Research Design, Biostatistics, and Literature Evaluation
Observation bias: Observers (research team) are aware of the research purpose and allow this
knowledge to influence interpretation of results.
Recall bias: Methodological error that is introduced in survey research when the participant is asked
to provide recall of a past event
| d. | Misclassification bias: Inappropriately categorizing a group of patients with, or without, the disease/ |
|---|
syndrome
Immortal time bias: Occurs when populations studied include an exposed and unexposed group
without a predefined time-zero. This results in a delay before a subject is considered treated.
Given that the subjects must survive up until the time of first exposure, the time before exposure is
Measurement bias: When information collected for use as a study variable is inaccurate
Confounding variables: Extraneous variables that influence both the dependent and the independent
variable, affecting how the overall result can be interpreted
Residual confounding occurs when unmeasured variables may affect overall outcome results.
This can only be avoided with randomization (Int J Public Health 2010;55:701-3).
ii.
Confounding by indication occurs when a contributor to the outcome is present in those at
high risk and is an indication for intervention. This results in care differences between the
exposed and nonexposed groups that may be based on differences in indication for exposure
Bias can be minimized with sufficient planning for design, data collection, and analysis (i.e., βeffect
modificationβ).
Sensitivity analysis: These may be used to determine robustness of study findings in different analysis
methods or in the presence of unmeasured confounding. They may also be used to detect misclassification
Approaches to control for bias:
Study design: Randomization, matching
Propensity score matching
| (a) | This method in theory allows for design and analysis of observational studies to mimic |
|---|
aspects of a randomized controlled trial.
| (b) | The propensity score itself distributes observed baseline covariates similarly between |
|---|
treated and untreated subjects to create a certain number of matched sets.
| (c) | Caution should be exercised because propensity scores ensure balance only in observed |
|---|
covariates, where randomization ensures balance between both observed and unobserved
covariates.
Analysis: Multivariable analysis, matching potential confounding variables with a controlled cohort
(discussed in detail in next section)