RESS 3 Flashcards
Why do we study healthcare practice?
- Provide better quality evidence rather than guesswork
2. Primary + secondary studies allow results to generate guidelines, standards and targets
Study types used in healthcare development and assessment
- Research
- Audit
- Service Evaluation
Wha does research allow us to do?
Generate new knowledge where there is no or limited research evidence available and which has the potential to be generalisable of transferable
What is an audit?
A quality improvement process that seeks to improve patient care and outcomes through systematic review of care against explicit criteria and the implementation of change
What is a service evaluation?
Evaluates a proposed service or current practice with the intention of generating information to inform local decision making
What method can be used to define the search terms used in research lit. searches?
PECOS
Patient Exposure Comparison Outcome Study design
What are the competing priorities which clinical studies require us to balance?
- design and conduct studies that provide the correct answer
- maximise the efficient use of resources
- comply with ethical and legal regulations
What are the 4 potential sources of bias?
Sampling/selection bias
Measurement bias
Analytical bias
Dissemination bias
How do each of the following study designs help address the issue of bias?
- cross sectional
- case control
- cohort
- trial
- meta analysis
- cross sectional: provides evidence of association within a sample
- case control: provides evidence of association between samples
- cohort: provides evidence of directionality of associations
- trial: provides evidence of causality
- meta analysis: provides evidence of reproducibility/generalisability
What 3 ethical concepts came from the Belmont Report’s
Respect for persons
Justice
Beneficence
What types of projects do not require formal ethical approval?
- Secondary research
- Non human research
- Audit/Service evaluation
- Service evaluation (collects info on existing service)
What types of projects do require formal ethical approval?
- Non human subjects/animals
- Vulnerable groups
- Experimental research
- Non experimental research where information on more than only existing service delivery is required
What is a sample?
A collection of data drawn from a population
What is a target population?
The total finite population we wish to know about from which your sample is drawn
What is the study sample?
The units/participants drawn from the target population that constitute our data set
What are the types of samples?
Complete: all
Unstratified Random: every member of the target population has the same chance of being sampled
Stratified Random: randomly sample from target pop.
Pros and cons of complete sampling
Pro: no bias introduced by design
Con: potentially expensive
Pros and cons of unstratified sampling
Pro: easy to design and conduct
Cons: Smaller groups may be under represented by chance
Pros and cons of stratified sampling
Pro: representative of population
Pro: unequal sampling of strata improves power for rare strata
Con: population may not be easily divisible
Con: strata may now be known until after sampling
Why do we statistically analyse data?
Estimation
Hypothesis testing
What happens to the confidence interval if the sample size is increased?
It decreases and becomes more reliable
What does hypothesis testing tell you?
How unlikely you are to see an effect this big by chance is there is no genuine effect
When is the association significant?
When the CI does not include the null at p=0.05.
What is power?
The probability of rejecting the null hypothesis when the null hypothesis is false
When is the power greater?
- the mean effect is bigger
- variation in effect is smaller
- sample is bigger (the only one that can be influenced
When should power be calculated?
Before a study to ensure it is well designed
After a study when no association is found - to demonstrate if the study was sufficiently powered
What is the ‘odds of an event’
= the probability an event occurs / probability that it doesn’t occur
What is the odds ratio?
The odds of an event for exposed / odds of an event for unexposed
Which of confounders, competing exposures and mediators, are not adjusted for?
Mediators
What are confounders
They cause both the outcome and exposure
Why must confounders be adjusted for?
They will generate a statistical relationship between the outcome and exposure even when none exist
What are mediators?
They cause the outcome and are caused by the exposure
Whey are mediators NOT adjusted for?
They are part of the causal path between O and E so should not be adjusted
What are competing exposures?
They cause the outcome but have no relationship with the exposure
When would a competing exposure be adjusted for?
If they cause a substantial amount of the variation in the exposure
A causal relationship can be:
- functional
- emperical
- theoretical
- speculative
When is a causal relationship possible
Only when the cause variable precedes the effect
What do prospective studies do?
Record variables over the study period, with the outcome measured subsequently
What do retrospective studies do?
Measure the outcome and then look backwards to measure the exposure and other variables
Pros and cons of prospective data collection
Pros:
- fewer sources of bias
- less chance of confounding
- some data can only be measured prospectively
Cons:
- time and resource intensive
- tendency to collect data on more variables than you need or can use in your analyses
- subjects can drop out
- usually unfeasible for rare outcomes
Pros and cons of retrospective data collection
Pros:
- data collection is less time and resource intensive
- allows oversampling of rare outcomes
Cons:
- more susceptible to bias
- some variables cannot be measured directly
- if data are from records, little control over these
What are some sources of measurement error?
Instability in the variable
Imperfect measurement
What allows for a better line of best fit
Less residual variation
What does linear regression allow?
Comparison of two variables
What is R squared?
Proportion of variation explained by the linear model. Values closer to 1 = more of the variation is explained by the model
What can R squared not calculate for?
Logistic regression
What do the results of the odds ratio mean?
>1 = control is better than intervention <1 = intervention is better than control
