HaDPop Flashcards
What are the two approaches to the concept of causality?
Deterministic and Stochastic
What does the deterministic approach measure?
NAME?
How does the deterministic approach of causality validate the hypothesis?
By systematic observations to predict with certainty future events
What does the stochastic approach measure?
NAME?
How does the stochastic approach to causality assess the hypothesis?
By systematic observations to give the likelihood of future events
Based on the stochastic approach, does a significant association mean causality exists?
No
What is population-based risk?
How individuals infer their personal risk
What do population-based observational studies do?
Investigate the causes of disease
What is the purpose of population based interventions?
They treat and prevent disease
What is the purpose of population based intervention trials?
They evaluate drugs and interventions
Where is critical appraisal of evidence necessary?
To decide about causality
How useful is laboratory bases evidence in determining causality?
It is contributory, but neither necessary not sufficient
What can population based evidence give?
Association, but not causality
Where are the ‘universal’ sources of information?
- Birth registration
- Death registration
- Population census
How often is the population census done?
Every 10 years
What is a census?
The simultaneous recording of demographic data by the government at a particular time, pertaining to all the persons who live in a particular territory
What does the census describe?
Both households and people
Who runs the census?
The government
What is the incentive to complete the census?
It is mandatory by law, and failure to complete is punishable by fine or imprisonment
What does the census cover?
A defined area at one time
Who fills out the census?
Personal enumeration, or a person in each household completes the census form
On what timescale is the census performed through a defined area?
Simultaneously
What does the census provide?
Universal coverage
What information can be obtained from the census?
NAME?
What can the population size be used for?
Measurements of rates
Why is it important to know the population structure?
To service needs
Give an example of population characteristics the government may need to know
Measures of deprivation
Give 5 measures of deprivation
- Unemployment
- Overcrowding
- Lone pensioners
- Single parents
- Lack of basic amenities
What influences a population size?
- Births
- Deaths
- Migration
Who provides birth notification?
An attendant at birth, usually the midwife
How quickly does a birth notification need to be submitted?
Within 36 hours
Where does a birth notification need to be submitted to?
The local Child Health Register
Why is birth notification important?
For relevant services such as immunisation
What is the incentive for carrying out birth registration?
It is required by law
Who registers a birth?
Parent
How soon does a birth need to be registered?
Within 42 days
Where should a birth be registered?
At a local Registrar for Births
What is the purpose of birth registration?
- Statistical purposes
- Makes you identifiable
How does birth registration make you identifiable?
You get a birth certificate
What are the measures of fertility?
NAME?
What is the CBR?
The number of live births per 1000 population, including men, women, children and old people
What is the GFR?
The number of live births per 1000 females ages 15-44
What is the TPFR?
The average number of children born to a hypothetical women in her life
What is TRFR not influenced by?
The size of population in different age groups
How is TPFR calculated?
∑(all current age-specific fertility rates)
What does a TPFR of 2 mean?
Replacement
What does a TPFR of >2 mean?
A growing population
What is GFR affected by?
Age specific birth rates (ADBR) and age distribution within the 15-44 year olds
What does TPRF give each age?
Equal weighting in it’s calculation
What are the determinants of fertility?
- Fecundity
- Fertility
What is fecundity?
The physical ability to reproduce
What decreases fecundity?
Increase in sterilisation and hysterectomies
What is fertility?
Realisation of the ability to reproduce
What is fertility based on?
Humans, not biological
What increases fertility?
- Sexual activity
- Good economic climate
What decreases fertility?
- Contraception
- Abortion
What does conceptions equal?
Live births + miscarriages + abortions
What is the CBR used for?
Describing the impact of births on populations
What is the GFR used for?
Comparing the fertility of female populations
What is TPFR used for?
Comparing the fertility of females without being influenced by age-group structure
Whos statutory obligation is death certification?
The attending doctor
What can happen if a doctor doesn’t provide death certification?
They can go to prison
What is a doctor legally required to do on the death certificate?
Provide information on likely cause(s) of death
What must a doctor do if the cause of death is unusual or uncertain?
Notify the Coroner’s Officer
Who must perform death registration?
A qualified informant, usually a relative
How quickly does death registration need to be performed?
Within 5 days
What does death registration require?
A Death Certificate from a doctor
What are the measures of mortality?
NAME?
What is the CDR?
The number of deaths per 1000 population
What is the ASDR?
The number of deaths per 1000 in an age group
What does the SMR do?
Compares the number of ‘observed’ deaths with the number of ‘expected’ deaths if the age-sex distribution of the populations were identical
What does SMR adjust for?
Age-sex distribution
What are the reasons for collecting mortality data?
NAME?
What do population estimates do?
Apply what is known about births, deaths and migration to the present
What do population projections do?
Estimate the future populations
What additional assumptions are made in population estimates?
About births, deaths and migration in the future
What questions does identification of health and healthcare necessitate?
NAME?
What does a trend involve?
The comparison of rates, which require a numerator and a denominator
What are rates often?
Per unit time
What does a trend imply?
A comparison over time
What can a trend be a comparison of, other that over time?
Comparison between places, across socio-economic groups etc., or a combination
What are the two types of errors that can occur in trend monitoring?
- Numerator errors
- Denominator errors
What are some possible opportunities for numerator errors?
- Death certification
- Disease diagnosis
- Classification or coding errors
What are some possible opportunities for denominator errors?
- Population used
- Population definition
- Population count or estimate
What can trends be due to?
- Chance (random) variation
- Artefactual (systematic) reaosns
- Real phenomenon
- ‘Natural’ (epidemiological)
- Medical care effects
What should be done when there is a dramatic change in trends?
Consider artefactual reasons before considering real phenomenon
What are some contentious issues?
- Purpose
- Users
- Quality
- Comparability
- Relationship
- Publication
- Access
- Funding
What are the potential purposes of scientific studies?
NAME?
Who are the possible users of information obtained from studies?
NAME?
What is the competition in quality of data?
Real-time data vs validated data
What is comparability of data in competition with?
Comparable vs customised
What are the possible relationships in data found in studies?
Integral vs indepedant
Where could data from studies be published?
NAME?
What is the conflict in access to data?
Data protection vs. freedom of information
What are the possible sources of funding for studies?
NAME?
What concepts does the ‘amount’ of disease have?
- The number of new cases that occurred
- The number of people affected by the disease
What does the concept of amount of new disease focus on?
New events
When is the concept of number of new cases that occur useful?
When monitoring epidemics
What does the concept of number of people affected count?
The number of people with the disease, counting both old and new cases
What does the concept of number of people affected by the disease describe?
The burden of disease
Where is the measure of number of people affected by a disease useful?
As a measure of need for services
How do you calculate incidence rate?
New events / (person * time(yrs))
What is the unit for incidence rate?
Events per persons per year
Is prevalence a rate?
No, it’s a proportion
What is the denominator for prevalence?
Persons (not persons per time)
What kind of study is use to determine prevalence?
Cross sectional
What does an increase in incidence lead to?
An increase in prevalence
What does a cure, or death of patients lead to?
Lower prevalence
What does a longer survival rate lead to?
Increased prevalence
How can prevalence be calculated?
- Incidence * length of disease
- Cases / population
What is incidence?
The measure of the populations average risk of disease
What exists within a population regarding risk of disease?
Variations in risk of disease between groups of people
Why are systemic variations in risk between people of great interest?
Because it can give clues about the aetiology (cause) of a disease
How can variations in prevalence be used to determine aetiology?
Can compare levels of exposure in two groups of people and try to identify the causal factor for a disease
What may be done after identifying the causal factor for a disease?
Try and prevent exposure, thus reducing incidence of disease
What is the incidence rate ratio (IRR)?
A comparison of incidence rates between groups with different levels of exposure
How is the IRR calculated?
Rate B (Exposed) / Rate A (Unexposed)
What is implied if the incidence rate in group B is higher than that in group A?
The difference in exposures was associated with the differences in rates of disease
How can efficacy of treatments be measured?
Incidence rate ratios can be used to compare the effects of two treatments, and decide which one is best
Give two examples of nuisance variations in risk of disease?
NAME?
What is the rate ratio for most diseases when comparing the rate old with rate young ?
> 1.0
Why is knowing that there is variation based on age and sex not that useful for prevention?
Whilst it may be possible to target prevention at particular age-sex groups, age and sex are not modifiable factors
What can confounding factors explain?
All or part of an apparent association between an exposure and a disease
Give two ways of dealing with confounding by age
- Use age specific rate ratios
- Use standardised mortality ratios
How can using age specific rate ratios help deal with confounding by age?
With narrow age bands, little confounding due to age occurs
What is the problem with using age specific rate ratios?
Results are difficult to interpret as you get too many answers, as there is one for each age band
What does the SMR look at?
The rate ratio for two populations if age-sex structure of the two populations was the same
What is indirect SMR comparing?
The levels of mortality observed in a study population with the level of mortality expected if a standard reference populations age-sex specific ratios were applied to the study population age-sex groups
What does SMR account for?
Any age-sex confounding
How is SMR usually expressed?
As a %
What would a SMR of 100 mean?
There there is the same risk in the study population as in the standard reference population
What does a SMR of >100 mean?
A higher risk in the study population
How can SMRs be expressed if not a %?
Relative to 1.0
Essentially, what is the ‘observed’ value?
Our best estimate of the ‘true’ or ‘underlying’ tendency
What is a hypothesis?
A statement that an underlying tendency of scientific interest takes a particular quantitive value
What must be calculated in formal hypothesis testing?
The probability of getting an observation as extreme as, or more extreme as, the observed, assuming the stated null hypothesis is true
What happens if the probability of getting an observation as extreme as the observed is very small?
It is reasonable to conclude that the data and the stated null hypothesis are incompatible
What has happened if there is a very small probability that getting an observation as extreme as the one you observed?
- Something very unlikely has happened or
- The stated hypothesis is wrong
What is the calculated probability of getting a value as extreme as the observed called?
P-value
When is an observation statistically significant?
When the p value≤ 0.5
What does a p value of >0.05 not mean?
That they null hypothesis has been proven
What are the limitations of hypothesis testing?
- Rejecting a null hypothesis is not always useful
- Statistical significance depends on sample size
- Statistically significant doesn’t mean it’s clinically important
Why is rejecting a null hypothesis not always useful?
P≤ 0.05 is arbitrary
What is meant by P≤ 0.05 being arbitrary?
Nothing special happens between p=0.049 and p=0.051
What is it usual practice to hypothesis test again?
A null hypothesis
What is a null hypothesis?
A hypothesis assuming that two things are equal, or that there is no effect or difference
What information may be required by epidemiologists, and health service managers?
- Underlying tendencies
- What tendencies imply about the patterns of disease
- Health care need in the general population
What is the 95% confidence interval?
The range within which we can be 95% certain that the ‘true’ value of the underlying tendency really lies
What is the range of the 95% CI centred on?
The observed value
Why is the range of the 95% CI centred on the observed value?
Because it is always out best guess at the ‘true’ underlying value, so the observed value always lies between the 95% CI
What are values in the 95% CI said to be?
‘Consistent with the data’
What happens if the null hypothesis value is consistent with the observed data?
Any observed difference from the null hypothesis may be due to chance
How can you decide wether the finding is statistically significant?
Using the 95% CI
How do you calculate 95% CI?
- Calculate observed value of whatever you’re interested in
- Calculate error factor
- Lower 95% confidence limit = observed value / e.f.
- Upper 95% confidence limit = observed value * e.f.
What happens as we get more data?
We get more sure about the ‘true’ underlying value
Why do we get more sure about the true underlying value as we get more data?
The e.f. gets smaller and the 95% CI gets narrower
What are the features of an ideal study?
- Basic scientific method comparing ‘like with like’
- Two identical groups differing only in exposure of interest
What could be done if a study was ideal?
Differences can then reasonably be attributed to the exposure
Why can an ideal study not be achieved?
It’s impossible to get two identical groups of people differing only the exposure of interest, when exposure is linked to other factors
What can be done in an experiment?
Force all other factors to be identical
How can a study be randomised?
A randomised control trial
What can be done using a cohort study?
Measure and record any non-identical features
What is best, an experiment, randomisation or a cohort study?
An experiment, then randomisation
What must be counted in a cohort study?
Outcome events and person years, in exposed and unexposed groups
What are person-years?
The sum of the total time of everybody followed up in the study
Who must be recruited in a cohort study?
Outcome free individuals
What must the individuals recruited for a cohort study be classified into?
Exposed and unexposed categories
What are the advantages of cohort studies over routinely obtained data?
- You can study exposures and personal characteristics that are not routinely collected
- You can obtain more detailed information on outcomes or exposures
- You can collect additional data on potentially confounding factors
What do all cohort studies involve?
Prospective follow up
What is counted on the follow up of cohort studies?
Person-years (p-y) and d (developed)
When may data collection begin in a cohort study?
NAME?
What is it called when data collection starts immediately or later in a cohort study?
Concurrent or prospective cohort study
What is it called when data is collected from the past in a cohort study?
Historical or retrospective study
How is a historical cohort study carried out?
Recruitment of outcome free individuals, classification of their exposure status and subsequent outcomes is done using historical data
How can comparisons be made in cohort studies?
Internally, or against external reference population
What does internal comparisons of cohort study data use?
IRR
What does external comparisons of cohort study data use?
SMR
Why is an SMR approach for a cohort study important?
Because cohort studies are usually conducted over long periods, often decades, so people age during the study
How is a ‘Lexis’ diagram produced?
- Calculate separately the number of ‘expected’ cases or deaths for each calendar time period
- The expected number of cases or deaths in each cell then summer over all the cells, i.e. over all age groups and for all calendar time periods, to give total number of expected cases or deaths
- Can also add additional classification variables, but you are limited to the variables recorded by routine data sources
How is the number of ‘expected’ cases or deaths for each age group in a time period calculated?
- Obtain reference populations age-specific rates for each calendar time period from routine data sources
- Multiply these rates by appropriate cells’ person-years to estimate the expected number of cases and deaths in each cell
What additional classification variables can be added to a Lexis diagram?
Age-sex specific rates at each calendar time period
When is comparison with external reference population is useful?
When you cannot use sub-cohorts
What are the limitations of external comparisons in cohort studies?
- Often limited data available for reference population
- Often no incidence data
- Usually have to make do with mortality data
- Study and reference populations may not be comparable
Why may study and reference populations not be comparable?
Selection bias
Give an example of a form of selection bias
Healthy worker family
What is the result of the healthy worker effect?
Many occupational cohorts yield SMRs of well below 10%
What causes the healthy worker effect?
Since employment is often restricted to healthy individuals
What is the advantage of concurrent cohort studies?
Enables detailed and prospective assessment of exposure, outcomes and confounders
What are cohort studies better than case control studies at?
- Studying a range of different outcomes
- Studying rare exposure
- Establishing that exposure(s) precede outcome(s)
Which conditions are cohort studies better for?
Those that fluctuate with age, both randomly or systematically
What are the disadvantages of cohort studies?
- Usually large and resource intensive
- Take long time
- Rigorous definitions of outcome and exposure can require expensive and sometimes intensive/invasive investigation
- Risk high number of losses to follow up
- Results take a long time
- Not good for rare outcomes
- Difficulty with confounding, especially unknown confounders
Which kind of cohort studies are quicker?
Historical
What is produced when there are a high number of losses to follow up?
Survivor bias
What causes survivor bias?
When those who remain in the study differ from those who left
What is the result of cohort studies taking a long time?
Potentially ethical dilemmas, can become politically charged
Why are cohort studies not good for rare cases?
You would get too few cases
What must you do to conduct a case-control study?
- Identify a group of cases
- Identify a suitable group of non-cases (controls)
- Ascertain previous exposure status of everyone
- Compare level of exposure in cases and controls
Why do we need case-control studies?
- Conventional cohort studies take a long time
- Cohort studies are expensive, especially if you need detailed information
- Cohort studies are not good for studying rare events
Why are cohort studies not good for studying rare events?
Because they would need to be impossibly large
What is the rare disease assumption?
IRR = AD/CB
When A= disease in exposed (person-years), B= no disease in exposed (person-years), C= disease in unexposed (person-years) and D= no disease in exposed (person-years)
What does AD/CB always equate to under the rare disease assumption?
The odds ratio
What is the odds ratio a valid measure of?
Excess risk in cases compared with controls
How do calculate error factor?
Error factor = e 2*√(1/a)+(1/b)+(1/c)+(1/d)
What is the precision of odds ratio affected by?
The number of healthy people, as well as the number of cases
What is the result of the precision of an OR being affected by the number of cases?
It is worth increasing the number of controls up to a point
Up to which point is it worth increasing the number of controls?
4 to 6 as many times as many controls as there are cases
In what direction do cohort studies look?
Always forward in time
In what direction do case control studies look?
Backwards in time
What happens in a conventional case-control study?
Retrospective collection of data
How is data obtained in a conventional case control study?
From recall
What happens in a nested case-control study?
Collection of data from the evolving outcome and exposure database of a ‘concurrent’ and ‘prospective’ cohort study
Why is a nested case control study so named?
Because it’s a case control study ‘nested’ within a cohort study
What are the advantages of nested case control studies over conventional?
NAME?
What are the advantages of conventional case control studies over nested?
Can collect more detailed information for a minority of participants
What are the key issues for case-control studies?
- Selection bias
- Information bias
- Confounding
What is the most difficult aspect of case-control studies to deal with?
Selection bias
What should cases selected be representative of?
All cases
What should controls selected be representative of?
The population from which the cases came
What can cause information bias?
NAME?
Give an example of a non-differentiated misclassification
Randomly incorrect measurement
Give 3 examples of systematic misclassification
- Recall bias
- Assessor bias
- Data collection methods differ
How can confounding be minimised?
By matching important confounders
How can confounding be adjusted for?
Analysing with logistic regression
What does disease result from?
The interplay of host, environment and the agent
What is a cause?
An exposure or factor that increases probability of disease
To exposures have to be necessary or sufficient to be important causes?
No
What is the aim of the use of knowledge?
To remove, avoid or protect against harmful factors
What are the two observational study designs?
NAME?
What are the two types of analytical studies?
NAME?
What are possible explanations for correlation?
Systematic and random variation
What can cause systematic and random variation?
NAME?
What is wrong with results due to confounding?
They are erroneous
What can confounding factors be?
- Known factors
- Possible factors
- Unknown factors
Give 2 examples of known confounding factors
NAME?
Give an example of a possible confounding factor
Deprivation
What could act as an unknown confounding factor?
Genetics
What is wrong with results due to bias?
They are incorrect
Give two types of bias
NAME?
What is the problem with selection bias?
- Unrepresentative of population being studied
- Group comparison not ‘like with like’
What can cause information bias?
- Differential recall
- Differential observation
- Differential measurement
- Differential classification
What measures chance?
The p-value
What is reverse causality?
If you believe X → Y, but actually Y → X
When does a true causal association exist?
When X → Y
What is Bradford Hill’s viewpoints or criteria for inferring causality?
- Association features
- Strength of association
- Specificity of association
- Consistency of association
- Exposure/outcomes
- Temporal sequence
- Dose response
- Reversibility
- Other evidence
- Coherence of theory
- Biological plausibility
- Analogy
What is meant by strength of association?
A causal link is more likely with strong associations
How is the strength of associations commonly measured?
By a rate ratio or odds ratio
What are strong associations unlikely to be explained by?
Undetected confounding or bias
Can weak associations be causal?
Yes
What is meant by specificity of association?
A causal link is more likely when an outcome is associated with a specific factor, and vice-versa
What does specificity of association strengthen?
The case for a causal link
Does a lack of specificity weaken the case?
Not necessarily
Why does a lack of specificity not necessarily weaken the case?
Current models of disease causation are multi-factorial
What is meant by consistency of association?
A causal link is more likely if the association is observed in different studies and different sub-groups
What is consistency of association between studies or groups unlikely to be due to?
The same confounding or bias
Why may inconsistency exist?
- Because of differences in other causal factors
- Features of study design
What is meant by temporal sequence?
A causal link is more likely if exposure to the putative factor has been shown to precede the outcome
Can a causal link exist if the outcome preceded exposure to the putative factor?
No
What are the optimal study designs?
- Randomised control trails
- Prospective cohort studies
What are the weak study designs?
NAME?
What is meant by reversibility?
A causal link is very likely if removal or prevention of the putative factor leads to a reduced or non-existent risk of acquiring the outcome
What is the significance of reversibility?
The strongest evidence for causal relationship
Why is reversibility often difficult to demonstrate?
- Many diseases have long time lags
- Ethical issues for a RCT of a prevention programme
- A public health programme to remove or prevent an exposure often requires action by society
What is meant by coherence of theory?
A causal link is more likely if the observed association conforms with current knowledge
What forms of coherence can strengthen a case?
With current paradigms, constructs or theories
What is the problem with coherence of theory?
It can lead to inappropriate rejection of ‘unfavored associations’
Does lack of coherence rule out a causal link?
No
What is meant by a dose response?
A causal link is more likely if different levels of exposure to the putative factor leads to different risk of acquiring the outcome
What is meant by biological plausibility?
A causal link is more likely if a biologically plausible mechanism is likely or demonstration
What limits biological plausibility?
Current knowledge
What is meant by analogy?
A causal link is more likely if an analogy exists with other diseases, species or settings
What is the advantage of an analogy over biological plausibility?
It is easier to infer
What is the problem with analogies?
They may be inappropriate
What is epidemiology?
The study of distribution and determinants of health-related states, or events in specified populations, and the application of this study to the control of health problems
What assumptions are made in epidemiology?
- Disease does not occur at random
- Disease has causal and preventable factors that can be identified through systematic investigation
What has the evolution of the concept of causality lead to?
The adaptation of the probabilistic approach to considering causality based on assessment of likelihood or risk of disease occurring, as now thought that multi-factorial web of factors causes disease
What does epidemiological reasoning involve?
- Hypothesis
- Analytical study
- Observed associations
- Cause-effect relationships
What is a hypothesis generated from?
Observations and/or theories
What is analytical study?
Systematic observations of comparisons
What are observed associations?
Possible explanations of non-causal associations
Give 3 examples of non-causal associations
NAME?
What is a cause-effect based on?
Judgement of how the observed associations fits in with information from other sources
What is a clinical trail?
Any form of planned experiment which involves patients and is designed to elucidate the most appropriate method of treatment of future patients with a given medical condition
What does a clinical trial measure?
The outcome of the new treatment compared to the outcomes of the standard treatment
What is the purpose of a clinical trial?
To provide reliable evidence of treatment efficacy and safety
What is efficacy?
The ability of a health care intervention to improve the health of a define group under specific conditions
What is safety?
The ability of a health care intervention not to harm a defined group under specific conditions
What does a clinical trial need to be to be able to give a fair comparison of effect and safety?
- Reproducible in experimental conditions
- Controlled
- Fair
What are clinical trials subject to?
Random variation
What differences found in clinical trials are more prone to chance?
Those observed in small trials of
What happens in the preclinical phase of drug development and monitoring?
Laboratory studies
What is being tested in the preclinical phaseof drug development and monitoring?
Pharmacology and animal toxicity
What is tested on in the preclinical phase of drug development and monitoring?
Cell cultures and animals
What happens in phase I of drug development and monitoring?
Volunteer studies
What is being tested in phase Iof drug development and monitoring?
Pharmacodynamics, pharmacokinetics, and major side effects
Who is tested on in phase Iof drug development and monitoring?
What happens in phase IIof drug development and monitoring?
Treatment studies
What is tested in phase IIof drug development and monitoring?
Effects, dosages, and common side effects
Who is tested on in phase IIof drug development and monitoring?
What happens in phase IIIof drug development and monitoring?
Clinical trails
What is being tested in phase IIIof drug development and monitoring?
Comparison with other treatments
Who is tested on in phase IIIof drug development and monitoring?
What happens in phase IVof drug development and monitoring?
Post-marketing surveillance, monitoring for adverse reactions and looking for potential new uses
Who is tested on in phase IVof drug development and monitoring?
The whole population
What do non-randomised clinical trails involve?
The allocation of patients receiving a new treatment to compare with a group of patients receiving standard treatment
What is introduced in non-randomised clinical trials?
NAME?
Who can introduce allocation bias in non-randomised clinical trials?
- Patient
- Clinician
- Investigator
What does comparison with historical cohorts involve?
The comparison of a group of patients who had the standard treatment with a group of patients receiving a new treatment
What is the problem with historical comparisons in clinical trails?
For the standard treatment group,
- selection often less well defined, and less rigorous
- treated differently from new treatment group
- less information about potential bias/confounding
- unable to control for confounders
What needs to be defined in a randomised control trial (RCT)?
- The disease of interest
- The treatments to be compared
- The outcomes to be measured
- Possible bias and confounders
- The patients eligible for the trial
- The patients to be excluded from the trial
What must be done to conduct a RCT?
- Identify a source of eligible patients
- Invite eligible patients to be in the trial
- Consent patients willing to be in the trial
- Allocate participants to the treatments fairly, without bias and confounding
- Follow-up participants in identical ways
- Minimise losses to follow up
- Maximise compliance with treatments
What are we comparing outcomes to determine?
NAME?
What is the importance of the size of the difference between groups?
Determines if it’s clinically significant
What needs to be determined before the start of a clinical trial?
- Protocol for data collection
- Agreed criteria for measurement and assessment of outcomes
Why must outcomes of a clinical trail be pre-defined?
To prevent data dredging and repeated analysis
What is preferable in clinical trails regarding outcomes?
That there is only one, primary outcome
What is the primary outcome used in?
Sample size calculations
What are secondary outcomes?
Other outcomes of interest
What do secondary outcomes often include?
Occurence of side effects
What are the types of outcome?
- Pathophysiological
- Clinically defined
- Patient focused
Give 3 examples of pathophysiological outcomes
- Tumour size
- Thyroxine levels
- ECG changes
Give 3 examples of clinically define outcomes
- Death
- Disease
- Disabilities
Give 4 examples of patient focused outcomes
- Quality of life
- Pathological well being
- Social well being
- Satisfaction
What are the features of an ideal outcome?
- Appropriate and relevant
- Valid and attributable
- Sensitive and specific
- Reliable and robust
- Simple and sustainable
- Cheap and timely
Who must an outcome be appropriate and relevant to?
Patient, clinician, society etc
What is meant by an outcome being valid and attributable?
That any observed effects can be reasonably linked to the treatments being compared
What is meant by an outcome being sensitive and specific?
The chosen method of measurement can detect changes accurately
What is meant by an outcome being reliable and robust?
That the outcome is measurable by different people in various settings with similar results
What is meant by an outcome being simple and sustainable?
That the method of measurement is measurement is carried out repeatedly
What is meant by an outcome being cheap and timely?
Not excessively expensive to measure, nor has a long lag time
When are measurements made?
- Baseline measurements
- Monitoring outcomes during the trial
- Final measurement of outcomes
What is the purpose of baseline measurements?
Monitoring for inadvertent differences in groups
What is being monitored for during the trial?
Possible effects and adverse effects
What is happening when the final measurements are being made?
Comparing final effects of treatments in trial
What is non-random allocation?
Allocation of participants to treatments based on on a personal, historical basis, geographical location, convenience, numerical order etc
What can result for non-random allocation
Allocation bias and confounding factors
What is the problem with non-random allocation?
It can unwittingly cause unidentified differences between treatment groups being compared
What is the advantage of random allocation?
It allocates participants to treatments fairly, and this minimises allocation bias and confounding
How does random allocation minimise allocation bias?
Randomisation gives each participant an equal chance of being allocated to each of the treatments in the trial
How does random allocation minimise confounding?
In the long run, randomisation leads to treatment groups that are more likely to be similar in size, and characterised by chance
What kind of confounding does random allocation minimise?
Both known and unknown
Give 3 methods of randomisation
- Toss a coin
- Random number tables
- Computer generated random number
What is it called when the treatment allocation is known?
Open label
What may knowledge of which participant is receiving which treatment lead to?
Bias of the result
Why may knowledge of who is receiving which treatment lead to bias?
- Patients may alter their behaviour, other treatment or expectation of the outcome
- Clinician may alter their treatment, care and interest in the patient
- Investigator may alter their approach when making measurements and assessing outcomes
What are the types of blinding?
- Single blind
- Double blind
- Triple blind
What happens in a single blind trial?
One of the patient, clinician or assessor (usually patient) doesn’t know treatment allocation
What happens in a double blind trial?
Two of the patient, clinician or assessor doesn’t know the treatment allocation (usually involved the patient not knowing)
What happens in a triple blind trial?
All do not know allocation
What must be done to blind a trial?
- Aim to make treatments identical in every way
- Use a designated pharmacy to label identical containers for the treatments with code numbers, and have code sheet detailing which code number corresponds to which treatment
In what respects must a treatment be identical?
- Appearance
- Taste
- Texture
- Dosage
- Regime
- Warnings
Where can blinding be difficult?
- Surgical procedures
- Psychotherapy vs. anti-depressant
- Alternative medicine vs. Western medicine
- Lifestyle interventions
- Prevention programmes
What is the problem with controlling with no treatment?
The effect of comparing ‘new’ treatment group with a group receiving no treatment is to leave one unsure as to whether any observed difference was due to the new treatment, or just to that group receiving care
What is the result of the placebo effect?
Even if the therapy is irrelevant to the patients condition, the patients attitude to their illness, and indeed the illness itself, may be improved by thinking something is being done about it
What is a placebo?
An inert substance made to appear identical in every way to the active formulation with which it is to be compared
What is the aim of a placebo?
To cancel out any placebo effect that may occur in the the active treatment
What are the ethical implications of a placebo?
Use of placebo is a form of deception
When should a placebo be used?
Only ben no standard treatment is available
What is ethically essential when using a placebo?
Patients in a placebo-controlled trial are informed that they may receive a placebo
Why may losses to follow up occur?
- Their clinical condition may necessitate their removal from the trial
- They may choose to withdraw from the trial
What must be done to minimise losses to follow up?
- Make follow up practical and minimise inconvenience
- Be honest about commitment required for participants
- Avoid coercion or inducements
- Maintain contact with participants
Why may non-compliance with treatment occur?
- May have misunderstood instructions
- May not like taking their treatment
- May think that their treatment is not working
- May prefer to take another treatment
- Can’t be bothered to take their treatment
How can compliance to treatment be maximised?
- Simplify instructions
- Ask about compliance
- Ask about effects and side effects
- Monitor compliance
How can compliance be monitored?
- Tablet count
- Urine level
- Blood level
What is not possible in practice?
To have 100% follow up, and to guarantee than 100% compliance took place
What is the result of the impossible to obtain 100% follow up and compliance?
Any analysis of outcomes should take this issue into account
What are the two interpretations of wether a new treatment is better than a standard treatment?
NAME?
What are the two types of analysis?
NAME?
Who does an ‘as treated’ analysis look at?
Only those who completed follow up and complied with treatments
What does ‘as-treated’ analysis compare?
The physiological effects of treatment
What is the problem with ‘as-treated’ analysis?
Non-compliers are likely to be systematically different from compliers, and so introduces selection bias and confounding
What does an ‘intention-to-treat’ analyse according to?
The original allocation of treatment groups, regardless of wether they completed follow up or complied with treatment
What does an intention to treat analysis compare?
The likely effects of using the treatments in routine clinical practice
What is the advantage of an intention to treat analysis?
It preserves the effects of randomisation
What do as treated analyses test to give?
A larger size of effect
What do intention to treat analyses tend to give?
Smaller and more realistic sizes of effect
How should clinical trials normally be analysed?
On an intention to treat basis
What are the ethical principles for medical research involving human subjects?
- The health of the patient must be the physicians first consideration
- A physician shall act only in the patients interest when providing medical care which may have the effect of weakening the mental or physical condition of the patient
What is a collective ethic?
That all patients should have treatments that are properly tested for efficacy and safety
What principles apply in the individual ethic?
- Beneficence
- Non-malifecence
- Autonomy
- Justice
What are RCTs for the benefit of?
Future patients
How does the collective ethic apply to RCTs?
They aim to properly test treatments for efficacy and safety
How does the individual ethic apply to RCTs?
- Do not guarantee benefit
- Could result in harm
- Allocate treatments by chance
- Place burdens and confer benefits
What issues should be considered for a clinical trial to be ethical?
- Clinically equipoise
- Scientifically robust
- Ethical recruitment
- Valid consent
- Voluntariness
What must be done for a trial to go ahead?
Approval by Research Ethics Committee
What is clinical equipoise?
When there is reasonable uncertainty or genuine ignorance about the better treatment or intervention (including the non intervention
What are the issues with clinical equipoise?
- Is the ‘uncertainty or ignorance’ by the individual clinician, or for the scientific community as a whole?
- What constitutes ‘reasonable uncertainty’
- How is ‘better’ defined for the individual patient or for society as a whole
What features must a study have to be scientifically robust?
- Addresses a relevant or important issue
- Asks a valid question
- Has an appropriate study design and protocol
- Has the potential to reach sound conclusions
- Can justify use of comparator treatment or placebo
- Has acceptable risks of possible harm compared to anticipated benefits
- Has provision for monitoring the safety and well being of participants in trial
- Has arrangements for appropriate reporting and publication
What must be true for a study to have an appropriate design and protocol?
It must address potential bias and confounding
Why must a study have the potential to reach sound conclusions?
To minimise inability to find a clinically important effect using a sample size calculation
Give 2 examples for provision of participant safety and well being monitoring
- Data monitoring
- Patient monitoring
What are the two issues of ethical recruitment?
- Inappropriate inclusion of;
- participants for communities that are unlikely to benefit
- participants with high risk of harm with respect to potential benefit
- participants likely to be excluded from analysis
- Inappropriate inclusion of;
- people who differ from an ideal homogenous group
- people who it is difficult to get valid consent for
What are the features of valid consent?
- Knowledgable informant
- Appropriate information
- Informed participant
- Competent decision maker
- Legitimate authoriser
How should appropriate information be given?
Verbal and written
What should be given after receiving information?
- Cooling off period
- Ability to opt out
What should be given as evidence of valid consent?
Signed consent form
Does a signed consent form equate to valid consent?
No
What is voluntariness a pre-requisite for?
Consent to be valid
What is meant by voluntariness?
The decision should be free from coercion or manipulation, or the perception that coercion or manipulation may take place
What is the result of perceived coercion or manipulation?
Invalidates consent as much as actual coercion or manipulation
What can coercion be?
Non-access to ‘best’ treatment, lower quality care, disinterest by clinician
What can manipulation be?
Exploitation of emotional state, distortion of information or financial inducements
What is the issue of voluntariness an issue of?
Autonomy
Why is the issue of voluntariness an issue of autonomy?
If undue influence is being exerted so that a potential participant acts uncharacteristically then influence is regarded as unethical
What are NHS Trusts / the PCD R&D Office concerned with
- Research governance
- Financial management
- Resource implications
What do the Research Ethics Committee concerned say?
‘The dignity, rights, safety and well being of participants must be the primary consideration in any research study’
What do the Research Ethics Committee focus particularly on?
- Scientific design and conduct of study
- Recruitment of research participants
- Care and protection of research participants
- Protection of research participants confidentiality
- Informed consent process
- Community considerations
What should healthcare services and interventions be based on?
The best available evidence
What should the best available evidence be based on?
Rigorously conduced research
What must be looked at in research into best available evidence?
- Primary research studies
- Literature review of studies
- Decision analyses
Give two types of literature reviews of studies
NAME?
What are narrative reviews looking at?
Implicit assumptions, opaque methodology
What is the result of a study not being reproducible?
It is considered biases and subjective
What do systematic reviews look at?
Explicit assumptions, transparent methodology
What do decision analyses look at?
- Harms and benefits
- Cost-effectiveness
What is required for systematic reviews require?
A clearly focused question
What are explicit statements made about in systematic reviews?
- Types of study
- Types of participants
- Types of interventions
- Types of outcome measures
What do systematic reviews involve?
- Systematic literature search
- Selection of materials
- Appraisal
- Synthesis (possibly including meta-analysis)
What is the advantage of a systematic review?
It is an extremely credible source of evidence
What are the key aspects of a systematic review?
- Explicit
- Transparent
- Reproducible
What is a systematic review an overview of?
Primary studies that used explicit and reproducible methods
What is a meta-analysis?
A quantitive synthesis of the results of two or more primary studies that addressed the same hypothesis in the same way
What is the purpose of meta-analyses?
- To facilitate the synthesis of large number of study results
- To systematically cellate study results
- To reduce problems of interpretation due to variations in sampling
- To quantify effect sizes and their uncertainty as a pooled estimate
What are the quality criteria of meta-analyses?
They should have a formal protocol specifying;
- Compilation of complete set of studies
- Identification of common variable or category definition
- Standardised data extraction
- Analysis allowing sources of variation
Does a systematic review always include a meta-analysis?
Not always
When may a systematic review not contain a meta-analysis?
When clinical heterogeneity is too great
How is a pooled estimate odds ratio for all studies calculated?
- OR and their 95% CIs are calculated for all studies in meta-analysis
- These are combined to give pooled estimate odds ratio using a statistical computer program
- Studies weighted according to their size and uncertainty of their odds ratio
In a pooled estimate OR, what does a smaller e.f. mean?
A greater weight to results
How are forest plots interpreted?
- Individual odds ratios are squares, with their 95% CIs being displayed as lines
- The size of each square is proportional to the weight given to the study
- The diamond is a pooled estimate with the centre indicating the pooled odds ratio (dotted line), and the width representing the pooled 95% CI
- The solid line is the null hypothesis CI
What are the problems with meta-analyses?
- Heterogeneity between studies
- Variable quality of studies
- Publication bias in selection of studies
What must be done to determine heterogeneity between studies?
- Modelling for variation
- Analysing the variation
How can variation be modelled for?
Fixed effect model vs random effects model
How can variation be analysed?
Sub-group analysis
Ideally, what should all studies in a meta-analysis be similar in terms of?
- Study design
- Participant profile
- Treatments or exposure
- Outcomes measured
- Statistical analysis used
What is the problem with finding identical studies for meta-analyses?
In practice, no two studies are identical
What are the two approaches to calculating the pooled estimate odds ratio and its 95% CI to model for variation?
NAME?
What does a fixed effect model do?
Assumes studies are estimating exactly the same effect size
What does a random effects model do?
Assumes that the studies estimate similar, not same, effect size
How do the results of a fixed effect and random effect model differ?
NAME?
What is the problem with hypothesis testing for heterogeneity using fixed effects and random effects models?
Low statistical power to detect heterogeneity
What is often used in hypothesis testing for heterogeneity using fixed effects and random effects models?
10% significance levels
What model is superior, fixed effect or random effects?
Much debate
What is true if, in a test for heterogeneity, the result is the null hypothesis?
There is no heterogeneity
What is true if, in a test for effect, the result is the null hypothesis?
There is no difference in the studies
What can random effects modelling account for?
Variation
Can random effects modelling explain the variation?
No
What can sub-group analysis do?
Can explain heterogeneity
What is the advantage of sub-group analysis being able to explain heterogeneity?
It can provide further insight into the effect of treatment of exposure
What are the two types of possible sub group analysis?
- Stratification by study characteristics
- Stratification by participant profile
What happens in stratification by study characteristics?
Subsets of ‘whole’ studies are defined by characteristics
What characteristics can sub-sets of whole studies be characterised by?
- Study design
- Length of follow up
- Participant profile
- Recruitment criteria
What happens in stratification by participant profile?
Data is analysed by type of participants
What is the advantage of stratification by participant profile?
Has greater statistical power than for individual studies
What is the disadvantage of stratification by participant profile?
Data is often unreliable
What can variable study quality be due to?
- Poor study design
- Poor design protocol
- Poor protocol implementation
Order studies from least to most susceptible to bias and confounding?
Randomised control trials → Non-randomised control trials → Cohort studies → Case-control studies
What are the two main approaches to variable quality of studies?
- Define basic quality standard, and only include studies satisfying these criteria
- Score each study for its quality, then
- incorporate quality score into weighting allocated to each study during the modelling, so that higher quality studies have greater influence on the pooled analysis
- use sub-group analyses to explore differences
- meta-regression analysis
Give an example of a meta-regression analysis
Weighted linear regression of effect size against quality
What are the main components for assessing the quality of RCTs?
- Allocation methods
- Blinding and outcome assessment
- Patient attrition
- Appropriate statistical analysis
Who assesses quality?
> 1 assessor
What is the purpose of having >1 assessor?
To handle disagreement
What is the problem with assessors being blinded to results?
It’s sometimes difficult
Why does publication bias occur?
Studies with statistically significant or favourable results are more likely to be published than those studies with non-stastically significant or unfavourable results
Where does publication bias particularly occur?
To small studies
What does publication bias tend to lead to?
Biased selection in favour of studies in favour of demonstration of effects
How can publication bias be identified?
- Check meta-analysis protocol for identification of studies- it should include searching and identification of unpublished studies
- Plot results of identified studies against a measure of their size
- Use a statistical test for publication bias
What do statistical tests for publication bias tend to be?
Weak
How is a funnel plot showing publication bias interpreted?
- A plot of measure of study size against measure of effect
- If no publication bias, then the plot will be a ‘balanced’ funnel
- Smaller studies can be expected to vary further from ‘control’ effect size
- Publication bias likely to exist if there are few small studies with results indicating small or ‘negative’ measure of effect