Bias Flashcards
Bias
Factor which causes SYSTEMATIC over- or under- estimate of a particular result
Types of biases that can occur in the designing of the study?
Selection bias, volunteer bias, channelling bias
Selection bias
Participants in research may differ systematically from the population of interest.
Biases that can occur when conducting the research?
Interview bias, recall bias, Hawthorne bias
Biases that can occur in writing and publishing the study?
Publication bias, confounding
What is confounding bias…?
Confounding is when you have an independent factor which is associated with both the exposure factor and the outcome of interest with the disease
How can you minimise confounding factors?
Minimise through stratification, regression analysis, randomisation
What is internal validity and external validity?
Internal validity - accuracy of conclusions
External validity - generalisability of results
Membership bias:
Explanation: Included patients are already participating in a study so might be more likely to look after their health, see the benefits of research, and educate themselves about treatment.
Neyman/survival bias
Explanation of an example: Included patients who had to be alive months after the treatment, out-selecting those with more rapidly evolving conditions who may have realistic idea of the treatment effects
Examples of confounders:
Age, ethnicity, sex, comorbidities (DM, alcohol intake), past surgery, previous family history, drug history, etc.
Diagnostic purity bias
Comorbidity is excluded in the sample population, such that it does not reflect the true complexity of cases in the population
Neyman bias (survival bias)
There is a time gap between the onset of a condition and the selection of the study population, such that some individuals with the condition are not available for selection.
Response bias
Individuals volunteer for studies but they differ some way from the target population e.g. they are more motivated to improve health
Lead-time bias
Screening/testing increases the perceived survival time without affecting the course of the disease
Publication bias
A study that shows a significant difference between two interventions is more likely to be published than a negative study
Observation bias
There is a problem with the way data is collected in the study, such that data has been unduly influenced by the expectations of the researchers and subjects
Response bias
The subject answers questions in the way in which they think the researcher wants them to answer e.g. subjects in the experimental arm are more likely to give favourable responses
Recall bias
Subjects selectively remember details from the past
Exclusion bias
Results from differences in dropout rates between groups
Hawthorne effect
Subjects alter their behaviour because they are aware they are being observed in a study
Performance bias
Differences in care provided aside from intervention
Detection bias
Differences in how outcomes are assessed between groups
Aggregation bias
Occurs when it is wrongly assumed that the trends seen in aggregated data also apply to individual data points
What is a method to assess the potential role of publication bias?
Funnel plot
Ecological fallacy
The failure in reasoning that arises when an inference is made about an individual based on aggregate data for a group. This is an association for aggregate data in which the unit of observation is the country.
Strategies to reduce confounding
Randomisation, restriction, matching, stratification, adjustment and multivariate analysis
How does randomisation reduce confounding?
Aims to randomly distribute confounders between study groups
How does restriction reduce confounding?
Restricts entry of individuals with confounding factors (risks bias in itself)
How does matching reduce confounding?
Matching of individuals/groups aim for equal distribution of confounders
How does stratification reduce confounding?
Confounders are distributed evenly within each stratum. Stratification allows the association between exposure and outcome to be examined within different strata of the confounding variable. This technique allows observation effects of an intervention in different subgroups (however, this becomes more difficult if there are several confounders). Creating several smaller groups reduces the power of the study.
How does adjustment reduce confounding?
Usually distorted by choice of standard
How does multivariate analysis reduce confounding?
Only works if you can identify and measure the confounders. This analysis is a statistical procedure for analysis of data involving more than one type of observation. Multivariate regression is an extension of multiple regression with one dependent variable and multiple independent variables.
What are the advantages of multivariate analysis
Limited loss of power and ease of combining several confounders in one study
