Epidemiological Research Flashcards
Detail some key considerations required when thinking about sampling
- Samples must be representative in order to be able to extrapolate findings to a wider population
- A representative sample is generally large, randomised and accounts for minorities and hard to reach groups
- Sample size: determined by the size of the effect, outcome of interest, level of observation/intervention
Samples should have:
- Statistical power: probability of detecting if an effect is real (80-90% power)
- Statistical precision: probability of detecting is an effect is not real (p-value 5%)
Name different types of sampling methods
- Simple random sampling: random selection from frame. Equal chance of selection.
- Systematic sampling: Every nth (regular intervals)
- Stratified sampling: groups and strata, equal proportions using simple random sampling
- Cluster sampling: uses hierarchical structure and samples all those within the structure
- Multi-stage sampling: hierarchical structure with more stages. Proportional.
Name the key types of data collection methods
Direct - questionnaires, interviews, clinical examinations
Advantage: can be performed prospectively and tailored
Limitation: costly and time-consuming, re-call bias, non-response
Indirect - medical records, census data, health surveys, registries, school/employment records
Advantage: readily available, easy access, no cost
Limitation: data may be missing or inaccurate
Describe the different types of data variables
Quantitive - numerical
- Discrete
- Continuous
Qualitative - Categorical
- Binary
- Ordered categorical
- Non-ordered categorical
- Binary or categorical data presented as a proportion (%)
- Descriptive = Bar chart or histogram
- Continuous variables represented as their average and variation (frequency distributions: normal, positive skew, negative skew)
- Analytical = scatterplots to show correlation
Explain the correlation coefficient
Used to demonstrate the strength of relationship variables as per trend lines
- Direct Relationship - r is positive
- Both variables increase together
- Indirect Relationship - r is negative
- One variable increases as the other decreases
- No Correlation - r = 0
- Variables are not associated
- Closer to 0 = a weaker correlation (conversely distance from 0 = stronger correlation)
Describe standardisation and its use
- Comparing crude descriptive data is inaccurate due to differences in population, time, place, confounding and modifiers
- Standardisation methods equals the structure of two populations with respect to an outcome modifying factor for more accurate comparison eg. age, gender, mortality
Describe the process of direct standardisation
- Used when strata-specific outcome frequency is known in the populations to be compared
- Standardised FREQUENCY = Frequencies are applied to a standard population to calculate a weighted average frequency for each population
- Not a real value but allows for comparison
- Multiply strata-specific frequency by percentage of standard population in the stratum to get weighted measure
- Sum all start weighted measures
- Divide by 100 to get standardised rate (eg. incidence rate) for comparison
Describe the process of indirect standardisation
- Used when strata-specific data is unavailable but population and total number of cases is known
- Standardised RATIO = strata-specific frequencies from a comparison population are applied to population of interest to calculate expected cases, observed outcomes and expected outcome frequencies and compared
- Multiply percentage of population in each stratum by the comparison population strata-specific frequency
- Small all strata = total expected cases
- Calculate SR = observed cases/expected cases
Describe the difference between ‘intention-to-treat’ and ‘per-protocol’ analysis
Intention-to-treat
- Outcome is compared between study subjects based on allocation regardless of withdrawal, loss or non-compliance
- Ensures comparability between intervention and comparison arms is maintained (bias and confounding remains minimised)
- Best to understand EFFECTIVENESS in real world conditions
Per-protocol
- Secondary analysis that only includes study participants who did receive intervention
- Subject to bias and confounding
- Best to understand EFFICACY of intervention under ideal conditions
Describe the common patterns of case distributions for different disease outbreaks
- Common source at one point (eg. food poisoning)
- Common intermittent source (eg. toxic waste)
- Common continuous source (eg. water contamination)
- Propagated epidemic (eg. infectious agent)
