RESS Notes Flashcards
Representative sampling
Choosing a sample representative of whole population
Exposure
Something participants are exposed to, an attribute or a behaviour
Outcome
Result being investigated
Variable
Characteristic that differs between individuals and can be measured
Dependant variable
variable that changes based on changes to another variable
ie the outcome
Independent variable
variable that causes changes in other variables
ie the exposure
Proxy variable
not normally of interest but has close correlation to variable of interest
Confounding variable
interferes with relationship between dependent and independent variable (causes outcome and exposure)
Mediator
State that occurs between exposure and outcome - no effect on outcome
Do not adjust for mediators
Competing exposure
Different exposure that causes same outcome
Nominal data
Categorical eg marital status
Ordinal data
Categorical in order ie good, very good, excellent
Discrete data
Quantitative, whole numbers
Continuous data
Quantitative, on a continuum
Data collection proforma
Questionnaire to ensure variable are consistently and accurately measured
Linear modelling
Indicates if there is an association between 2 variabls
Confidence intervals
Range of values within which researches are confident that the results of repetition of the study would fall
When applied to odds ratios, if CI includes 1 then conclusion is that exposure is not related to outcome
Confidence level
probability that repeated results will fall within confidence intervals
Number needed to treat
Average number of patients who need to be treated to prevent one additional bad outcome
Ideal = 1
Lower the better
Number needed to harm
average number of patients who need to receive treatment in order for one bad outcome to occur
Absolute risk
Difference between control group’s event rate and experimental group’s event rate
Relative risk
AKA risk ratio
ratio of probability of an event occurring in exposed group to the probability in the non exposed group
Odds ratio
Used to quantify how closely linked an exposure and an outcome are
>1 = association more likely in control group 1 = no difference to control group <1 = association less likely in control group
Adjusted odds ratio
Odds ratio after adjusting for confounding variables and competing exposures
P value
Probability of obtaining a result as extreme as the one observed, given the null hypothesis is true
P <0.05 = statistically significant
P >0.05 = not significant
Randomised control trials
Participants are randomly allocated to receive treatment or not
Can be double blind, single blind or non blind
What observational studies are there?
Cohort
Cross sectional
Case control
Ecological
Cohort study
Analysis of risk factors, follows cohort of people without disease who share characteristic over period of time
Can be prospective - will control all testing but can lose patients in follow up
Can be retrospective - immediately available data but can only examine prognostic factors
Can be time series - repeated observation of the same variables in the a cohort at designated points in time
Cross sectional
Collect data from population at particular point in time
Descriptive studies
Used to describe odds ratios, absolute risks and relative risks
Useful to analyse prevalence
Ecological study
Studies of risk-modifying factors on health defined geographically or temporally
Case control study
Two groups with different outcomes are compared on the basis of a causal attribute
Meta analysis
Contrast and combine results of multiple studies to identify patterns
Difficulties: differences in data collection + differing criteria
Requires careful data extraction, manipulation and assessment
Aetiological studies
Provide info on risk factor and causes of a condition
Can only prove correlation
Diagnostic studies
Shows specificity and sensitivity of a diagnostic test
Prognostic studies
Provide info on outcome and morbidities of patients after exposure
Therapeutic studies
Provide info on effectiveness of treatment
Pros and cons of prospective data collection
Follows patients who share a characteristic
Pros: less missing data, reliable info
Cons: time consuming, large amount of data to analyse
Pros and cons of retrospective data collection
Pros: immediately available data
Cons: no control of variables, often missing data
How can you identify confounding variables?
Draw a directed acyclic graph to show causative relationships
Describe linear modelling
Uses parsimony (prefers the simplest explanation) Mean part = the relationship Residual part = unexplained and error (normally distributed) Have to adjust results to account for effects of confounding variables and competing exposures
PECOS
Patient Exposure Comparison Outcome Study design
Controlled vocab search
Medline uses MeSH - medical subject headings
Principles of critical appraisal
Is research: Relevant Robust Objective Important
How is research robust?
Good design, no bias:
Sampling bias - random sampling, multi site
Measurement bias - info bias, observer bias, recall bias?
Analytical bias - loss to follow up, omitted variables?
Dissemination bias - publication bias?
How is research important?
If there is good quality of design (no of patients) and good strength of evidence
What is required for effectiveness of evidence?
Controlled trials Randomisation Concealment and masking Low drop out and intention-to-treat Sufficient statistical power
Describe a critical appraisal of a prognostic study
Routinely collected
Bespoke measurements
Ignore studies with >80% LTFU (lost to follow up)
Compare baseline data of LTFU with non LTFU
Conduct sensitivity analysis
Describe a critical appraisal of a diagnostic study
What is sensitivity and specificity of test? What is predictive value of positive and negative test result?
Sensitivity
Ability of test to pick up true positives
Specificity
Ability of test to avoid false positives
What is external validity?
Appropriate spectrum of patients - participants must have same range of severity, symptoms and prevalence of condition as target population
What are the key issues with evidence from an aetiological study?
External validity of participants
Exposures/ outcomes measured blind
Follow up - was sufficient time left for outcome to occur?
Loss to follow up - assess potential differential bias
