Epidemiology Flashcards
what are the 2 types of epidemiology? and define them
Descriptive epidemiology: providing measures of frequency
Analytic epidemiology: testing hypotheses and associations
What is confounding? and what does it lead to?
effect of an extraneous variable
that wholly or partially accounts for the apparent effect of the study exposure or that masks an underlying true association
- Can lead to biased findings
- Can produce misleading results
what are the ways of identifying confounding in am epidemiological study?
Knowledge of subject matter
See whether the variable follows the 3 conditions for confounding
Stratification
Compare crude and adjusted estimates
N.B- You only need one method to identify confounding
Methods of identifying confounding:
How do you expand your knowledge of subject matter
- Explore literature
- Knowledge of similar biological pathways can be applied
- Not always possible, however, especially when investigating novel associations
Methods of identifying confounding:
What are the 3 conditions for confounding in a variable?
Check whether the variable is:
- Associated with the exposure in the source population
- Associated with outcome in the absence of the exposure
- Not a consequence of the exposure (in the causal pathway)
Methods of identifying confounding:
How do you use stratification and describe what it entails
- Stratify data by the variable of interest
- Compare stratum specific estimate with the estimate from the data analysis
- When a pooled estimate is significantly different (10%) from stratum specific estimates it is reasonable to think there is confounding
Methods of identifying confounding
How do you compare crude and adjusted estimates?
- Create a regression model adjusted for the variable
- If adjusted odds ratio differs from the crude odds ratio by 50% or more this may indicate confounding
- Not the optimal method though as adjusting for stuff may introduce confounding.
What is effect modification?
- Exists when the strength of the association varies over different levels of a third variable
- After controlling for confounding there is still a variable which affects the exposure or outcome
- This is a natural phenomenon
What are the stat tests for effect modification (to confirm that stratum specific estimates are truly different between them)
- Breslow-day test
- Q test
- Interaction terms in regression models- very frequently used. Interactions is synonymous to effect modification
what can you do about effect modification?
Do not try to control it; it is not a problem as it occurs in nature.
Instead take it into account and present stratified results
This effect can occur when you further stratify groups of exposure
in effect modification what is Synergism and Antagonism?
- Synergism = effect modifier potentiates the effect of the exposure
- Antagonism = effect modifier diminishes the effect of the exposure
what is the difference between confounding and effect modifier
Addressing a confounded relationship by addressing the exposure exclusively is very unlikely to yield a gain.
Addressing an exposure where effect modification is apparent may be useful. Hence interventions could be targeted ti a more homogenous pool of participants.
Effect modification affects exposure or outcome but not both whereas confounding could independently affect both exposure and outcome
what is a crude model of analysis
Univariate
It simply looks at the impact of the exposure on the outcome with no consideration of anything else
what are the features of multivariate analysis
Uses adjusted models- multiple exposures have been included.
The inference is that the outputs of these analyses mean that holding all other adjusted variables equal, X is the association between exposure and outcome.
e.g adjusted odds ration or adjusted hazard ratio.
it can help us to find confounding
what are koch’s postulates for infering causation
- Microorganism must be found in abundance only in diseased
- Microorganism must be isolated from diseased and grown in pure culture
- Cultured organism should cause disease when introduced to healthy organism
- Ethical problems here
- Must be reisolated from experimental host and identified as same causative agent
We do not use koch’s postulates for inferring causation, hence what criteria do we use to infer causation from both observational and interventional methods?
LIST THEM
Bradford-Hill Criteria
- Strength
- Consistency
- Specificity
- Temporality
- Biological gradient
- Plausibilty
- Coherence
- Experiment
- Analogy
Bradford hill criteria- EXPLAIN the following terms and give any relevant details:
- Strength
- Consistency
- Specificity
- Temporality
Strength
- Stronger association increases the confidence that an exposure causes an outcome
Consistency
- Consistent findings across settings tend to rule out errors or fallacies that might befall one or two studies
- Meta-analysis is a summation of this approach
Specificity
- Describes an association between specific causes and specific effects
- One of the most criticised criteria
- Lack of specificity does not necessarily invalidate a causal relationship
- Difficult when the disease is multifactorial
Temporality
- Insufficient for exposure A and Outcome B to co exist; A must precede B
- Not useful for cross-sectional studies
- Longitudinal studies are more useful
Bradford Hill criteria:
Explain the following terms and give relevant details:
- Biological gradient
- Plausibility
- Coherence
- Experiment
- Analogy
Biological gradient
- Dose-response effect (in the ‘right direction’) is a compelling argument for causality. e.g smoking and cancer
Plausibility
- This more intuitive
- Relationship should be biologically plausible where the science is understood
- However, where there is deficient understanding, assessing whether a relationship is plausible or not may not be possible
Coherence
- Association should be consistent with the existing theory and knowledge
- Can be an issue when challenging current beliefs or questioning the status quo
Experiment
- Evidence from experimentation should be supportive of the proposed link
- However scientifically desirable, experimentation is often not ethical when dealing with public health issues
Analogy
- Drawing upon analogous findings, we may make inference on the relationship
- Important in understanding emergent diseases and new associations
Define correlation
Correlation is a statistical term describing a linear relationship between two variables
Validity and bias help us to determine whether a results from a study is relevant or trustworthy
What are the two types of validity and explain them
Internal validity
- The extent to which findings accurately describe the relationship between exposure and outcome in the context of the study . i.e. if an association truly exists in the study
External validity
- The extent to which these inferences can be applied to individuals outside the study population
- Internal validity is a prerequisite for this
- Sometimes referred to as generalisability
what is bias
Inference is valid when there is no bias
Bias is any trend in the collection, analysis, interpretation, publication or review of data that can lead to conclusions that are systematically different from the truth
What are 2 types of errors and can it lead to
Random and systematic error
- Random error can be overcome with a large enough sample size
If there is a systematic error, this leads to incorrect results regardless of sample size
Systematic error can introduce bias into a study
This reduces its validity
what are the types of bias
- Selection sias
- Information bias
- Confounding
what is selection bias and what studies are particularly susceptible to this type of bias
An individual’s chance of being included in a study sample may be related to both exposure and outcome
This leads to a biased estimate of the association between exposure and outcome.
Case-control studies are more susceptible to this
Describe and explain the variations/examples of selection bias
Berkson’s bias
- When there is a hospital-based case-control study
- Controls are selected among the hospitals patients
Healthy worker effect
- Active workers are more likely to be healthy than those who are retired
Non-response bias
- People who do not respond are systematically different to the people who do respond
what are the ways of mitigating selection bias
- Choose controls representative of target population
- Keep non-response to a minimum
- Compare respondents with non-respondents and explore any systematic differences between them
what is information bias ?
Misclassification of exposure or the disease status or both (placing people in wrong groups)
Usually to poorly defined variables or due to flaws in data collection.
what are the types of information bias and how do you mitigate them?
Interviewer bias
- Common flaw occurs when interviewers ask about exposure status
- May be more thorough with those how have the outcome desired
- Leads to misclassification of exposure status and biased odds ratio
Recall bias
- Specifically refers to the differentially inaccurate recall of past exposure between cases and controls
- Patients with outcome may be more likely to recall exposure
- We often forget past exposure, but try harder remember if we have disease
If all participants cant remember exposure there is no recall bias
How do we mitigate recall and interviewer bias?
Interviewer
- if the interviewer does not know the disease status
- If interview process is standardised so interviewers follow strict protocol
Recall:
- Prevent by using objective ways to assess exposure
- Eg. medical records or biomarkers
what are the two different types of misclassification and explain how it affects the odd ratios
Non-differential misclassification. when there are errors in determining the outcome but they happen equally among exposed and non-exposed groups.
Odds ratios ALWAYS biased TOWARDS the null.
Differential misclassification- when errors in determining an individual’s exposure status occur unevenly amongst cases and controls.
odd ratios may be bias towards or away from the null
what is the Lake Wobegon effect and Hawthorne effect?
Lake Wobegon effect
- Illusory superiority
- If you ask a class of students how good their driving is, most of them will believe themselves to be better than average
Hawthorne effect
- Consequence of participants realising they are being observed and therefore acting differently
what are the types of disease prevention? explain them
Primary- prevent disease by controlling exposure to risk factors
Secondary- apply the available measures to detect early departure from health and giving treatment and interventions
Tertiary- introduce treatment to reduce long term impairments and disabilities
Draw out the Demographic transition model with all the relevant stages
what are the stages of EPIDEMIOLOGIC transitions?
Draw it out
Describe the features of Stage 1 epidemiological transitions
Pestilence and famine
There is constraints on food supply
High birth rate and mortality
Life expectancy low at birth
what are the features of stage 2 of epidemiological transitions
Receding pandemics
A huge natural increase in population
High BR and reducing mortality
Agricultural development improves nutrition
Vaccination emerges at the end of this
what are the features of stage 3 of epidemiological transitions
Degenerative and man-made diseases
lifestyle factors and non-communicable disease dominate
Tech reduces need for physical labour
Addiction, violence and other issues emerge
What is Stage 4 of epidemiologic transitions
Give features of this
Delayed degenerative disease and emerging infections.
Health tech defers morbidity
Emerging zoonotic disease presents new threats
Inequalities within and between countries come to the lore
outline the hierarchy of evidence
what is DALY
(DALY) is a measure of disease burden that combines years of life lost from ill-health, disability or premature death.
People with low socio-economic status suffer the most as they don’t have resources to mitigate it
what are the 4 measures of frequency?
- Odds
- Prevalence
- Cumulative incidence
- Incidence rate
what is odds and how do you calculate it
Definition: a ratio of a probability of an event (P) to the probability of it’s complement (1-P)
Equation: (number of people with the disease)/ number of people without the disease.
what is the equation for prevalence and what are the features of it
Timepoint is very important in this
it reflects both the occurrence and duration of the disease
Can be used to monitor trends of disease overtime and hence to allocate health diseases
Not suitable for diseases of a short duration
what is cumulative incidence, give the formula and what factor is very important in this
used to measure how many new cases of the disease are there over a time period.
Time period is very important
it is a proportion; 0= no new cases in that time period and 1 suggest all individuals developed it during the time period
What is the other name for cumulative incidence and what must you do to make sure your calculation is correct
What are the limitations of using this:
Can also be called risk or incidence proportion
You have to follow up ALL participants
NO new participants should enter.
Limitations:
- People can drop out of study or die
- New participants may enter
what is the incidence rate? give formula
Number of new cases per unit of person time.
Ranges from 0 to infinity
Person time- a measure of the time spent in the study by participants. it starts when they enter the study until they get the disease, die or leave the study
what are the two types of standardisation? explain them
Direct standardisation- gives comparable incidence and allows us to adjust for differences in population
Indirect standardisation- this gives a ratio out of 100
You can standardise for age, sex, etc
For hospital deaths, what is the name of indirect standardisation used when comparing data
Standardise mortality ratio when it is adjusted for types of procedures
This is observed death/expected deaths using indirect standardisation
why are the reasons for a high death rate in a hospital? what data stats can you use to help mitigate these differences
Reasons are:
- Unwarranted variation
- Explained variation like higher amounts of high risks procedures
- Statistical artefacts- recording deaths
Indirect standardisation helps to account for these variations especially the standardised mortality ratio
Calculate and outline how you DIRECTLY adjust for age differences in a population.
explain the meaning behind direct standardisation
If you standardise for age and it goes DOWN then you’re population is OLDER.
Contrast aggregated data and person-level data
Aggregated data: for example, 5% of the population died.
Person-level data: for example, participants 1, 7 and 15 died.
Contrast primary and secondary data
Primary data
- Collected by researcher first-hand
- Collected for pre-specified purpose: test hypothesis
- Higher financial and time cost
- Will take up most of time and budget
Secondary data
- Data collected for another purpose and ‘recycled’
- Make have to make assumptions
- Can introduce critical limitations
- Usually faster and cheaper
in public health what are the two ways of collecting data?
Contrast them
Routinely collected data:
- Large administrative datasets
- Understanding and monitoring population/ local areas by using GP register or electoral register
- Prescribing data
- Hospital episodes stats like: outpatient, admissions etc
Non routinely collected data:
- E.g surveys
- very expensive
what is data linkages? what is it’s potential in public health?
Joining two or more datasets together
E.g. GP records and hospital records
it would be good for tracking disease progression, however we need to act lawfully and with transparency
what are the issues with data linkages in public health?
Technical issues
- don’t have the tech platform or solution necessary
Privacy
- Ethical and legal constraints on information governance
- Most patients find it strange
what are the features of ecological studies?
it doesn’t require data from individuals as it is a comparison of groups
You can only analyse aggregate level data that cannot be linked to a particular person
what are the issues with ecological data?
Ecological fallacy- when you assume group-level associations are applicable to individuals.
it uses secondary data sources
Only suitable when there’s little variation in outcomes between groups
You cannot tell if exposure preceded the outcome; you collected secondary data
what is the relationship between a statistic and a parameter?
Statistics are estimates of a parameter
Parameter- a fixed, unknown value which describes an entire population
what are the features of cross sectional studies
Snapshot in time
Hence no follow up, you can only assess the prevalence, NOT INCIDENCE
it is usually a survey
what is a seroprevalence study?
This is routinely used in epidemiological surveillance to understand the level of various infections at a population-level.
How is a case-control study conducted?
Case eligibility must be clear and cases are chosen from the target population
Case source could be hospital or community but it must be representative of everyone with disease under investigation.
Controls should come from same population and be representative of population at risk
Exposures should be measurable with similar accuracy (for both cases and controls)
Good for rare disease
what are the advantages and disadvantages of case-control studies?
Advantages of case-control studies:
- Good for studying rare diseases
- Relatively inexpensive to conduct
- Quick to obtain data (measure exposure and outcome at the same time
Disadvantages:
- Can be recall bias
- Often difficulty in selecting a control group
- Limited to assessing just one outcome
- Number of cases limited by rare disease
- Cannot provide information about temporal relationship between an exposure=re and disease
How do you mitigate against the low number of cases (due to the rarity) of a disease in a case-control study?
you can have more than one control per case?
How do you calculate odds ratio
(A/B)/ (C/D)
or
(AD)/ (BC)
Cohort studies/longitudinal studies
What type of evidence is it and what are the key features of it?
Most robust of the observational study
Defining characteristic is that you track people forward in time from exposure to outcome.
Always assess exposure prior to disease and track forward in time
what are the limitations of a cohort study
Cannot always keep a track of all participants for follow up
- May die or move away
- Population may be slightly smaller
How do you set up a cohort study?
Select target population
Assemble cohort
Assess exposure
Outcome ascertainement
How do you select a target population in a cohort study?
If you want to study chronic disease associated with ageing, you may restrict target popn
Exposure of interest may be rare so may need to target a specific population
Should initially attempt to identify as many subjects as possible without any restrictions to make findings generalisable
How may you assemble a cohort?
- By geographic región
- By occupation
- Based on disease
- Risk group
- Birth cohort
How do you assess exposure in a cohort study?
Often multiple exposures assessed
When analysis is conducted you want to be able to control for other exposures
May be sources for confounding
- Exposure must be well-defined
- Binary
- All exposed but different levels
- Eg obesity can be yes or no or categorised based on BM
- Variety of methods can be used
- Self report
- Taking physical measurements
- Using existing records
What are the methods of outcome ascertainment for cohort studies?
- Routine surveillance
- Death certificates
- Medical records
- Directly from the participant
Methods used must be the SAME for both exposed and unexposed
what stats can you collect from a cohort study and what outcome statistic cal you calculate
- Start with exposure in health subjects and check for outcome of interest
- Calculate incidence of disease in both groups
- Relative risk shows likelihood at developing disease if you are exposed relative to if you are not exposed
How do you calculate relative risk
Draw the table
Formular:
(A/A+B) divided by (C/C+D)
How do you deal with attrition in cohort studies?
This is unavoidable. It may not be an issue if attrition rates are low.
You must do two things:
- BE TRANSPARENT- report it in your write up
- BE conservative - if there’s the potential for the loss to bias away from the null hypothesis, then reporting the output statistic which biases towards the null is generally the more sensible course of action.
what are the features of retrospective/historical cohort study and what outcome statistics can you use for it?
- Group of individuals from cohort with a distribution of exposures and outcomes
- Measured contemporaneously or extracted from health records
- Typically lower quality than prospective cohort studies because there is a greater risk of both selection and information biases
- Odds ratio
when might Odds ratio be similar in value to Relative risk?
when the population is small therefore when there are rare diseases
what are the issues with observational studies?
Observed groups may differ in many characteristics in addition to the one being investigated
Clinical medicine puts most emphasis on robust experimental studies or clinical trials
what are the features of RCTs
Groups are assigned randomly
Choice of control is critical if you want to maximise the value of the RCT
The new treatment must be tested against the best current treatment available
Gold standard study design for evaluating the impact of treatment on an outcome
The process of randomisation alone doesn’t protect against bias.
what are the other potential sources of bias? and how do you mitigate
- Incorrect data analysis
- May reach an incorrect and biased assessment of results
- loss of data
- Loss of participants
Mitigate by :
- Intention-to-treat analysis -Analysing patients according to the group they were assigned-
- Look at consort diagram!
The randomisation process involves random allocation.
Which group(s) of people cannot determine allocation?
- Investigator
- Clinicians
- Participants
Allocation is not predictable
what are the overarching steps involved in the random allocation process?
Generation of allocation sequence
- Use random numbers table or a computer software program to carry this out
Implementation of allocation sequence using allocation concealment
what are the features of allocation concealment
Use of sequentially numbered opaque envelopes:
- Envelopes with treatment allocation opened by clinician on enrollment
- Must be opened in correct order and not in advanced
- Must ensure the envelope seal has not been opened
what are the potential problems with allocation concealment and how do you mitigate against it
It cannot be guaranteed that an opaque letter hasn’t been opened.
Therefore, If allocation can be seen investigators, they can preferentially assign treatment to patients they believe will do better
Use Central randomisation- a central randomisation service which issues treatment allocation
why is randomisation important in an RCT?
- Vital for preventing selection bias in the trial
- Ensuring integrity of randomisation process throughout trial is essential
what are the potential sources of bias in a trial
- Patient being treated
- Clinical staff administering treatment
- Physician assessing treatment
- Team interpreting results
why is blinding very important in an RCT?
- Used to prevent conscious or unconscious bias in the design and delivery of clinical trial
- Help to prevent withdrawals from study- pts who know they are receiving inferior treatments are more likely to drop out
what are the types of blinding in a trial and give what type of biases do they prevent?
Single blinding- only 1 party is blinded (participants)
Double blinding- - both participants and staff
- Help to reduce performance and detection bias
Triple blinding- not always possible
Contrast performance and detection bias
Performance bias refers to systematic differences between groups in care that is provided, or in exposure to factors other than the exposure of interest
- Eg if investigators know experimental group are given active drug they may focus attention to this group
Detection bias refers to systematic differences between groups in how outcomes are determined.
- May recieve more frequent exams and diagnostic tests
what are the potential limits to blinding
- Not always possible
- Eg surgical technique - this is impossible
- May be ethically impossible
- Makes things more costly
- Difficult if drugs require titrating
- If you need to increase of decrease the dose
what is the power of a study and give some features of it
Power of the study is ability to detect and effect of association if one truly exists.
it increases with sample size
A study with a power of 90% means that if a true association exists there’s only a 10% chance the study will not detect it.
what are the 3 main statistical factors that will push your sample size up or down?
The difference between the groups you’d be looking to investigate- a larger difference will require a smaller sample size
The study’s power- you should aim for at least 80%
The study’s alpha- Reducing alpha from 0.05 to 0.01 will increase sample size.
what is a study’s alpha
- How much you want to rule out chance causing a positive finding
- Equivalent of specifying the p-value and it’s also connected with one- and two-tailed testing
What % of participants are usually lost to follow up?
- Usually 15% loss of participants
- Longer studies have higher loss
You must also consider this loss when accounting for sample size
what are the two types of errors? give their features
Not random or systematic
- Type I error/ ‘False positive’ finding
- Can use the p-value to determine statistical significance
- Reduces the alpha
- However must bear in mind
- With multiple analyses one is likely to come back significant
- P-value can be mistaken as implying clinical significance
- Type II error
- ‘False negative” finding
- Can use statistical power- going from 80% to 90% power halves the risk of you deriving a false negative
what 2 things should you bear in mind when interpreting findings when considering p values
With p values between 0.01- 0.1, interpret data with caution. Unless you see multiple tests with a p-value of more than 0.05 or single tests of less than 0.01
Consider clinical significance
Contrast outcome statistics and outcome variable
Outcome variable- the variable you’re interested in (e.g diabetes, step count)
Outcome statistic – the number you got after interpreting; odd ratios, relative risk
what is a study’s beta
The probability of type 2 error.
The probability of not detecting a difference when one actually exists.
Power= (1 - beta)
Most medical literature has a beta cut off of 20%
how does the effect of interest affect sample size
Increasing the effect of interest decreases the sample.
We use literature reviews when we want to give an opinion on topics we are not expert in
Contrast the two main approaches to literature review
Narrative review
- Brings together published literature into a single article
- Enables reader to rapidly understand issues
Systematic review
- Similar to narrative
- Sets out highly structured approach to searching, sifting, including and summarising the literature
- Often presented as the basis for meta-analysis, but also exists separately
What are the strengths of a narrative review
- Agile
- Easier and dater to write
- More up to date than systematic reviews
- Particularly useful when looking at areas with with limited research or higher levels of variation in research approaches
- Useful in interdisciplinary work
what are the limitations of a narrative review?
Narrative review have lost ground to systematic reviews recently
Subject to potential bias:
- Author can select work that make support their opinion
- Ethical authors must prevent overly-speculative or unbiased reviews
- No search is specified, so important evidence may be omitted
what are the strengths of a systematic review?
- Aims to collate all available evidence that relates to a highly focused research question
- Implements a highly specified protocol that enables reproducibility
- ‘Includes’ evidence based on pre-specified criteria: inclusion criteria
Can take months to design the search and synthesis findings
What are the limitations of a systematic review
- Can take sometimes 18-24 months from start to publication
- Only as good as the method employed
- Only as good as indices searched
- Only as good as the evidence that is incorporates
- Very quickly out of date:
- Look at the search date not publication
- By the time article is published the data may be 18-24 months old
- Only published every 2-10 years
what are the steps of a systematic review
what does step 1 and step 2 of systematic review protocol entail
- Research question
- Narrowly defined
- Structured search
- Works to develop a series of searches with justification
- Think about operators you can use
what does step 3 of a systematic review entail? what does it prevent
Indices
- Eg, medline, mbase, psycinfo, ovid search - based on published research
Must be specific to allow reproducibility and transparency
Registries - registration of research yet to be completed or published
- Prevents of duplication, omission or publication bias
Describe how you would carry out step 4 of the systematic review protocol
What does it involve and what does it allow you to do
Step 4 - Screening
use PRISMA flow diagram
- Shows how many articles have been found through search
- How many have been removed as duplicates
- Screening process of titles and then text
- Determine how many studies will be included
- Shows the ‘n’ at each stage
how many papers will normally pass the screening process of a systematic review
and what should you do to systematise the screening process
- Often systematic reviews will identify 1000s but only end up with 10-30
- Applying inclusion and exclusion criteria is important to systematise the process
- Allows a comprehensive review
Give the features of grey literature
Not everything that is known is published literature
- Eg government reports, evaluation of processes
Google scholar/opengrey
- Take a more holistic view of what is considered evidence
Typically not peer-reviewed; evidence is not robust
Rapid understanding
How do you make decisions when so many papers are published everyday?
Start with the Cochrane Collaboration- for medicine
- Narrative and systematic review are helpful as a starting point
- Can snow-ball out from reviews to find individual literature
- Can look to see more recent citations of these reviews using online indices
- Must be careful about timeliness
what is a meta analysis ?
quantitative, formal epidemiological study design
used to systematically assess previous research studies
used to derive conclusions about that body of research
Can use forest plot to pool estimates of association
look at the forest plot below, describe in detail what it shows.
First column shows identities of studies
2nd and 3rd show number of participants in studies involved
4th column shows forest plot
- Midline shows line of no effect
- Rows show estimates of relative risk presented as a square
- Square is proportional to number of people included in study
- Whiskers show the level of confidence/uncertainty around these estimates
- With bigger squares, there is lower uncertainty
- Position of squares to either side of the line of no effect indicates effect
- Total (pooled) estimate is represented by a diamond
- This combines power of previous studies and left and right angles show uncertainty
5th column shows weight of each study on the pooled estimate
6th column shows point estimate as a numerical value
- MAKE SURE TO ASSESS THE SCALE
what does meta analysis require?
Meta-analysis requires that the pooled studies are sufficiently similar: or else your results are meaningless
what are other considerations that can affect your systematic review/meta-analysis
Heterogeneity
Weighting
Publication bias
What is Heterogeneity and give sources of it.
How do you interpret it
‘Difference between studies that are included’ -Only QUANTITATIVE analysis
Sources:
- Clinical: patients, selection criteria
- Methodological: study design, blinding, intervention approach
- Statistical: reporting differences
There are statistical ways of evaluating the heterogeneity between studies
Degree of heterogeneity must be conveyed, and how this may affect the results.
- if I2 (heterogeneity) is less than 50% then use fixed effect model for analysis becuase it is considered homogenous
- if more than 50% then use random effect model
what is weighting (systematic review) and give features of it
- Mostly proportional to study size
- We can build in assumptions to account for differences and similarities
- Fixed effects
- Random effects
What is publication bias and how can you mitigate it
Studies with positive findings are more likely to be submitted for publication
Publication funnels plot can show balance of evidence between studies assuming an overall effect size
- Assess likelihood of publication bias
- More publications on one side of the line may suggest that some studies may not have been reported
what are the trial endpoints
An outcome that usually describes a clinically meaningful outcome.
Measure of efficacy
what are the clinical trial outcomes/goals?
Efficacy – how well a therapy works in achieving the desired outcome
Safety – how well a therapy works in not causing adverse events.
Each have their own endpoints
what are the efficacy endpoints? give features of them
Primary endpoint –
- endpoint for which the study has been powered
- number of trial participants (sample size) will have been recruited on the basis of the pre-specified power and difference.
Secondary endpoint –
- slightly different endpoint in addition to the primary endpoint
- Eg. a study seeks to examine survival – may record often ‘softer’ - measures such as recurrence of disease or hospital admission might also be measured
if 20 endpoint is achieved but not primary then the results may contribute to understanding of the disease
what are the safety endpoints
More intuitive
could be anaphylaxis or direct mortality associated with the therapy (rare)
- Such major issues should usually be detected early in the trial process
- Commonly, the safety endpoints will be more nuanced: potentially measuring commonly observed adverse events (AEs) and grading them into a hierarchy of significance
- A large proportion of patients reporting AEs will require investigation
what are composite endpoints
- Multiple potential endpoints
- Common when the outcome is uncommon
- Eg combine MI and stroke to give composite outcome ‘cardiovascular event’
Give features of survival analysis
- Interested particularly in survival with novel cancer therapies
- Survival analysis is used to get around issue of losing precision by generating arbitrary timepoints
Flat line indicate censorship hence no follow up continued. We don’t count it!
