Epidemiology Flashcards
What makes global mental health multidisciplinary?
It draws upon
- social sciences
- clinical disciplines
What are the 3 roles of epidemiological methods?
- Quantify burden of disease
- Examine determinants of disease
- Evaluate interventions designed to prevent disease
Who was John Snow?
- Medical doctor
- ‘father of epidemiology’
What did John Snow suggest was the cause of the common cholera outbreaks (1849)?
‘It’s a water-borne disease’
What did John Snow do during the 1854 cholera outbreak?
> By talking to local residents, he identified the Broad Street pump in Soho as the most likely source of the disease
> He persuaded the council to disable the pump
-> the epidemic came to an end
> He demonstrated a link between the quality of the water source and cholera cases using statistics
What are the two key approaches to ascertain cause and effect?
- Inductivism
2. Refutionism
What does the inductivist approach consist of?
Bacon (17th century):
- observation leads to development of theory of the observed patterns
- Iterative: further observation leads to the development of more advanced theories
- number of times a pattern is replicated does not prove causality
- causality can not be directly observed
- > spurious notions of cause and effect
What does the refutionism approach consist of?
> Hume (18th century):
- uncertainty between cause and effect relationship was impossible to eliminate
> Popper (20th century):
- scientific knowledge is generated as a result of this uncertainty
- role of observation is to attempt to criticise and refute existing theories
- > building scientific knowledge
Which approaches does modern science use to determinate cause and effect?
- Inductive (hypothesis generation)
- Deductive (hypothesis testing)
-> inductivist and refutionist approaches
How are the inductive and deductive processes of modern science used in epidemiology?
- Ideas for exploratory research often emerge from clinical observations
- Results from studies lead to generate of hypothesis that might be tested in later research
What characterises a hypothesis?
- Iteratively builds corroboration for theories by seeking evidence to refute them
- Central to all scientific practice (including epidemiology)
- It is a statement predicting an outcome
- Commonly emerges from inductive research
What are the three criteria that make a hypothesis?
- Clearly stated
- Testable and refutable
- Statement, not a question or objective
What are the advantages procured by hypotheses?
- Support validity of findings
- Less risk of type 1 error (false positive)
- Ease of replication of findings by others
- Ensures research is driven by theory and observation
- Ensures research is a scientific process (rather than ad-hoc or data-driven)
What are the three types of epidemiological studies?
- Descriptive epidemiology
- quantify the burden of disease
(e. g. cross-sectional studies) - Analytical epidemiology
- examine determinants of disease, course and outcome
(e. g. cohort studies) - Evaluative epidemiology
- evaluate efficacy and effectiveness of interventions
(e. g. RCTs)
How do hypotheses help us structure our research practice?
Hypotheses shape methodology
- design and outcome measures
- statistical analysis plan
What are the three key observational study designs used in global mental health research?
- Cross-sectional
- Longitudinal (cohort)
- Quasi-experimental
What does the 10/66 dementia research group study consist of?
Dominican Republic
- series of cross-sectional surveys examining the health of older people
Research questions:
- how does this vary across time?
- how does this vary between regions?
How is the analysis conducted in cross-sectional studies?
> Participants recruited at one particular point in time
> Data collected at single timepoint
- face-to-face, clinical, census records
Why are we provided with a ‘snapshot’ of the data in cross-sectional studies?
Data for baseline characteristics and outcomes are ascertained at the same time
-> ‘snapshot’
What type of data is collected in cross-sectional studies?
- Frequency
- Characteristics
- Distribution of outcomes
What are cross-sectional studies useful for?
- Planning services
- Raising awareness
- Comparing regions/time
- Identify association with risk factors
(e. g. weight loss and dementia) - gather evidence to support generation of hypotheses for future investigation using different study design
What is the limit of cross-sectional studies?
We cannot be sure of the direction of cause and effect
What does the ‘snapshot’ of current disease cases represent in cross-sectional studies?
A reflection of survival and aetiology of disease
How is the prevalence calculated?
Rate of new cases (incidence)
+ rate of cases disappearing (recovery, death, new diagnosis)
What is the incidence?
Rate of new cases
What happens to the prevalence of a disease in the case of greatly improved effectiveness in treatment?
Reduced number of incidence (new cases)
+ Reduced number of cases disappearing (reduced mortality, improved life expectancy with the disease)
= Increased prevalence (current cases)
-> less new cases, but more people living with the disease
What is needed to ensure results are generalisable and representative of the base population in cross-sectional studies?
- Define base population
- person (age, gender)
- place
- time - Define and enumerate the sampling frame
- is the aim to cove the Toal base population or part of it?
- routine date (e.g. census, clinic lists) or your own sampling frame (e.g. going door to door)?
Which bias needs to be avoided in cross-sectional studies?
Selection bias:
- sample differs from base population
What were key features of the 10/66 dementia research group study?
- Careful consideration of ethics
- Communication with potential participants
- To reduce the burden to participants to maximise inclusion
What is the key aim of cross-sectional studies?
Accurately calculate burden or prevalence of a particular outcome
How do you achieve an accurate burden or prevalence number of a particular outcome in a cross-sectional study?
- Accurately determine X: number of people with particular outcome at particular time
- Accurately determine Y: number of people in base population at particular time
How is the analysis conducted in longitudinal studies?
- Recruitment (exposure status determined)
- Data collected at several points (follow-up)
What is a key element of data collection in longitudinal studies?
The outcome cannot have occurred at the time of baseline data collection
e. g. UMEED cohort study:
- people who had accessed HIV services were ineligible since the study examined the association between depression status and use of HIV services
What are the aims of longitudinal studies?
- Compare outcomes
- Determine the number of new occurrences of the outcome
What are the six steps to carry out a longitudinal cohort study?
- Hypothesis
- Identify base population
- Define and measure exposure
- Define and measure confounders
- Define outcomes
- Follow-up, analysis, interpretation
What are confounders?
Factors which we think might influence results
- providing we measure confounding factors, we can adjust for their effects in our analysis
Which elements should be controlled when collecting data in longitudinal cohort studies?
> Avoid selection bias
> Reduce potential for bias
> Avoid systematic errors in the classification of outcome status
- so exposure is not associated to misclassification
> Minimise observer bias: blinding
- researchers should have no knowledge of exposure status
> Collect comprehensive info on potential confounders
-> adjust for their effects in analysis
What are the methods used to collect exposure data?
- Routine records
- Interviews
- Clinical exam
- > depends on the nature of exposure and available resources
What happens when the misclassification of exposure status is non-random?
Outcome is linked to errors in the determination of your exposure status
What happens when the misclassification of exposure status is random?
Errors in determining exposure occur randomly with no link to outcome status
Why should the quality of data collected from the exposed and unexposed be the same?
Helps us quantify its effect on our results
What are the methods used to collect outcome data?
- Routine data
- Self report
- Standardised assessment
How do you reduce potential for bias?
Researcher collecting outcome data should be blind to the exposure status
- e.g. use different researchers to collect exposure and outcome data
What is a loss to follow up?
Bias that can occur due to flagging (people losing interest)
- common when there are multiple waves of data collection
- participants also lost due to migration, death, being uncontactable
How does the loss to follow up influence results?
- Attrition (loss to follow up) varies between exposed and unexposed
- Random or differential attrition influences sample sizes
- > reducing statistical power
How can you reduce loss to follow up?
- Maintain contact with participants
- Make multiple attempts to contact those who are lost
- Effects of loss to follow-up can be adjusted for in analyses (input missing data, sensitivity analysis)
What are the three types of quasi-experimental designs?
> Non-randomised controlled trials
> Uncontrolled before and after studies
> Time series
What do quasi-experimental designs consist of?
The observation of what happens to participants after the introduction of an intervention
What do observational study designs consist of?
The observation of what happens to participants in a naturalistic manner
What are the three types of observational study designs?
> Cross-sectional
> Longitudinal cohort studies
> Case-control studies
What are the two types of experimental study designs?
> Pragmatic RCTs
> Scientific RCTs
What do experimental study designs consist of?
> Study carried out under tightly controlled conditions in order to minimise bias
> We can isolate the effects of an intervention upon outcomes
-> Gold standard study designs
How is the analysis conducted in quasi-experimental studies?
- Recruitment
- Baseline measures
- Introduction of the intervention
- Outcome measures collected
What is the assumed origin of observed differences in quasi-experimental studies?
Assumed to be due to the intervention
How are non-randomised controlled trials conducted?
> 2 groups (intervention, control)
> Outcomes compared between the 2 groups
> Ensure groups are as similar as possible
- population
- measurement
- data collection
> Data collected before and after intervention introduced in treatment group
How are uncontrolled before and after studies conducted?
> One group
> Measures are compared before and after the intervention
What is the limit of quasi-experimental studies?
It is often difficult to determine that observed differences are as a result of the intervention (cause and effect)
How can a relationship be assumed to be truly causal?
Meet the Bradford Hill’s criteria of cause and effect
What are the Bardford Hill’s criteria of cause and effect?
- Strength (appropriate tests)
- Consistency (of results with different studies)
- Specificity
- Temporal sequence (exposure before outcome)
- Dose response (increased exposure -> increased risk)
- Experimental evidence (condition can be altered by appropriate experimental design)
- Biologic plausibility (association agrees with currently accepted understanding of pathobiological processes)
- Coherence (association compatible with existing theory and knowledge)
- Analogy (analogous associations between similar factors and similar diseases)
When are quasi-experimental studies useful?
- When randomisation is not possible (e.g. roll-out of intervention is outside researcher’s control)
- Useful as proof of concept studies
e. g. Is the change only due to the intervention
- confounders, natural remission, service setting changes
What are the limits of quasi-experimental studies?
- It is difficult to identify a sufficiently similar control group to eliminate confounder’s influence on results
- It is difficult to attribute cause and effect
- > Researchers must protect against external changes
- > Caution when interpreting results
How are randomised controlled trials (RCTs) conducted?
> Participants are randomly allocated to receive either the intervention or the control
> Follow-up
> Compare outcomes
What is the power of randomisation found in randomised controlled trials (RCTs)?
It’s possible to say that observed differences in outcomes are the result of the intervention with a greater degree of certainty
What is the simplicity in a randomised controlled trial (RCT)?
The allocation of participants to intervention or control can be done by tossing a coin
What happens to the factors that may influence outcomes in a randomised controlled trial (RCT)?
They are randomly distributed between groups apart from the allocated interventions
- > ‘perfect’ control for confounding
- > Outcome should be attributable to the intervention alone
How can biases still be introduced in a randomised controlled trial (RCT)?
- Randomisation is not properly implemented
- > non-random allocation
- Randomisation is properly implemented BUT group difference may still occur
How can we avoid potential biases in randomised controlled trials (RCTs)?
- Allocation is random and concealed (e.g. opaque envelopes)
- Carry out randomisation independently of the research team (e.g. outsource to clinical trials unit)
- Consider and measure a priori confounders and their distribution between groups
- Use different techniques to ensure balance between groups
(e. g. block randomisation or stratification)
What is the role of chance in randomised controlled trials (RCTs)?
We cannot eliminate this possibility and the consequent differences in outcomes
e.g. Random sampling error
- difference is not represented in the population that the sample is supposed to represent
(difference in sample, not in base population)
How can we reduce a sampling error?
By increasing the sample size
Why is it better to have a large sample size?
Larger trials increase precision and power to detect a treatment effect of given size
- reduced range of confidence intervals
- reduced chance of type I error
How do we quantify the level of uncertainty associated with the role of chance in determining results?
Using statistics: 95% confidence intervals
-> range of possible true effects sizes
What is a type I error?
False positive
How can we minimise the information bias?
Blinding
What it the limitation of blinding?
With mental health interventions it is often difficult
- e.g. intervention is talking therapy, control is an info. leaflet -> difficult to conceal
(compared to placebo controlled trials)
