Flashcards in HaDPop Deck (93):
What is a census?
A simultaneous recording of demographic data by the govt at a particular time, pertaining to all the people living in a particular territory.
What are the main features of a census?
Run by government
Occurs at regular intervals
What are censuses used for?
1). Allow allocation of resources
2). So we can make projections on populations
3). Work out trends in populations
What is crude birth rate?
The number of live births per 1000 of the population.
What does crude birth rate give us?
It gives us the impact of births on the population size.
What is general fertility rate?
The number of live births per 1000 females of ages 15-44.
What does GFR compare and how is it affected?
It compares fertility of fertile female populations and is affected by age distribution.
What is total [period] fertility rate?
The average number of children that would be born to a hypothetical woman in her lifetime.
What does TPFR compare and how?
It compares the fertile females fertility without being influenced by age structure, by giving each age group equal weighting in its calculation.
What is the calculation for TPFR?
TPFR = sum of (all age specific fertility rates)
e.g. If 24 15yr olds have 6 babies = 6/24 = 0.25
What is fecundity and what affects it?
The ability of a population to reproduce.
Affected by sterilisations and hysterectomies.
What is fertility and what affects it?
It is the recognition of fecundity as a potential for live births.
Affected by sexual activity, the economic climate, abortion and availability of contraception.
What is crude death rate?
The number of deaths per 1000 of the population.
What is age specific death rate?
The number of deaths per 1000 in a particular age group.
What is the standardised mortality rate calculation?
Number of observed deaths in study pop
--------------------------------------------- x 100
Number of expected deaths in study pop
What is standardised mortality ratio and what is its purpose?
A comparison of the observed deaths with the number of deaths expected if 2 populations had the same age-sex distribution.
It adjusts for age-sex confounding by comparing a population with a standard reference.
How do you calculate person years?
It is the sum of total years exposed per person. e.g. 1 person for 10 years, 3 people for 0.1 years
= (1x10) + (3x0.1)
What is a population estimate?
Applying birth rates, death rates and migration to the present.
What is a population projection?
Applying birth rates, death rates and migration to make predictions about the future along with additional assumptions.
What is incidence and its units?
Units are rate
Incidence is the amount of new cases of a disease in a population in a time.
How do you calculate incidence rate?
IR = ------------------------------------
persons x time (years)
What are incidence measurements useful for?
Giving a measure of the populations average risk of disease (although people vary)
What is prevalence and its units?
The number of existing cases in a population
~ (incidence x length of disease)
Unit is a proportion (no. cases/population) = no time element.
When can variation aid a study?
1). Systematic variation can give us clues about the cause of a disease e.g. Exposure levels between 2 groups can be used in studies.
2). IRR allows us to compare groups.
How do you calculate IRR and what is it useful for?
Rate of exposed
IRR = -----------------------
Rate of unexposed
It allows us to compare incidence of disease between 2 groups or to compare the efficacy of two treatments.
Why does IRR only give us relative risk?
The value it gives us is comparison between 2 groups only (risk of developing the disease n exposed group compared to the risk in the 2nd group, which I selected). This means it isn't taking into account the general population, and the 2 groups may be systematically different from these.
Why does IR provide absolute risk?
Because it compares the number of new events to the general population
When is variation a nuisance?
Confounding can explain all/part of an apparent association.
How do you eliminate age-sex confounding?
Age-sex confounding always needs to be eliminated.
Use the SMR (compares number of observed deaths to a reference population like the one being studied)
Could use age specific rates but this would be very time consuming as you would obtain many different values.
When is a p value statistically significant?
What are the limitations of the p value?
1). The values p=0.049 and 0.051 are very close together = is there really that much difference?
2). The statistical significance needs to have a big enough sample size to start with or chance can't be down to 0.05 (e.g. Flipping a coin).
3). Statistical significance does not necessarily mean clinical importance.
What do confidence intervals give us?
The range in which we can be 95% certain our true value lies, allowing us to make rational inferences despite random variation.
How do you calculate confidence intervals?
1). Calculate the observed value (IRR, OR, IR).
2). Calculate the error factor.
3). LCL = observed value/error factor
UCL = observed value x error factor
What are the two types of case control study?
1). Conventional - collect data retrospectively
2). Nested - case control study within a cohort study
--> take people with/without the disease from the cohort and ask about their exposures, side effects etc.
What are the advantages of nested case controls?
1). More detail about the participants e.g. Side effects
2). Allows calculation of an IR still as it is within a cohort study so is based on a defined population (case control alone is not).
What is the calculation for the IRR in a common disease of a case control study if...
Exposed diseased = A
Exposed undiseased = B
Unexposed disease = C
Unexposed undiseased = D
= rate of exposed incidence/rate of unexposed incidence.
What is the calculation for the IRR or OR in a rare disease of a case control study if...
Exposed diseased = A
Exposed undiseased = B
Unexposed disease = C
Unexposed undiseased = D
What is the error factor for case control studies?
[2x(root of 1/a + 1/b + 1/c + 1/d)]
When is the OR used in case control studies?
In the rare disease assumption.
What are the three types of bias in case control studies?
1). Information bias
2). Selection bias
What is selection bias?
Bias in how controls are selected.
- if controls become a case, must be classed as one.
- healthy worker effect
How can you minimise confounding in case control studies?
1). Match cases to controls by important confounders.
2). Adjust for it using logistic regression.
What are the 2 types of information bias in case control studies and what do they cause?
1). Non differentiated misclassification (recall bias)
---> inaccurate measurement often when people forget if they were exposed.
---> causes shrinkage to the null (OR moves towards 0)
2). Systematic bias
---> where the cases have different assessor bias, data collection methods etc
---> those unexposed think they were exposed
---> movement to or from the null depending on if the diseased or undiseased got it wrong.
3). Publication bias
---> when ones that don't prove something aren't published
What is information bias in a case control study estimated at?
What is a prospective or retrospective case control study?
Prospective - the study can be designed so that controls are also monitored for any disease arising and side effects etc are checked.
Retrospective - the usual - people are selected and asked about exposure.
What is a prospective or retrospective cohort study?
Prospective - start with disease free individuals and follow up for a period of time.
Retrospective - recruit past disease free individuals and classify on their exposure status and subsequent disease status.
What is an internal or external cohort study, which calculations are used and how does cohort size affect them?
- follow 2 groups ourselves
- if either sub cohort is small, e.f. increases
- IRR is used
-follow 1 group and compare to a reference population
- e.f. is based on SMR (which is one value rather than 2 in IRR) so isn't affected by size.
What are the differences between prospective and retrospective studies?
Prospective = slower, less unknown confounding (can measure outcome and confounders)
What are the advantages and disadvantages of external comparison?
+ works better in small groups
- more prone to selection bias if groups aren't comparable
- limited data for reference population
What are e advantages and disadvantages of cohort studies?
+ good for conditions that change with age
+ good for rare exposures
+ measures multiple outcomes
+ establishes temporal sequence
- survivor bias
- takes a long time
- large/resource intensive
- falls to unknown confounding
- results can be influenced by ethics or politics
- no good for rare diseases
How is SMR calculated in external cohort studies?
1). Obtain the age specific incidence rates for reference population for each time period
2). Multiply rates by correct cells person years for cohort = expected deaths
3). Calculate SMR
How do you account for age change when calculating the SMR in an external cohort study?
By plotting a lexis diagram.
What are the 3 points a RCT must be?
1). Fair - free from bias or confounding
2). Controlled - comparison of interventions
What are the advantages and disadvantages?
+ random allocation = little confounding or bias
- goes against collective ethics (all patients should have safe, efficient treatment)
- goes against individual ethics (principles of justice, non malice, benefice and autonomy)
Briefly, how is a RCT conducted?
1). Define factors - disease, outcome, treatment
2). Identify, invite and gain consent from participants
3). Randomly allocate treatment
4). Follow up - maximise compliance and minimise drop out
5). Observe outcomes
What are the types of outcome and examples of these?
1). Clinical - death
2). Pathophysiological - tumour size
3). Patient centres - quality of life
What a primary and secondary outcomes?
Primary = the one being measured in sample size calculation
Secondary = any other interests e.g. side effects.
What are the features of an ideal outcome?
Specific for accurate detection of differences
What can differences in treatment in an RCT be due to?
2). Patient, clinician or assessor bias
3). One is better than the other
What is blinding and what does it avoid?
Single, double or triple.
The patient/clinician/assessor does not know what treatment has been assigned to who.
This avoids the placebo effect.
What are the features of a placebo and under what conditions can it be used?
They are inert and identical in EVERY WAY to the real drug.
The patient must consent to knowing they may not receive a real drug, and a placebo can only be used when no standard treatment available.
What is an RCT?
A planned experiment that tries to elucidate the most appropriate method of treatment for future patients for a specific disease.
What are the two types of trial analysis in RCTs?
1). Pragmatic trial - intention to treat
2). Explanatory trial - as treated
What is a pragmatic trial?
Intention to treat
Includes everybody who began the trial = preserves randomisation.
Gives a better indication of how effective the treatment would work in a clinical setting.
What is an explAnAtory trial?
As treated analysis
Includes only those who have complied with the treatment and stayed in the trial.
Gives a better indication of the physiological effect of the drug
Loses effect of randomisation
Why do some not comply in an RCT and how do we maximise compliance?
May not comply if:
They don't understand
They don't like treatment
They feel no benefit/prefer another treatment
Maximise compliance by:
Making it simple
Asking about side effects and compliance
How do we minimise losses to follow up?
1). No coercion into being a participant
2). Be honest about commitment etc
3). Maintain contact
4). Make follow up simple
What are the 5 ethical points to be considered in an RCT?
1). Clinical equipoise
- there must be reasonable uncertainty/genuine ignorance as to which treatment is best
2). Scientifically robust
- safety must be monitored and the trial be of appropriate study design and question.
3). Ethical recruitment
- those included in the trial must not be at high risk of harm and must benefit from the trial in future.
4). Valid consent
- must be authorised by a knowledgeable informant
- consent must be given and a cooling off period allowed.
- no coercion or manipulation into involvement
What are the Bradford hill criteria linked to the association?
1). Strength of association-
Stronger the association, the more likely its a cause.
2). Specificity of association -
If the outcome is associated with only one specific factor, the more likely its a cause.
3). Consistency of association-
If the association occurs in multiple similar studies, the more likely its a cause.
Why may there be inconsistency between association of outcome and exposure in similar studies?
The exposure is not a cause
Study design is poor
Outcome may have multiple causes
What are the Bradford hill criteria about the link between exposure and outcome?
1). Temporal sequence -
If the exposure always precedes the outcome, the more likely its a cause.
2). Reversibility -
If removing the exposure reduces the risk of acquiring the outcome, the more likely its a cause.
3). Dose response -
If an increase in exposure increases risk of acquiring the disease, the more likely its a cause.
What are the Bradford hill criteria referring to other evidence?
1). Coherence of theory -
If the observed association conforms with current knowledge, the more likely its a cause.
2). Analogy -
If the disease is like others, the more likely its a cause.
3). Biological plausibility -
If a biological mechanism is demonstrated, the more likely its a cause.
How can coherence of theory and biological plausibility be limited?
Biological plausibility may be limited by current knowledge.
Coherence of theory may lead to inappropriate rejection of unfavourable association = publication bias.
What are the explanations for association?
2). Selection bias
--> people selected are unrepresentative of the population
3). Information bias
4). Alternate associations
--> reverse causality
--> common cause between the associated things
What are the henle-kochs postulates?
1). Agent is necessary - present in all cases of the disease
2). Agent is specific - present in cases of that disease only
3). Agent is sufficient - alone can cause the disease
What is the purpose of meta-analysis?
Systematically collate study results
Quantify effect sizes and their uncertainty as a pooled estimate
Facilitate synthesis of a large number of study results
Aid interpretation of results with variation in sampling.
What do meta-analyses specify?
How set of studies is collected
Common categories/variables the study defined
Standardised data extraction used
How sources of variation were allowed for
How do you interpret a meta-analysis?
Dotted line = pooled OR
Diamond = pooled CL
Line = null value
Size of study's square = size of error factor
What would you talk about when nterpreting meta-analysis results?
1). Overall trend
2). How many studies are statistically significant
3). Conclusion from the pooled estimate
What are the 3 problems with meta-analyses?
1). Heterogeneity - no 2 studies are ever the exact same.
2). Publication bias - not involving unfavourable studies in the results
3). Variation in study quality
What could variable quality of study design be due to and how can we solve this?
Poor study design or poor protocol implementation (= bias).
Solve by having a set standard which the studies must meet before being involved in the meta-analyses, or scoring each study on its quality, and incorporating this into the study's weighting.
How can you identify publication bias?
Funnel plot - size of study plotted against the result. If there's a lack of studies on the bottom left hand side, publication bias has probably occurred.
Check to see if unpublished trials have been included in the analysis.
How can heterogeneity variation be modelled for?
A fixed effect model - assumes all studies are estimating th same true value and variation is variation within the study, not between studies.
A random effects model - assumes all studies have their own true value that lies in a range (the distribution of true effect). Variation is between studies also.
Compare the fixed effect model and random effects model.
Fixed effect =
less equal weighting (more given to smaller studies)
similar point estimate to random effects model
Random effects =
more equal weighting (less given to smaller studies)
Why is the test for heterogeneity weak?
Heterogeneity is hard to detect - only works at a 10% significance level.
When will a meta-analysis not be performed on a systematic review?
When the studies are heterogeneous = non comparable
What is a systematic review?
An overview of primary studies that used similar methods to study the same topic.
What is a meta-analysis?
Quantitative synthesis of results of multiple primary studies on the same topic.
What is reporter bias?
When if a patient is receiving a placebo, their symptoms etc may be ignored by an assessor who knows they receive the placebo.
What is clinical equipoise?
The genuine uncertainty over which treatment is better in an RCT.
What is ethical recruitment?
Recruitment from the region in which the drug will be used
No unethical exclusion from the trial
What makes a clinical trial ethical?