Observational studies Flashcards
What are the different types of epidemiological studies?
Descriptive = distribution of disease therefore help to generate a hypothesis
- ecological
- case reports/series
- cross-sectional studies
Analytic = uncover causes of disease to help analyse hypotheses
- observational = cohort and case control studies
- intervention = clinical trials
Define bias:
systematic deviation of results from the truth, or the processes leading to such deviation
What are the different types of bias?
Selection bias - sampling, responder and follow up bias
Information (obs) bias - recall bias / social acceptability bias, recording bias / interviewer bias
What is sampling bias?
occurs if each potential member of the population doesn’t have an equal selection into the study - more of a problem in an interventional study
What is responder bias?
some people are more likely to take part in these studies and therefore they may differ systematically to those who don’t and therefore they aren’t representative of the population
What are confounding factors?
effects of additional variable that might be responsible for an observed association
- situation where another factor associated with the exposure of interest and independently of that relationship the outcome
They can..
- cause spurious association
- exaggerate association
- or mask association between exposure and disease
Define incidence:
the total number of new cases commencing during a specified period in a defined population
Define prevalence:
total number of individuals who have the disease at a particular time
= incidence x duration of disease
What is the incidence rate?
number of new cases of a specific disease arising within a population over a specified time period divided by the person-time accumulated
What is the incidence risk?
number of new cases of a specific disease arising within a population over a specified time period divided by the number of persons at risk at the beginning of the time period
What is the point prevalence?
number of persons with disease at some point in time / total population at risk of disease at the same time point
What is period prevalence?
number of persons with disease at any time during a specified period / total population seen over the period of time
What does the ratio measure?
measures the strength of association
- risk ratio or odds ratio
What does the difference measure?
measures the magnitude of effect
- risk difference (therapy studies - interventional studies)
- attributable risk (aetiological studies)- measures how much disease in the disease group is caused by the exposure
What are the criteria for assessing causality?
1) biological plausibility - does it make biological sense
2) time - logically a cause must precede its potential effect
3) strength of association - stronger the association of an exposure with disease occurrence then the harder to conceive of likely confounders which might explain the association
4) biological gradient or dose-response relationship - causality as a plausible interpretation is strengthened if there is a strong dose-response
5) consistency - consistent with other studies in different populations, places and times add weight to a cause-effect interpretation
6) specificity - if supposed cause is associated with one disease only, or the disease is associated with one cause only, can add weight to causal interpretation
7) coherence - should not contradict what is already known about the natural history and biology of the disease
8) experiments - occasionally natural experiments offer themselves
What factors are important to consider when thinking about a cause and effect relationship?
First consider- chance, bias, confounding
Second - use criteria for causality as an aid to inferring causation
What are the functions of descriptive observational studies?
characterise disease in terms of time, place and person with several aims:
- alert medical community to what types of persons were most/least affected by disease
- assist in the evidence-based planning of health and medical care facilities
- to provide suggestions concerning disease aetiology for further investigation using analytic studies (hypothesis generation)
What are case reports/series?
careful detailed report of individual patients or a series of patients
suggest hypotheses
no control groups
e.g. MMR and autism
When are cross-sectional studies important?
if in a particular geographical region we were to determine the point prevalence of a particular disease
e.g. determining need for specific health or social services
How is the study population acquired for a cross-sectional study?
drawn from the target population using a sampling frame
- critical sampling because it affects the generalizability of the findings
What are the random and non-random methods of selecting your sample?
Random
- simple random sampling
- cluster sampling
- stratified sampling
Non-random - may produce a population that is atypical to your population
- systematic sampling
- snowball sampling
- street survey
What are the advantages of cross-sectional studies?
inexpensive
quick
What are the disadvantages of cross-sectional studies?
time sequence and causation (disease and exposure measured simultaneously and so sometimes hard to interpret results in terms of cause and effect)
Cohort effects when interpreting relationships with age
Problems with interpretation of prevalence which is a mixture of incidence and survival relating to a disease
What analyses can be carried out for cross-sectional studies?
Compare means = t-test
Compare proportions= chi-square
Correlation = pearson
Strength of association = prevalence ratio, odds ratio
What sources of potential bias are there in a cross-sectional study?
Selection bias - sampling bias - responder bias Information bias - observer bias - recording bias - subject bias
What are ecological studies?
correlation studies - measure rate of death or disease in a population and the population rate of a risk factor
- group of individuals as units of study
How are ecological studies analysed / interpreted?
scatter plots- visually inspect the relationship between exposure and disease
correlation (pearson’s correlation coefficient)
- numerically describes the association between exposure and disease, values range from +1 to -1, statistical significance of r is shown by a p value: null hypothesis is r=0
What are the advantages of ecological studies?
quick
inexpensive
hypothesis generation
What are the disadvantages of ecological studies?
subject to confounding
prone to sampling bias
prone to information bias
group of people described by disease data are often not the same people as those described by the exposure data
time relationship between the measurements for exposure and disease is often unclear
What 2 features distinguish cohort and case-control studies from cross-sectional and ecological studies?
both cohort and case-control studies:
- the unit of observation is at the individual level
- the period at risk tends to be prolonged
What are cohort studies?
studies in which a group of individuals who are defined on the basis of their exposure to a risk factor and are then followed up over time to determine who develops the disease of interest
- can be defined either prospectively or retrospectively
What are the key features of a cohort study?
exposure status known at start
- exposed group + unexposed group
prospectively follow up cohorts to see who develops disease
What are the key features of case control study?
disease status is known at the start
- have disease (cases) + disease-free (controls)
Retrospectively determine exposure in the 2 groups
What are the advantages and disadvantages of prospective cohort studies?
Adv
- exposure accurately measured
- confounders can be assessed
Disadv
- time consuming
- expensive
What are the advantages and disadvantages of retrospective cohort studies?
Adv
- cheap
- short-time scales
Disadv
- exposure records may be incomplete
- confounders may be unknown
How do you choose the exposed cohort?
- for a common exposure there is a wide choice, but for a rare exposure occupationally or therapeutically defined groups have been proved useful
How do you choose the unexposed cohort?
general population (national surveys / stats) and the use of SMRs and SIRs
another occupational group
a sub-cohort of the original exposed cohort
In cohort studies what sources of measurement of exposure can be used?
existing records, questionnaires to study subjects, clinical investigations
In cohort studies how can the measurement of exposure be refined?
- Dichotomous - ever / never exposed
- Categorical - no / low / medium / high
- Quantitative measure exposure (continuous variable)
What are the pros and cons of using pre-existing records?
cheap, collected before disease occurs (reduces bias) but level of the info may not be sufficient to answer specific questions
What are the pros and cons of using clinical investigations?
time consuming and invasive
prevent responder bias (could be influenced by stigma / individuals who are ill or preconceive connotations with the disease may provide answers that suit investigators)
What factors need to be in place for efficient follow up and outcome ascertainment in a cohort study?
well defined end points
minimise and record losses to follow up
follow up should be the same for all cohorts
length of follow up needed related to latency of disease
outcome measurement / diagnosis should be blinded
How are adjustments made for confounding factors in a cohort study?
achieved by use of standardisation, stratification or use of a suitable regression model (linear regression model, logistic regression model, cox regression model and poisson regression model)
What analysis should be undertaken for cohort studies?
compare mortality in exposed group to that in the general population
use SIR/SMR (standardised incidence ratios/mortality ratios)
comparison adjusted for age and sex
What are survival curves?
time taken for a disease related event to occur
- compare survival of subjects exposed and not exposed
- differences in survival may be due to chance
- large difference at the end may be based on very few patients
- statistical test is needed for comparing survival curves
What is the log-rank test used for?
test for equality of survival in 2 or more curves
What are the advantages of cohort studies?
provides estimate of absolute and cumulative risks of disease
particularly useful for following up rare exposures
eliminate some of the sources of bias which are unavoidable in case-control studies which rely on recall of past events
exposure is measured before disease occurrence - so little doubt about the cause and effect relationship
No limit to the diseases for which the risk of occurrence may be related to exposure
What are the disadvantages of cohort studies?
Not useful for very rare diseases as few of the individuals exposed will develop the disease
As they often involve following up large numbers of people over decades they are expensive and difficult to maintain
If disease is rare and we only need relative risks associated with different levels of exposure to risk factors then unnecessarily expensive
What are case-control studies?
studies in which a group of individuals with a particular disease (the cases) are retrospectively compared with an appropriately selected group of controls in relation to their previous level of exposure to the risk factor of interest
When are prevalent and incident cases used in case control studies?
If interested in determining aetiological factors then incident cases are preferable
Prevalent cases are influenced by risk factors for incidence and survival and use incident cases removes the latter
How are controls selected in case- control studies?
controls should be selected from subjects who if they had developed the disease would have become cases in the study
Should be selected from an underlying cohort from which the cases were ascertained
more difficult it is to conceive of an underlying cohort from which controls are bein selected, the more difficult it is to be sure that the selection has been unbiased
Biased here = sampling of controls has over or under represented individuals with respect to exposure than in the underlying cohort
What are some sources of controls?
From the underlying cohort from which the cases were ascertained hospital controls general practitioner controls neighbourhood controls random digit dialing by telephone number
What happens when the number of controls increases per case and what is the max?
As number of controls per case increases the power of the study increases - max number of control : cases = 4:!
What can be a useful way of matching controls to cases?
can be advantageous to match controls to cases on factors which may confound the relationship between disease and exposure
- this forces cases and controls to be identical in regards to the matched factors, which controls for confounding factors
- important to use methods of analysis which account for such matching otherwise it will be biased
What are the different forms of potential bias in case control?
selection bias - sampling bias - response bias info bias - recall bias -recording bias - interviewer bias
What can’t be measured in case control studies and what does this mean?
incidence can’t be measured and therefore you can use relative risk you have to use odd ratio
What regression model is used for adjusting for confounding factors for matched and unmatched studies?
matched - conditional logistic regression
unmatched - logistic regression
What are the advantages of case control studies?
effective use of resources to study the aetiological factors for rare diseases
Explore multiple exposures for a disease
study of disease with long latency periods
What are the disadvantages of case control studies?
it is a retrospective study - temporal relations may not be clear - which came first the cause or effect?
subject to selection and information bias
address single disease only
can’t estimate absolute risk