1 research Flashcards
(48 cards)
routine data
Routine data describes non-targeted information that is obtained in a standardised and consistent manner.
strengths and weakness of routine data
epidemiological trends
Time: secular, periodic, epidemic
person
place
DALY Vs QALY
DALY: the LOSS of the equivalent of one year of full health.
sum of the years of life lost to due to premature mortality (YLLs) and the years lived with a disability (YLDs)
QALY: one year of life GAIN in perfect health
Random error
Chance differences in the true and recorded values may result in an apparent association between an exposure and an outcome, and such variations may arise fro:
- unbiased measurement errors
- biological variation within an individual
Bias
any systematic error in an epidemiological study that results in an incorrect estimate of the true effect of an exposure on the outcome of interest
Either: Selection and information bias
Misclassification (Information bias)
classification of an individual, a value or an attribute into a category other than that to which it should be assigned.
differential or non-differential.
non diff fine as tends to null.
Types of information bias (differential)
- observer bias
- interview bias
- recall bias
- social desirability bias
- performance bias
- detection bias
selection bias
- loss to follow up
- sampling bias
- allocation bias
- attrition bias
- healthy worker effect (occupational cohort studies)
control bias (case-controls)
confounding control
- design:
- randomisation, matching, restriction - analysis
- stratification (mantel Haenszal)
- multivariate analysis
- standardisation
reduce bias
- randomisation
- blinding
- training
- protocol
- ease of follow up
- high risk cohort
- duplication
confounder
Independently associated with exposure and outcome but not on causal pathway
residual confounding
residual confounding: unknown confounding left after taking into account known
should be equal if RANDOMISATION done
Descriptive studies
example
pros/ cons
ecological fallacy
case studies and ecological studies
ecological studies: population level
- aggregate, environmental, global
- geographical and time series
Good:
rapid, low-cost, if only aggregate data available, exposures in different areas, spatial framework
bad:
control: selection
no causation, publication bias, confounding control, recording difference
ECOLOGICAL FALLACY
effect on population level not seen at individual level
- no possible to link exposure to individual
- data collected for another reason
- average exposure not linear
- confounding control
example: new migrants states with high literacy: individual migrants lower literacy
study designs
Observational: cross section, case-control, cohort
interventional: RCT, non randomised trial
cross sectional
collect exposure and disease outcomes at the same time
- descriptive, analytical (or ecological level)
prevalence and odds
pros:
Can study multiple exposures and outcomes. Rapid and cheap.
Useful for rare diseases.
Useful for detecting disease burden.
cons:
Because cross-sectional studies measure prevalence, not incidence, the findings cannot differentiate between the determinants of aetiology and survival.
difficult to determine if an outcome or an exposure came first, because both were assessed simultaneously (i.e. risk of reverse causation).
May be subject to recall bias
cross-sectional ecological study: the study by Drain et al. [4] that compared HIV prevalence and rates of male circumcision in 118 developing countries and found that HIV prevalence was lower in countries where male circumcision
Cohort
looks at association between exposure and outcome by following a group of exposed individuals over a period of time (often years) to see whether they develop the disease or outcome of interest.
prospective: follow until get outcome or until time limit reached
retrospective: exposure and outcome have already occurred at the start of the study. Pre-existing data, such as medical notes, can be used to assess any causal links, so lengthy follow-up is not required.
pros:
- prospective (temporal)
- multiple effects
- rare exposure
- long latency
cons;
- time consuming
- expensive
- rare disease
- recall bias
nested case-control: can be within a cohort
examples:
- farringham cohort study:
variations of RCT
Parallel (traditional)
crossover (person their own control, will need to ensure Rx washout)
factorial design
wo or more interventions are compared singly and in combination against a
comparison group (i.e. there may be four groups: intervention A, intervention B,
interventions A and B, and control).
This design allows the investigator to study
eg PEACE study 2x2 facotrial for prostate cancer (4 groups)
cluster:
eg ochomo et al: spatial malaria nets in Kenya
types of randomisation
systematic allocation
simple randomisation (random number generated)
block randomisation (AABB)
stratified (separate age, sex then randomise)
stepped wedge (pop divided then gradually introduce in random)
small area analysis
Small-area analysis (SAA) permits the examination of data for groups, such as towns, which tend to be more homogenous in character compared with larger populations that are likely to be more diverse.
- prevalence might be different
- better local knowledge
- support decision making
issues:
little variation
chance
limited data
eg OHID fingertips,
NWL WISC
(instrument) validity
- Criterion
- concurrent: how well it compares to a gold standard
- predictive: predict likelihood to have a disease - Face: expert opinion
- Construct
- extent to which the instrument specifically measures what it is intended to measure, and avoids measuring other things. For example, a measure of intelligence should only assess factors relevant to intelligence and not, for instance, whether someone is a hard worker. - Content
- is systematically and comprehensively representative of the trait it is measuring. For example, a questionnaire aiming to score anxiety should include questions aimed at a broad range of features of anxiety.
reliability
inter rate /inter observer reliability : same subject 2 observers
intra rate/ intra observer reliability:
same observer same subject
test-retest reliability
under the same conditions and in the same test population
equivalence/inter method reliability: 2 instruments measure the same thing (equivalence reliability coefficient)
internal consistency: This is the degree of agreement, or consistency, between different parts of a single instrument.
FOR internal consistency
(Cronbach’s alpha: a statistic derived from pairwise correlations between items that should produce similar results. The usual range for the alpha will be zero to one, with values above 0.7 generally deemed acceptable)
inter relator reliability
Kappa statistic
- independent
- doesnt say why variation
Kappa indicates how well two sets of (categorical) measurements compare.
Kappa values range from -1 to 1, where values ≤0 indicate no agreement other than that which would be expected by chance, and 1 is perfect agreement. Values above 0.6 are generally deemed to represent moderate agreement.
Statistical implications of clustered data
clustered data are more similar than each other –> loss of independent
work out ICC (intra-cluster correlation coefficient)
= p (rho)
between-cluster variability divided by the sum of the within-cluster and between-cluster variabilities.
iIf ρ = 1, all responses within a cluster are identical and the effective sample size is reduced to the number of clusters rather than the number of individuals
If ρ = 0, there is no correlation of responses within a cluster, and individuals within and amongst the group are independent with respect to that variable
As the ICC increases, the sample size required to detect a significant difference for the variable under investigation increases.
design effect
DE = 1+(n-1)ρ
The DE can then be used to calculate the ‘effective sample size’. This is the ‘real’ sample size in a clustered trial, compared with the number of participants actually enrolled in the study.
ANOVA:
- see ss difference between clusters
- random and fixed
