Epidemiology

Question 1

What is a cross-sectional study?

Answer 1

A study performed at one point in time, where the prevaalance (Outcome) is measured at the same time as the exposure.

Question 2

Cross-sectional study uses what calculation?

Answer 2

Prevalance risk ratio, or odds if rare outcome

Question 3

Cross-sectional study disadvantages? (6)

Answer 3

1. It is at one point in time so lacks the temporality aspect needed to infer causation (can only see an association)- reverse causality risk.
2. Prevalance not incidence- unsure whether this is high infection rate or people just not being treated, if short lived missed?
3. Selection bias- no response bias if a survey, and careful selection of controls needed for descriptive.
4. Recall bias if self confessing their exposure.
5. confounding high possibility.
6. Not useful for rare outcomes.

Question 4

Cross-sectional study advantages? (3)

Answer 4

1. Quick and easy
2. Can repeat study over time- see timeseries of
3. Provides prevalance and risk factors numbers- useful for budgeting and resource allocation for.

Question 5

How is a cross-sectional study ususally undertaken?

Answer 5

Survey usually- descriptive or analytical

Question 6

Two types of cross-sectional study designs? definitions and examples?

Answer 6

Descriptive: Often no Hypothesis- wants more information on. focus on either exposure or outcomes and find information on the other e.g. MSK disorders in waste workers or controls- focus on jobs and then find the prevalance of MSK.

Analytical: Hypothesis testing. Investigate the outcome and exposures simultaneously without focusing on either for sampling. E.g. Questionnaire for mothers on breastfeeding amount and childs BMI- see if an association?

Question 7

Why could selection bias be a problem in descriptive cross-sectional studies?

Answer 7

Selection of controls needs to be mindful of other confounders. E.g. the MSK in waste collectors- the control chosen were office workers- likely of higher socio-economic background.

Question 8

Why could bias be a problem in analytical cross-sectional studies?

Answer 8

Non- response bias, recall bias.
E.g. in the example of mothers questionnaire on breastfeeding and childs BMI- those whose child has a high BMI less likely to take part? Also mothers may lie or misremember breastfeeding amount- stigmatised and may be about 5 years prior.

Question 9

What is the problem with using a workplace as a sample population?

Answer 9

Healthy-worker effect. If do surveys etc and use a workplace as a study population, those working are likely healthy- if was ill wouldnt be working- also other confounders may likely be one socio-economic group, more one age group? Been to university etc

Question 10

What is an ecological study?

Answer 10

A study at population/group level with no access to the individual data. This could be cross-sectional or as a longitudinal study, to see if incidience or prevalance increases with exposure etc. Find association not causality.

Question 11

Why would an ecological study be used? (4)

Answer 11

1. If data is not available at the individual level.
2. If the data is more helpful at the group level e.g. seeing the effectiveness/need for an intervention targeted at that group.
3. Compare different groups e.g. countries
4. If the risk is at population level e.g. pollution in a city

Question 12

Advantages of an ecological study? (5)

Answer 12

1. Group level data may be more reliable e.g. country level salt intake average- rather than having people estimate their salt intake, or pollution level (can use routine collected data)
2. Useful for hypothesis generation- see an association.
3. Cheap, easy, quick (can use secondary data sources)
4. Can do over time to see if changes (timeseries)- useful if want to study one set group.
5. Useful if data varies more between groups than within.

Question 13

disadvantages of an ecological study? (4)

Answer 13

1. Causality- cannot prove as dont know that the individuals with the outcome have the exposure etc
2. High confounding likelihood e.g High pollution and lower LE- could be other factors- those in city less well off?
3. ECOLOGICAL FALLACY- cannot say that the association at group level means association for the individuals.
4. Information bias- some countries may record information better/ use different criteria- this can also change over time.

Question 14

The big problem that can happen with ecological studies?

Answer 14

ECOLOGICAL FALLACY- cannot say that the association at group level means association for the individuals.

• don’t know those with the outcome and those exposed
• people migrate
• confounders
• bias- different definitions/ measuring over time/place.

Question 15

Uses of a Time-trend study/ timeseries?

Answer 15

Can investigate how incidence/prevalance changes with exposure over time e.g. seasonally, annually, daily variation etc- strengthens the association (one of the 8 Bradford Hill criteria for causation)
-Disadvantage if over extended time- measuring reliability may change, use different definitions. thresholds etc

Question 16

WHat is a cohort study?

Answer 16

Focus on the exposures of interest, then follow up study investigates whether get the outcome. It is natural observation experiment with no intervention.
Prospective: Measure exposures, follow up to measure future outcome incidence.
Retrospective: Look at past exposure of participants from historical records, and then measure the incidence of outcome.

Question 17

Equations used for cohort study?

Answer 17

Risk, Rate (if follow up times differ) or odds (if rare- but usually common or need v large sample) from incidence cases. Can dp relative ratios exposed/unexposed risk to work out association.
Attributable Risk, or Population Attributable Risk can be perfomed to work out excess risk due to the exposure.

Question 18

Advantages of a prospective cohort study? (5)

Answer 18

1. Very accurate measuring of exposures- measure so no past records needed.
2. Can follow up over time, to see when develop outcome/ track confounders/ exposure changes
3. Clearly know exposure precedes the outcome- temporality for causation (Bradford Hill criteria)
4. Can study a range of outcomes from a given exposure.
5. No recall bias- measure everything

Question 19

Disadvantages of a prospective cohort study? (3)

Answer 19

1. TIME, money- follow-up can be for decades- if outcome takes years to develop- lag time.
2. Loss to follow up- if long follow-up times. Loss of power if everywhere, or one group more than others? cause selection bias e.g. those without exposure less invested in study- selection bias
3. confounders- but can often measure these.

Question 20

Advantages of a retrospective cohort study? (4)

Answer 20

1. Quicker- if large lag time don’t need to wait, see past exposures instead.
2. Can look at lots of exposures on records.
3. Clearly know exposure precedes the outcome- temporality for causation (Bradford Hill criteria)
4. No loss to follow-up

Question 21

disadvantages of a retrospective cohort study? (3)

Answer 21

1. Rely on the accuracy of historical records- definitions/ criteria can change over time, or be poorly recorded
2. May have recall bias, if records not used.
3. Confounders not measured (low chance though).
4. Systematic misclassification bias- people systematically put in the wrong groups if outcome knowledge known.

Question 22

cohort study controls?

Answer 22

Select controls that dont have the exposure of interest who are similar to the study population in every other way.

Question 23

PAR equation?

Answer 23

Incidence in population- incidence in unexposed= incidence in exposed

Question 24

AR equation?

Answer 24

Incidence in exposed-incidence in unexposed= excess risk from exposure

Question 25

What is a Case-control study?

Answer 25

Starts with identifying those with and those without (control) the outcome, then work back to see the exposures- focus on the outcome not exposures like in a cohort.

Question 26

Case-control advantages? (8)

Answer 26

1. Useful for rare outcomes. 2. Can look at multiple exposures for an outcome. 3. Cheaper and faster than cohort. 4. Useful when log latancy period. 5. Useful when rapid results needed e.g. an outbreak. 6. Must have prior hypothesis so avoids fishing for results 7. Low probability of loss to follow-up 8. Useful for genetic markers/ mutations in a family

Question 27

Case control disadvantages? (6)

Answer 27

1. No longer a representative sample- can only do odds ratio of exposure, as cant calculate risk without having a population total. Only say 100 with disease, 100 without. 2. Control selection difficult: have to be ssame besides the disease state e.g. if cases are in hospital a lot, should the controls be? confounding risk. 3. Overmatching risk- if match too closely may also match accidentally for the exposure e.g. geographical area and pollution. 4. Recall bias if reliant on individuals saying their exposure- and patients may be more likely to remmeber exposure than healthy. 5. reverse causality- disease could shape exposure e.g. in hospital not in polluted environment. 6. cannot distinguish between factors causing the disease and prolonging it- if prevalance is used.

Question 28

Calculation used for case-control

Answer 28

Odds ratio of exposure- instead of risk= those with disease/ total, don't have the total here so calculate the exposure/unexposed and do a ratio of these for with and without outcome to work out whether the outcome sample have a higher proportion who are exposed than the control group. = odds of exposure in cases/odds of exposure in controls

Question 29

WHat is an observaition study?

Answer 29

One that observes and does not intervene, but (intervention), but just observes.

Question 30

What is an intervention study design?

Answer 30

A study where participants are actively allocated to an intervention by the investigators e.g. a Randomly controlled trial.

Question 31

Advantages of an RCT? (3)

Answer 31

1. 'Gold standard' ideal for conferring causalty. 2. Minimises confounding factors (known and unknown) by randomly allocating participants (or matching) 3. Multiple outcomes can be studied by the interventions.

Question 32

Disadvantages of an RCT? (6)

Answer 32

1. Expensive. 2. Ethical problems associated. 3. Long follow-up drop out problems. 4. How generalisable, low external validity? Often fit males. 5. Conflicting trial evidence e.g. a metaanalysis better as summarises. 6. Need large sample size

Question 33

Ethics behind a RCT? (4)

Answer 33

1. Need informed consent. 2. Compare to treatment as usual TAU instead of no treatment. 3. Participants can leave the trial at any time 4. The trial can be stopped at any time e.g. If unexpected outcomes, can see no benefits, or even if clearly can see huge benefits and witholding treatment from controls.

Question 34

Efficacy vs efficiency studies?

Answer 34

Efficacy study: Effectiveness of a study under optimal conditions 'per protocol analysis' use risk ratio. Selection bias- if high side effects may drop out and not be included in the study. Efficiency: Under routine conditions- more transferable say if have poor infrastructure, 'intention to treat' real life conditions- keep random composition- so not keeping partparticipants in the calculations even if drop-out. E.g. Rate ratio

Question 35

How are confounders and bias reduced for in an RCT?

Answer 35

Confounding: Random allocation, matching known confounders. Bias: Blinding- investigator (observer bias), participant (reporter bias), analyst (analyst bias).

Question 36

Equations used in intervention studies?

Answer 36

Preventable fraction: (incidence in unexposed- in exposed)/ Incidence in unexposed. Numbers needed to treat: 1/ARR Absolute Risk Reduction= Risk outcome in intervention-risk in controls.

Question 37

Intention to treat vs 'per protocol analysis'

Answer 37

Intention to treat: Keeps everyone in the participant group even if they stop the treatment/ not completely compliant as people not always are in real life- avoids biases and keeps the random composition. May give a conservative estimate. Per protocol analysis: Looks only at the people who stick to the protocol- could only be looking at those who dont get side-effects.

Question 38

4 other reasons that a significant result may not be true causality? (not definitions)

Answer 38

1. Bias-information or selection- 2. Confounding 3. Chance 4. Reverse causality

Question 39

Bias definition:

Answer 39

Where a difference is introduced that is not due to the expossure but difference between case and controls. A systematic difference. Information bias: difference introduced in the data. Selection bias: difference introduced in the selection process.

Question 40

Three examples of selection bias?

Answer 40

Non- response bias Loss- to follow up Control selection bias

Question 41

WHat is non-response bias?

Answer 41

Differential loss from a certain group. This is where a difference is introduced by the fact that one group may be less likely to participate so will skew the results e.g. the study on breastfeeding and BMI of children, the mothers with the children overweight may be less likely to participate. In a RCT often healthy males, no children, elderly or ill can take part ususally.

Question 42

What is loss-to follow up?

Answer 42

Where a systematic difference is introduced due to a greater drop out from one group e.g. those with lots of side effects may drop out, and this skew the results. Could result in an underestimate or overestimate of the results.

Question 43

How could selection bias be introduced in cohort, case control, RCT, cross-sectional?

Answer 43

Where other differences could be introduced that are not just due to the differences in exposure. Cross-sectional: Non-response bias. Cohort: Healthy worker effect. e.g. two different socioeconomic classes- Waste collectors vs office workers- not just job dangers but other confounders. Loss-to-follow-up. Case control- want controls similar in every other way so if cases spend a lot of time in hospital, should the controls? RCT: self-selected sample, loss to follow up

Question 44

Information bias measures? (5)

Answer 44

1. Validity: Internal- degree that the equipment measures what it is supposed to i.e. thermometer for temperature. 2. Validity: External- degree to which the study is generalisable to the wider population. 3. Reliability: Degree to which the results can be replicated. 4. Accuracy: can have small SE but off-target from the true value (may not know) E.g. dart board close together but off centre- BIASED. 5. Precision: SE large bad precision but the average would be around the true value- not Bias. E.g. dartboard all over the board.

Question 45

Causes of information bias?

Answer 45

1. Missclassification: If differential then causes bias- only one group being misclassified . Non-differential will underestimate effect but not bias. 2. Participant bias/reporter/recall: Unreliable/unaccurate reporting- if unblinded may have placebo effect, or misremmeber past exposures. 3. Instrument bias: Wrongly calibrated instrument for one group introduces bias. 4. Observer bias: e.g. leading questions from interviewer

Question 46

What can reduce the risk of differential misclassification and observer bias?

Answer 46

Blinding the assessor so they do not know who is in which group

Question 47

What can reduce participant bias?

Answer 47

Blinding to reduce the placebo effect, giving a placebo, and if recall bias- use records to determine exposure where possible.

Question 48

How can instrument bias be reduced?

Answer 48

Use modern automated instruments instead of observations- ensure the same machine type is used to measure both groups- if possble same machine.

Question 49

What is confounding?

Answer 49

Another factor that is independently associated with both the outcome and exposure which influences the association.It is not on the causal pathway. e.g. triangle with white middle aged men and lung cancer but real cause is the middle aged men smoke more which causes cancer, not the association of interest.

Question 50

Common confounders?

Answer 50

Age, sex, socio-economic group

Question 51

How can confounding be controlled for at the trial stage? (3)

Answer 51

1. Randomisation- RCT controls for unknown confounders as well as known by randomly allocating people to a group. 2. matching: Could match age and sex in trial participants. 3. Restriction- do not extrapolate data beyong the study e.g. if study on all under 50s, dont presume works for 60 year old.

Question 52

How can confounding be controlled for at the analysis stage? (3)

Answer 52

1. Stratify for- work out associations separately for each group e.g. split into age groups, or into smoking frequency. 2. Standardise e.g. age standardisation. 3. Statistical modelling e.g. multiple linear regression models.- simultaneously adjust for other factors and find if there are associations.

Question 53

What is the definition of chance?

Answer 53

The possibility of random error. E.g. the possibility of accepting/rejecting the null hypothesis when the true value would not reflect this. The risk of a Type I error is usually at 5% (p<0.05) and the risk of a Type II error is usually at 20% (80% power)

Question 54

What is effect modification?

Answer 54

If the association only explains half of the story e.g. it lies on the causal pathway.

Question 55

WHat is a confidence interval?

Answer 55

A 95% CI contains 95% of the data (+ or- 2xSE), if the study was repeated 100 times, the true value will lie within 95 times.

Question 56

What is reverse causality?

Answer 56

where the outcome causes the exposure, not the other way round- more in a cross-sectional study. E.g. Diabetes type II (outcome) and healthy eating (exposure), but likely the other way round where get diabetes then eat healthier.

Question 57

Bradford Hill Criteria for causality?

Answer 57

1. Strength association 2. Consistancy- repeated. 3. Temporality (exposure before outcome) 4. Dose response 5. Plausability 6. Reversibility: if remove exposure, reduce outcome. 7. Coherence: consistant with other studies 8. Analogy: Cause and effect relationship. 9. Specificity: E.g. helmet reduces head injuries not whole body

Question 58

What is the CMR?

Answer 58

The ratio of two directly standardised mortality rates (DSR) is called the Comparative Mortality Ratio (CMR).

Question 59

WHy use incidence cases instead of prevalance in say a case-control?

Answer 59

1. Prevalent cases tend to be those with moderate disease (mild cases recover fast, severe cases die) 2. Prevalent cases might have changed their exposure (‘reverse causality’) 3. Prevalent cases might show recall bias in how they recall their exposures, especially if they are aware of the link between exposure and disease 4. Prevalent cases cannot be used to distinguish between factors causing a disease to persist in the community and factors causing a disease.

Question 60

What is selection bias?

Answer 60

Selection bias occurs when there is a systematic difference between the characteristics of individuals sampled and the population from which the sample is taken, or a systematic difference between the comparison groups within the study population. Most important in case-control studies: cases and controls should only differ on the outcome and exposure of interest.

Question 61

what is information bias?

Answer 61

Information bias occurs when there is a systematic difference between comparison groups in the way that data are collected. Can be introduced by those measuring the outcome (observer bias), the study participants (responder bias), or measurement tools (measurement bias).

Question 62

The screening test has a low positive predictive value. How could this be improved?

Answer 62

Predictive values are dependent on prevalence of disease. Therefore selecting a higher risk group (with a higher prevalence of disease) to screen would improve the positive predictive value.

Question 63

Two potential outcomes in a study of recall bias?

Answer 63

If the same degree of underreporting occurs both in cases and controls, then non-differential misclassification occurs and the direction of association will not change. But this may lead to underestimation of the strength of the association between obstetric events and maternal death. If the degree of underreporting differs between cases to controls then differential misclassification occurs and this may lead to information bias and over- or under-estimation of the true magnitude of the measure of association due to the change in the direction of the association. .