Causality Flashcards
What are kochs postulates?
- agent must be necessary (in everything)
- agent must be specific (in every case)
- agent must be sufficient (capable of inducing disease)
What are the two types of bias?
- selection (not representative population- external validity, cannot compare- internal)
- information bias (recall, observation, measurement, classification)
Give 5 of bradford hills criteria for establishing causality?
- strength of association (more strong= more likely)
- specificity of association (more likely if outcome only associated with specific factor)
- consistency of association (more likely if link seen in all groups& studies)
- temporal sequence (exposure must come before outcome)
- dose exposure (more likely if high exposure= higher risk)
- reversibility (removal or prevention of exposure leads to stopping outcome)
- coherence of theory (more likely if conforms w/ current knowledge)
- biological plasuability
- analogy
What is a systematic review?
An overview and amalgamation of primary studies using a clear and reproducible method. They have clear focussed question, make explicit statements about types of studies, types of participants, types of inteventions, outcome measure etc. They may include metaanalysis
What is a meta analysis?
a quantitative synthesis of results of two or more primary studies that address the same hypothesis in the same way.
What is a decision analysis?
comparison of harms and benefits and cost effecitveness
How are meta analysis done?
The CI and odds ratios for all studies are calculated and combined to give a pooled estimate of odds ratios using s statistical computer program. The weighting of the odds ratio depends on the size of the study and the uncertainty of their odds ratio (IE if their CI is large)
What is the fixed effect model?
Assumes all studies are estimating the same truth and differences from the overall mean are random chance.
What is the randoms effect model and when is it generally used?
It assumes all studies are measuring a similar but not exactly the same truth- one study may be finding truth for luton and another for london. It is generally used when there is great variation between studies, and makes studies weighted more evenly. Usually both fixed effects and random effects are used.
What are the two types of subgroup analysis?
- stratification by study characteristics (length of follow up, recruitment criteria etc, to see how similar the studies are)
- stratification by participant profile (types of participants- age, sex, wealth etc)
What is assesed when analysing quality of studies?
- study design
- design protocol
- protocol implementation
- type of trial (RCT best)
- allocation methods
- blinding and outcome assement
- pt attrition and intention to treat
- statistical analysis
What is publication bais?
studies with significant or favourable results are more likely to be published so systematic reviews less likely to include unfavourable or insignificant results.
how can publication bias be assesed?
Funnel plots- plot standard error against odds ratio and studies should look like funnel (symmetrical)- this will mean youre not excluding small studies proving little effect or large studies proving little effect
Which stages of drug development are clinical trials conducted? What is their purpose?
Stage 3- used to determine efficacy and saftey
Whats the disadvantage of not randomising a clinicial trial?
subject to allocation bias and confouding
Whats the disadvantage of comparing new study results with historical controls?
The historical controls are less rigorously selected- more comorbidities, they’re also treated slightly differently to eachother and less information is available about potential confounders.
Briefly outline how a clinical trial is conducted?
- define factors: disease of interest, outcome measures, confounders, pts eligible, controlled variables
- conduct trial: find pts, consent them, randomly allocate them to treatment groups, follow them up in identical ways minimise follow up losses and maximise compliance
- compare outcomes: are differences present, statistically significant and clinically significant?
Give features of an ideal outcome measure?
- appropriate and relevant
- valid and attributable
- sensitive and specific
- reliable and robust (reproducible)
- simple, cheap, sustainable, timely
- ability to measure before, during and after
Whats the difference between a primary and secondary outcome measure?
primary: outcome looking at changing with new treatment
Secondary: other effects of treatment, often side effects or unforeseen effects
Why is blinding good for a trial?
- reduced behaviour effect: pt alteres behaviour or expectations of outcome due to knowledge of treatment group
- reduced non treatment effect: clinicians alter treatment, care, interest
- reduced measurement bias
Why can blinding be difficult?
- where one intervention has clear and distinguishable side effects
- where one intervention is inherently different (eg pill vs surgery etc)
- where comparing against lifestyle interventions
- where one intervention is aimed around prevention
What is the placebo effect?
Where the pts attitude to illness or even illness itself improves as a result of the feeling that something is being done, even if the therapy is irrelevant.
When should a placebo be used?
When there is no standard treatment available, otherwise use a control group
How can follow up losses be minimised?
make follow up easy, be honest about commitment required at start, avoid coercion and inducements, maintain regular contact
How can compliance be maximised?
simplify instructions, ask about compliance regularly, ask about effects and side effects, monitor tablet count, test them using urine and blood samples
What is an explanatory/ as treated analysis?
Compares only those who completed follow up and complied with treatment so compares physiological effect of treatment well. But subject to bias and confouding.
What is intention to treat analysis?
analyses outcomes from everyone who took part in trial, trying to get more pragmatic results
What is ethical issues should be considered when conducting clinical trial?
- clinical equipose: where youre really not sure which intervention is better
- scientifically robust: valid question, acceptable risk, important issue
- ethical recruitment: dont inc communities who wont benefit, high risk ppl, but dont exclude non ideal candidates (elderly)
- valid consent: informed of risk, pull out whenever, verbal and written
- voluntariness: participants must be free of perception of coercion or manipulation that lead them to entering a trial.
