SPMM - Study Designs Flashcards
What is ecological fallacy?
Ascribing the characteristics of a group to all individuals in the group –> thus drawing subject level conclusions based on group observations
Name some study designs that belong to:
- Descriptive research
- Analytical research
- Experimental research
Descriptive (describe and do not test causal relationships)
- Case reports/series
- Cross sectional studies
- Prospective longitudinal studies
Analytical:
- Case control studies
- Cohort studies
- Ecological studies
Experimental
- controlled trials
- uncontrolled trials
Name some pros and cons of Case-Control studies?
Pros:
- Cheaper
- Less time consuming
- No exposure to new risks
- Can assess different exposures at once
- Good for rare cases where exposure may be common but chance of disease occuring following exposure is not certain
- No attrition problems
Cons:
- Difficulties with selection and recall bias
- Selecting controls is difficult
- Establishing temporality is challenging
- Incidence cannot be assessed - odds ratio provided rather than relative risk
- Cannot prove causality
Name some important features to adhere to for a well designed case-control study?
- Strict diagnostic criteria to identify cases
- Appropriately matched controls (for sex, age and other characteristics). Can be done group wise/pair wise. Note overmatching may actually reduce risk by controlling for factors related to the exposure/causal factors. Overmatching also makes it difficult to recruit controls
- Not necessary to increases ratio from 4 controls : 1 case - increases above this do not increase statistical power
When may a cohort study be appropriate? (in terms of exposure frequency & chance of disease after)
For rarer exposures but ones where chance of disease after is high
What does an internal control mean in a cohort study?
Where control is still exposed but exposure is under a threshold dose (i.e. 1 pack a day vs. 2 packs a day)
Can cohort studies calculate relative risk?
Yes
What are the pros and cons of cohort studies?
Pros:
- Less selection and recall bias compared to case-controls
- Good for studying multiple effects of one exposure
- Temporality can be established more accurately
- Better assessment of causal relationship than case-controls
- Can calculate incidence and relative risk
- Dose response relationship of exposure can be assessed
- Can assess natural disease progression
Cons:
- Expensive and time consuming
- Not good for rare diseases (need to recruit significant numbers)
- Cannot calculate incidence
- Drop out and attrition
- Can only assess one etiological factor at a time
What is a nested cohort study?
One which cases and controls are recruited from within one existing defined cohort (i.e. blood results)
They are cheaper and easier conduct but can still be precise
Describe verification (work up bias)
A type of bias whereby the decision to perform a gold-standard diagnostic (reference) test is influenced by the results of the initial assessed test
Commonly negative results of the initial test may mean the gold-standard test is not performed (Negative d-dimer - no CTPA)
Thereby more false negatives may not be detected
For biological efficacy what analysis may be best conducted?
Per protocol analysis - to see the biological efficacy need to analyse those who took the drug
Note this does not relate to clinical efficacy where ITT analysis would be best conducted
What do secondary outcomes test?
An additional hypothesis
What is a minimisation allocation scheme?
Matching candidates based on the criteria of those already allocated
What is best case/worst case scenario?
Sensitivity analysis conducted for when patients are lost to follow up
Firstly all patients who are lost are assumed to have good outcomes (best case) then assumed to have bad outcomes (worst case). Each analysis is run - if positive results persist after worst case we can be reassured trial results favoured
Attrition bias is when those lost to follow up differ from those left in the trial (migration bias also termed)
Outline the different types of randomisation?
- Simple - i.e. computer generated list
- Block - randomisation occurs in groups of certain number (larger the block size the harder to predict assignment)
- Stratified - baseline characteristics are stratified to ensure an equal spread in intervention/control groups. Larger the sample size the less gain in precision through stratification
- Cluster - units of randomisation are treatment centres/catchment zones. As patients in each cluster are not independent of each and thus more likely to respond to a treatment there is loss of statistical power. To try and resolve this effective sample size is used > actual sample size. This is calculated using the degree of similarity between participants in each cluster on an outcome measure.
Quasi-randomisation - date of birth, alternate numbers etc
When is minimisation used?
- Wanting to do stratified randomisation but small sample size and worried about mismatch of confounders across treatment groups
- Cluster randomisation
Minimisation is not true randomisation - computerised schemes can reduce bias
Why is randomisation used?
- Allows us to use probability theory
- Allows blinding
- Random distribution of baseline characteristics
- Eliminates effect bias
Name some features of sound randomisation
- Future sequences not predictable from past
- Reproducible
- Allocation is concealed
- Can detect departures from documented procedures
- Can document assignment processes
What is a patient preference trial?
Patients get to choose which group randomised to, if no preference they get randomised. Analysis should be limited to those randomised however data obtained from both wider observational groups. Can analyse the impact of motivation for a certain treatment on outcome
Can help recruitment however complex to conduct the trial
Describe when informed consent is done in Zelen’s modified RCT?
Following randomisation - normally consent done before for ethical reasons however participants may drop out if they feel they are not going to get their preferred treatment.
MRC does not accept Zelen modified RCT designs
Are larger or smaller sample sizes required for non-inferiority RCTs?
Larger - since the alternative hypothesis is no difference between groups, effect size of 0 (in theory sample should be infinity).
In these trials it is best to report per-protocol analysis and intention to treat as ITT may blur actual differences between groups
What do pragmatic RCTs assess for?
Effectiveness of a drug - how it works in the real world (not efficacy how it works in ideal circumstances)
Name some features of pragmatic RCTs?
(1) Select clinically relevant alternative interventions to compare,
(2) Include a diverse population of study participants,
(3) Recruit participants from heterogeneous practice settings
(4) Collect data on a broad range of health outcomes.
(5) They do not tell how well a drug works (EFFICACY) – they inform how EFFECTIVE is the intervention in the real world.
(6) Recruit based on presentation > diagnosis
Name the 6 conditions of causality as proposed by Bradford and Hill
- Temporal association
- Dose response relationship
- Specificity of association - of order
- Consistency of association
- Biological plausibility
- Strength of association
- Absence of reverse causality
Which is the most cost effective quadrant on an incremental cost effectiveness plane?
Southeast
