Observational Epidemiology Flashcards
When might an observational study be needed?
When an association between a potential cause and effect is found
What are the two types of epidemiological studies?
Descriptive
Analytic
What are the three types of descriptive study?
Ecological
Case reports/case series
Cross-sectional
What are the two categories of analytic study?
Observational
Interventional
What are the two types of observational study?
Case-control
Cohort
What is the one type of interventional study?
Clinical trial/RCT
How do observational studies different from interventional studies?
Observational studies look at naturally occurring causal relationships, whereas RCTs study controlled exposures
What are the three main things to consider in an observational study before looking at causality?
Bias
Confounders
Chance
What are the two broad types of bias which might be encountered in observational studies?
Selection bias
Information bias
What is selection bias?
The selection process of exposure groups is systematically different, therefore groups cannot be validly compared with each other or the general population
What types of bias fall under selection bias?
Sampling bias
Responder bias
Concerning occupational exposure what is the healthy worker effect?
The idea that people who are healthy enough to work cannot be compared with general practice, which includes elderly people and people unable to work - therefore the participants do not reflect the general population
What is information bias?
When the process of obtaining data from different groups differs systematically
What are 5 types of information bias?
Recall bias
Recording bias
Interviewer bias
Lost-to-follow-up bias
Social acceptability bias
What is recall bias?
The idea that unhelathy patients think more about a disease more than health patient,s therefore have more knowledge
What is recording bias?
The idea that unhealthy patients have more extensive medical notes than healthy patients
What is interviewer bias?
When the interviewer knows whether the patient has had the disease or not, and they subconsciously change their questions
What is lost-to-follow-up bias?
Patients lost to follow-up are more likely to be unhealthy patients with data of interest
What is social acceptability bias?
When the interviewee attempts to make their answers more socially acceptable
What is a confounding factor?
Something associated with both the disease and exposure, which could cause the disease and explains the apparent association between the exposure and the disease
What method can be used to check to see if factors are confounding or genuine?
Stratification
What does stratification involve?
Grouping results into subgroups of an exposure factor, and comparing how other exposure factors match up with that factor (or not)
What is the aim of stratification?
Identify and remove confounding factors by disentangling their effects
Give three ways in which disease can be measured (epidemiologically)
Incidence rate
Rate
Point prevalence
How is incidence rate calculated?
Number of new cases in a specified time period/Number of person years accumulated
How is rate calculated?
Number of new cases in a specified time period/Number of people at risk at the start
What sort of a measure is rate?
Probability
How is point prevalence calculated?
Number of people with the disease at a given timepoint/Total population at risk of the disease at the same timepoint
Give three things according to which descriptive studies characterise disease
Time
Place
Person
Give three aims of a descriptive study
Find out what type/characteristics of people are more or less affected by a disease
Assisted in evidence-based planning of healthcare
Give suggestions of disease aetiology, for further investigation with analytic studies
Give three types of descriptive study
Ecological
Case series
Cross-sectional
What is a cross-sectional study used for?
To identify the point prevalence of a particular disease in a particular population at a given point in time
What is a cross-sectional study useful for?
Determining need for specific health and social needs
Develop hypotheses concerning risk factors for a disease.
What are the pros of a cross-sectional study?
Quick
Cheap
Give a disadvantage of using a cross-sectional study
Disease and exposure are measured simultaneously, therefore it is hard to interpret cause and effect
Give a disadvantage of using a cross-sectional study
Point prevalence is a mixture of acute incidence and chronic patients surviving with the disease, which cannot be separated
Give a disadvantage of using a cross-sectional study
Cohort effects interfere when interpreting relationship of disease with age, as people born in the same year are likely to share an exposure which might have caused disease, rather than age
What is the unit of observation in ecological studies?
Groups of people, not individuals
What are ecological studies useful for?
Hypothesising the association between a disease and a chosen group, and a factor linking all the groups
Give three pros of ecological studies
Quick
Cheap
Generates hypotheses
Give a disadvantage of ecological studies
The group described by the disease data may not be the same as the people describe by the exposure data in the chosen group
Give a disadvantage of ecological studies
The time relationship between the measurement of exposure and of disease can be unclear
What is the main issue with using ecological studies?
One cannot assume that the relationship found at the group level will hold at the individual level
What are two key ways in which analytic studies differ from descriptive studies?
The unit of observational is individuals
The period of exposure is prolonged
What are the two types of observational analytic study?
Case-control
Cohort
What is the basis for defining a group in a cohort study?
Exposure to a risk factor
What is required to perform a retrospective cohort study?
Good records of historic exposure to risk factors
What type of exposure are cohort studies useful for?
Rare exposures
Which type of bias is eliminated by using a cohort study that is present in case-control studies?
Recall bias, because the study follows patients contemporaneously
Give three pros of cohort studies
Cohort studies produce estimates of absolute and cumulative risk of disease
Exposure is chosen first, so there is little doubt about cause and effect between exposure and disease
Cohort studies can assess any number of diseases with a risk related to the exposure
Give three cons of cohort studies
Cohort studies are not useful for rare diseases
It is expensive and difficult to follow a large cohort over a long time
If a disease is rare and only the relative risk associated with different levels of exposure is needed, then a cohort study would be unnecessarily expensive
Give three methods in which the exposure can be measured
Dichotomous
Categorical
Quantitative
What might be a good source of participants for a rare exposure?
Occupationally or therapeutically defined groups
What are two sources of bias in cohort studies?
Healthy worker effect
Lost-to-follow-up bias
How can lost-to-follow-up bias be designed out?
Over-recruitment of changing the recruitment methods
How is absolute risk calculated for the exposed population?
Patients with the disease and exposure/Patients with the disease and exposure + Patients with th exposure but without the disease
How is relative risk calculated?
Absolute risk of the exposed population/absolute risk of the unexposed patients
How is risk difference calculated?
Absolute risk of the exposed population - absolute risk of the unexposed population
Which values must risk be calculated from?
Rate of incidence
How can confounding be adjusted for?
Standardisation
Stratification
Use of a suitable regression model
How is the study group defined in a case-control study?
On the basis of a disease
What is the ideal for recruiting a population for a case-control study?
Use an existing, well defined cohort with a high number of cases
Give three pros of case-control studies
They are an effective use of resources to identify aetiology for a rare disease
More than one exposure can be investigated for the same disease
The relative risk of the disease’s association with different levels of exposure can be estimated
Give three cons of case-control studies
Recall bias: patients with the disease are likely to be more aware of it, therefore give more detailed histories
Only a single disease can be addressed
Absolute risk cannot be estimated, as one is starting with a population in the disease status.
Why must incidence be used over prevalence in case-control studies?
Prevalence cannot distinguish between new cases and chronic survivors, so incidence must be used to determine aetiology
How can bias be avoided when selecting the control group?
Select the control from the same underlying cohort as the case group
How can an unbiased control group be maintained if members of the control group get the disease?
If an individual in the control had got the disease, they would be entered into the case group
Give three methods which can be used to adjust for confounding
Standardisation
Stratification
Regression model
Give three sources of a control group
Underlying cohort from which cases were selected
Hospital/GP
Random digit telephone dialling
How can differences in result due to age/sex etc be avoided?
Match control participants with case participants according to age/sex etc
What steps must be taken to avoid bias is control and case groups are matched by age/sex etc?
Use of analyses which account for the matching
Which regression model is used if the case and control groups are unmatched?
Logistic regression
Which regression model is used if the case and control groups are matched?
Conditional logistic regression
Give five types of bias which can be encountered in case-control studies
Recall
Recording
Interviewer
Response
Sampling
What is recall bias?
The extent of recall about exposures differs systematically: people with a disease will have better recall than people without it
What is recording bias?
Subjects with the disease will have more detailed medical records than subjects without
What is interviewer bias?
Interviewers can have a non-neutral approach between case vs control participants
What is response bias?
Response rates may differ systematically between case and control participants, as those with the disease might be more health-conscious
How is relative risk theoretically calculated in case-control studies?
Incidence rate of exposed/incidence of unexposed
Why can’t the exact relative risk be calculated in case-control studies?
The incidence rates are unknown in case-control studies
How is risk ratio approximated in case-control studies?
Odds ratio is calculated and used to approximate risk ratio
How is odds ratio calculated in case-control studies?
Odds of disease in exposed/odds of disease in unexposed
How is odds of disease in the exposed population calculated?
Individuals with the disease (and exposure)/Individuals without the disease (and exposure)
How is odds of the disease in the unexposed population calculated?
Individuals without the disease (and exposed)/individuals without the disease (and unexposed)
What does a risk ratio of 1 represent?
There was no excess or deficit of disease in the risk of being exposed, therefore there was no effect of the exposure on the outcome, or the risk was equal in the exposed and unexposed
What does a risk ratio greater than 1 represent?
There was an excess of disease among those who were exposed, therefore there is a high risk of the exposure causing the disease
What does a risk ratio less than 1 represent?
There was a deficit of risk among those exposed, therefore there is a low risk of the exposure causing the disease
What do the Bradford-Hill criteria for causality assess?
Whether the association between exposure and outcome is actually causal or not, once chance, bias and confounding have been removed
What are the nine Bradford-Hill criteria for causality?
Biological plausibility
Time
Strength of association
Biological gradient (dose-response relationship)
Consistency
Specificity
Coherence
Experiments
Analogy
Which two Bradford-Hill criteria are particularly important?
Strength of association
Biological gradient (dose-response relationship)
What does biological plausibility question?
Whether the exposure causing the outcome actually makes biological sense or not
What does ‘time’ question?
Whether the cause precedes the future disease outcome or not
What does strength of association mean?
The greater the association between the exposure and the disease, the harder it is to attribute the association to confounder
What does ‘biological gradient/dose-response relationship’ question?
Whether the risk of disease increases with an increase in exposure or not
What does consistency question?
Whether the results match the outcomes of previous studies or not
What does specificity mean (as a Bradford-Hill criterion)?
An association between an exposure and a range of more than one disease is less likely to be causal than the association between an exposure and a single disease
What does coherence question?
Whether the results contradict what is already know about the natural history of the diease, or not.
What does ‘experiments’ mean (as a Bradford-Hill criterion)?
Occasionally experimental evidence can be used to confirm or reject causality
What does analogy mean (as a Bradford-Hill criterion)?
Use of analogies/similarities to compare the observed association and any other possible association
What do Forest plots from a meta-analysis have the statistical power to do?
Remove and examine chance and unadjusted confounders to the extent that the actual strength of the relationship between exposure and outcome can be identified