, except participants are allocated random to EG or CG
Ideal study, but only if it’s both ethical and practical
Only done when pretty sure beneficial and not harmful
Benefits of RCTs
Participants have equal chance of being allocated to EG or CG, so any differences between the groups are likely to be due to effect of the drug they are given
Unblinded RCT
Both patients and investigators know which intervention was given to the patient
Single-blind RCT
Only investigators know which intervention was given to which participant
Double-blind RCT
Neither participants nor investigators know which intervention was given to which participant
Benefit: investigators can’t accidentally give away any hints to patients
Risk difference
Difference in risks
EGO - CGO or vice versa
Risk difference vs relative risk/risk ratio
Relative risk often less useful than risk difference - all decisions should be based on RD not RR alone
Relative risk/risk ratio
Ratio of risks
EGO/CGO or vice versa
Benefit of a treatment is dependent on…
Risk of dis-ease BEFORE treatment is started
Error in epidemiological studies
Occurs when the wrong people are recruited into a study, or the right people are put in the wrong GATE frame –> EGO and CGO incorrect
Random error
Error that occurs by chance
Why do non-random errors / bias occur
Due to poor study design, processes, or measurements
Study validity
A study with only a small amount of random or non-random error is considered to be a valid study
Where do non-random errors occur?
RAMBOMAN Recruitment of participants into study Allocation of participants to EG and CG + Adjustment in Analyses Maintenance of participants in EG and CG during study period Blind or Objective Measurement of Exposures/Outcomes ANalyses
Main ways of allocation
By measurement/observation
By random allocation
Confounding
When exposure is mixed with another factor that is also associated with the outcome
To deal with this, divide/stratify the study into sub-studies/strata so participants with the confounder are all in one sub-study i.e. adjustment
Cross-sectional studies and maintenance
There is no follow-up time in cross-sectional studies because participants aren’t followed up, so maintenance error is not a problem
What to do if participants aren’t a representative sample of a population
Instead, can compare sample with other similar samples in different countries
Risk difference and Risk ratio - units
RD: units (/people/time)
RR: no units
RAMBOMAN - Maintenance
Did participants remain in their allocated groups (EG or CG)?
Were many participants lost to follow-up (from either EG or CG) (can’t happen in cross-sectional study as not followed over time)
RAMBOMAN - Blind or Objective Measurement
If outcome is death, it’s objective
If outcome is cause of death, less objectively measured - requires personal interpretation
Measurement errors often not applicable to RCTs
Blinded studies reduce subjective measures
Paper vs video questionnaires
Video questionnaires measure more objectively compared to paper questionnaires
Ecological randomised study reduces chances of…
Confounding
Extreme events are often ____ events
Chance/random
Why is measuring the exact ‘truth’ not possible in biology/epidemiology?
The participants are moving targets, so identical measurements of exposures and outcomes in the same or similar people can change from moment to moment
Called random measurement error
Random measurement error affects…
Measurement of both exposures (E and C) and outcomes (O)
‘Identical studies’
Identically designed and implemented studies will never include participants with identical characteristics
‘Identical studies’ will produce different results
Random sampling error
The smaller the sample, the greater the chance the sample will be different from the whole population, i.e. the greater the random sampling error
95% confidence interval
A measure of the amount of random error in our estimates of EGO, CGO, RR AND RD in the whole population when you only have done one study
95% confidence interval - actual definition
In 100 identical studies using samples from the same population, 95/100 of the 95% CIs will include the true value for the population
“There is about a 95% probability that the true value of EGO in the whole population of interest, from which the study participants were recruited, lies between __ and __”
Confidence intervals describe…
The range of results likely to include the true result in the whole population
Every epidemiological measure has random error which can be estimated by a…
Confidence interval
Other types of confidence intervals (not 95%)
90% and 99%
99% CI is wider (more uncertainty); need a bigger net to catch the true effect of 99% of the time compared to 95% of the time
If CI of CGO doesn’t overlap with CI of EGO…
There’s a statistically significant difference between them
If CI of RD doesn’t overlap with no-effect line…
There’s a statistically significant effect
No-effect line
Where EGO = CGO so RD = 0, or RR = 1
If CI for EGO and CGO do overlap…
There’s probably no statistically significant difference between EGO and CGO
How to reduce random sampling error
Do a bigger study; bigger sample –> less chance sample will be different from whole population –> less random sampling error –> narrower CI
How to reduce random measurement error
Regression to the mean; repeat measurement/study - often results are less extreme
Meta-analysis
Combining studies in a meta-analysis is the next best thing to doing a larger study and reduces random error
Mainly done using RCTs
Reverse causality
Common error in cross-sectional studies, as you measure exposure and outcome at the same time
e. g. must ask if ‘diet’ was before ‘heart attack’
i. e. effect may come before cause
Individual participant studies - confounding
Investigators measure as many confounding factors as they can –> adjust as much as they can
Conservative measure bias
Where RR is closer to 1.0 and RD is closer to 0
RRI and RRR
Relative Risk Increase = (RR - 1) x 100% = > 1
Relative Risk Decrease = (1 - RR) x 100% = < 1
ARR and ARI
The RD is an Absolute Risk Reduction if the risk is lower in the EG
The RD is an Absolute Risk Increase if the risk is higher in the EG
EGO and CGO are measures of ______
Occurrence
RR and RD are measures of ____
Effect
