Quantitative Study Design Flashcards
Sampling/selection bias
sample does not represent population of interest (may colllect more extreme views)
Recall bias
inaccurate recall of past events/exposures/behaviours
Information bias
incorrect measurement eg miscalibrated machine
The Hawthorne effect
participants change their behaviour when they know they are being observed
Attrition bias
differential dropout from studies eg sicker participants drop out so our outcome is only measured on healthier participants
Experimental study design
researchers have intervened in some way (prospective)
Observational study design
the researchers have not intervened, merely observed. Can be:
• retrospective- looking back into the past
• Cross-sectional- a single snapshot of time
• Prospective- following up over time
Types of data
• categorical variables: binary, ordinal, nominal
• numeric variables: discrete, continuous
Binary
Only 2 categories
Ordinal
categories with natural order eg stage of cancer
Nominal
categories with no natural order eg blood group, ethnicity
Discrete
observations can only take certain numerical values eg number of children
Continuous
observations can take any value within a range eg age, body temperature
Restriction is precision of measurement tool
Proportion
- the number with a characteristic or outcome divided by the total number. Used to describe the probability or risk (scale 0 to 1)
Risk
probability of event occurring - probably occurring divided by total
Odds
the number with an exposure or outcome divided by the number without. The ratio of the probability of an event occurring to the probability of it not occurring- occurring divided by not occurring
Rate
incidence of health-related events or outcomes. Allows account for variation if follow-up time or time at risk of an outcome
Risk difference
absolute risk difference (subtraction)- no difference = 0
Risk ratio
relative risk- risk in one group divided by the risk in the other
Ratios
No difference = 1
Ratios > 1 indicate higher risk/odds in group of interest
Ratios < 1 indicate lower risk/odds in group of interest
Numbers needed to treat (NNT)
1 divided by absolute risk difference
• always round up- has to be a whole number
Prevalence
number of existing cases in a population at a defined time point
Incidence
number of new cases in a population over a defined time period
PICO framework
• Intervention studies: PICO
• Observational studies: PECO → ‘exposure’ rather than ‘intervention’
• Non-comparative studies (e.g. in qualitative research): PEO
• Example research question framed using PICO:
Population
In middle-aged women (>40 years old) with raised cholesterol (>5 mmol/L)
Intervention
…does new statin x
Comparator
…compared to current statin y
Outcome
…provide greater reduction in cholesterol?
(Ideally with clinically meaningful reduction in cholesterol defined)
9 Bradford Hill criteria
- Strength of association – the stronger an association, the more likely it is to be causal
- Consistency – association shown across different studies in different locations, populations, using different methods, etc.
- Specificity – specific exposure-outcome relationship, e.g. asbestos and asbestosis
- Temporality – exposure must precede outcome
- Biological gradient – dose-response, i.e. increase in exposure = increase in outcome
- Plausibility – biological mechanism that would explain outcome development
- Coherence – compatible with existing theories
- Experiment – outcome altered with experimentation, e.g. reversible
- Analogy – similar cause-effect relationships established
When designing studies must consider
• Ethics: are the rights, safety and wellbeing of participants protected?
• Feasibility: will it give the best quality evidence given our resources (time, money, personnel, etc.)?
• Efficiency: is there a faster/cheaper way to address the question without compromising quality?
Which type of study can only use odds ratio
Case-control study
Absolute risk
Probability of event within time period
Relative risk
Probability of event relative to exposure
Proportion
Outcome/ total number
Scale : 0-1
Odds
Outcome with exposure/ outcome without exposure
Probability / (1-probability)
Odds ratio
Odds that case was exposed/ odds that control was exposed
Ratio >1
Higher risk
Ratio <1
Lower risk
Risk ratio/ relative risk
Disease incidence in exposed/ disease incidence in non-exposed
NNT (numbers needed to treat)
1/ |relative difference|
Number needed to take drug for one success
NNH (number needed to harm)
1/|relative difference|
Number needed to take drug for one adverse effect
Parametric
Make distributional assumptions
95% reference range
Mean +/- 1.96*SD
What percentage of values lie within 1SD of the mean
68%
What percentage of values lie within 2SD of the mean
95%
What percentage of values lie within 3SD of the mean
99.7%
What is the y axis height of normal distribution determined by
SD variation
What is the x axis location of a normal distribution determined by
Mean
Normal distribution
Bell-shaped curve
Mean = median = mode
Report mean +/- 2 SD
Negative skewed distribution
Skewed negative (left): median> mean
left skewed distribution: same mean but median is higher
Positively skewed distribution
right skewed distribution: same mean but median is lower
Mean > median
Skewed distribution
Report median and interquartile range (centiles)
Meta-analysis
Combining systematic reviews to address a question
Ecological fallacy
Can’t make individual-level inference
What does PICO stand for
Patient
Intervention
Control
Outcome
Systematic sampling
Selected at equal intervals
Advantages and disadvantages of systematic sampling
+ve - easy
-ve - need list, large standard error, periodicity pattern
Random sampling
Sampling error can be managed
Cluster random sampling
Cluster population then sample whole clusters
Advantages and disadvantages of cluster sampling
+ve - list not needed, cheap
-ve - increases standard error
Simple random sampling
Computer generated numbers
Advantages and disadvantages of simple sampling
+ve - easy, quick
-ve - poor representation of minorities, bigger standard error than stratified
Stratified random sampling
Samples taken from each group
Advantages and disadvantages of stratified sampling
+ves - increases representation, decreases standard error
-ves - expensive, require prior population info
Types of random sampling
Systematic
Cluster
Simple
Stratified
Advantages and disadvantages of Non random sampling
+ve - convenient
-ve - sampling error can’t be measured, high potential for bias
What is bias affected by
Sample size
NOT population size
Observer bias
Researcher
Confounding variables
Relate to exposure and outcome but not on causal pathway or with intervening variable
Preventing confounding
Randomisation: generates comparable groups
Restriction: eliminates variation in confounders eg only recruiting females
Matching: control group matches confounders in case group
What is the Most appropriate study design to investigate an infectious outbreak
Case-control study
