What are the pros and cons of UK Biobank?
Advantages
- Large sample size
- Standardised data collection
- Lots of information about genetics, environmental and lifestyle
- Longitudinal
Disadvantages
- Low response rate (around 5%) resulting in selection bias
- Healthy volunteer bias
- Limited ethnic diversity and mainly middle aged individuals
- Ethical concerns about vague consent
What are 3 key advantages of combining studies in a systematic review or meta-analysis?
- Greater power and precision
- Greater generalisability
- Efficiency and cost savings
What are the 3 key steps in a systematic review?
- Literature review
-Critical appraisal - Amalgamation (qualitative and/or quantitative)
What is a systematic review?
It is the review of a clearly formulated question that uses systematic and explicit methods to identify, select and critically appraise research and to collect and analyse data from studies
Advantages and disadvantages of a systematic review
Advantages
- Systhesises evidence
- increases sample size of available data, increasing power and precision of summary estimate
- Highly influential with policy makers
- Reasons for heterogeneity can be identified and new hypotheses generated about specific subgroups
- Explicit methods applied limit bias in identifying and rejecting studies
- Helps define what is known and unknown and helps formulate hypotheses for further investigation
Disadvantages
- Knowledge of an average treatment effect may not apply to an individual patient
- Important to ensure methods are valid and reliable
- Inappropriate aggregation of studies that differ in terms of intervention used or patients included can lead to the drowning of important effects
- Findings of SR may not be in harmony with findings of large scale clinical trial
What is a meta-analysis? And what type of results do you get from them?
- Meta-analysis is the quantitative amalgamation of the findings from the studies included in a systematic review
- Summary estimate: weighted average of the effect sizes in each study included. Weight is usually based on sample size
- Forest plot: visual representation of the results of the individual studies and the summary estimate, including the confidence intervals
What two key issues lead to bias in a meta-analysis?
- Poor study quality
- Publication bias
What are the two main approaches to meta-analysis?
- Fixed effects meta-analysis: can only use when there is no evidence of heterogeneity. Assumes all variation between studies is due to sampling variation. Narrower confidence intervals and smaller p value
- Random effects: can be used when there is heterogeneity. Assumes treatment effect follows a distribution whose mean is overall effect. Wider confidence interval and larger p value.
Key feature of case-control study and strengths and weaknesses
- A group of individuals with the outcome of interest (cases) are compared to a group without the outcome (controls). Ideally this is 1 to 1 but can increase number of controls up to 1:4. Retrospective, can be matched for other features.
Strengths:
- Rapid and cheap
- Efficiency for rare diseases and those with long latency periods.
- Can look at multiple exposures
Weaknesses:
- Can only look at one outcome
- Prone to selection and recall bias
- Bad for rare exposures
- Temporal relationships difficult to establish
Cross-sectional
- Determines the simultaneous prevalence of exposure and outcome
Strengths:
- Rapid and cheap
- Can study multiple exposure and outcomes
- Prevalence can be measured, useful for looking at burden of disease
- Good for descriptive analysis and generating hypotheses
Weaknesses:
- Recall and response bias
- Unable to measure incidence, can’t assess factors associated with survival
- Can’t establish temporality
Cohort studies
- A group without the outcome chosen according to exposure. Prospective or retrospective. Example is Framingham study following residents in Masechussets town for cardiovascular outcomes
Strengths:
- Allows direct measure of incidence
- Useful for rare exposures
- Can investigate multiple outcomes
Weaknesses:
- Expensive and time consuming
- Not good for long latency diseases or very rare diseases
- Bias from loss to follow up
RCTs
- Where investigators allocate the exposure randomly
Strengths:
- Gold standard for causality
- Powerful tool for controlling confounding
- Enables blinding and therefore minimises bias
- Can measure disease incidence and multiple outcomes
Weaknesses:
- Ethical constraints
- Expensive and time consuming
- Generalisability
Time-series
- both exposure and outcome are measured over time.
Strengths:
- Can identify time trends and serial variation
- Can be used to forecast future trends
Weaknesses:
- Migration of populations between groups during study period may dilute the difference between groups
- Routine data sources may have been collected for other purposes
- Regression to the mean (refers to the tendency for extreme values in a dataset—whether unusually high or low—to be followed by values that are closer to the average over time. This happens purely due to random variation)
Delphi
- Rounds of anonymised questionnaires of experts to develop a consensus on complex or uncertain issues
Strengths:
- Rapid consensus can be achieved without the risk of Groupthink
- Low cost
Weaknesses:
- Does not cope well with widely differing opinions
- Can be time consuming
- Success depends on quality of participants
Cluster RCT
- Randomised by groups rather than individuals, but individual outcomes measured
When would you use them?
- Interventions implemented at group level e.g group counselling
- Normal randomisation would lead to “contamination”
Strengths:
- Useful when intervention targeted at group level
- Can be easier for staff to only deliver one intervention
- Useful when group could influence eachother
Weaknesses:
- Intra-cluster correlation therefore need larger sample size, therefore more expensive
- Increased risk of imbalance in baseline characteristics
Statistical issue
- Sample size calculation needs to be inflated appropriately to account for intra-cluster correlation (called the design effect)
- Involves quantifying the homogeneity of response between individuals in a cluster - termed the intraclass correlation coefficient (ICC)
- Can use robust standard errors
How do you calculate power?
This is the probability that you will be able to detect a difference if one truly exists in the population. Power is 1 minus type 2 error rate (1- beta). Influenced by:
- Effect size (larger effect size, higher the power)
- Sample size (larger sample, higher power)
- Alpha level usually set at 0.05
