MDM - Biases Flashcards
goals of epidemiological and clinical research
identify true effects of putative causal factors
obtain valid epidemiological measures or causal inference
reduce both random and systematic (bias) error through study design, subject selection, information collection and classification, and data analysis
biases affecting external validity
generalizability - has to do with the publication and application of research findings
- publication bias
- spectrum bias
- random error
- systematic error (bias)
biases affecting internal viability (systematic error)
has to do with the study design, data collection, and statistical analyses of biomedical research
- selection bias
- misclassification bias
- confounding bias
publication bias
trials reporting statistically significant positive findings are more liekly to be published and be published faster than those that report negative findings
spectrum bias
occurs when diagnostic test performance varies across patient subgroups and a study that tests performance does not adequately represent all subgropus
also occurs when a results from clinical trials vary among subgroups
random error
occurs from subject sampling variation and is limited by increasing sample size
systematic error (bias)
occurs if there is a difference between what is studied and actually estimating what it is intended to measure
would be present even if it is an infinitely large study, due to study design or analysis
selection bias
stems from the procedures used to select subjects and from factors that influence their participation
occurs when comparison groups differ because of the selection or sampling process
occurs when disease or exposure status influence participation of subjects to a different extent in compared groups
most often occurs in cohort studies with variable lost to follow up and in case-control studies when the exposure influences the selection of case or controls
nonparticipants are often different form participants
presence must usually be inferred, rather than observed
efforts should be made to prevent selection bias rather than adjust for it
mis-classification bias
arises because information collected about or from study subjects is erroneous
also known as information bias
bias in the effect estimation resulting from exposure or disease misclassification
confounding bias
inherent differences in risk between exposure groups that distorts the estimate of effect
assessment criteria for biases regarding systematic error
presence, direction, and magnitude
When does selection bias most often occur?
most often occurs with variable lost to follow-up in cohort studies, improper selection of controls in case-control studies, improper use of ris factor to identify cases
non-differential misclassification error
proportion of subjects mis-classified is the same for comparison groups, but does not depend on other study variables
differential misclassification error
proportion of subjects misclassified is different for comparision (exposure or outcome) groups and depends on other variables in the study
variable measure - precision
degree to which a variable has nearly the same value when measured repeatedly
assess by comparison of repeated meausres
improves ability to detect differences
also known as measurement reproducibility
affected by random error contributed by the observer, subjects, or instruments
variable measurement - accuracy
degree to which a variable actually represents what it is supposed to represent
also known as measurement validity
assessed by comparing with a referenced standard (gold standard)
affected by systematic error contributed by the observer, subject or instrument
measurement accuracy components
criterion validity, content validity, face validity, construct validity
criterion validity
extend to which the results of a measure or test agree with another gold standard criterion
content validity
extent to which a composite measure includes all the important aspects or domains regarding the theoretical construct of interest
face validity
extent to which a single item or test is judged to reflect the construct of interest
construct validity
the extent to which the measures or tests agree with other measures that are consistent with theoretically derived hypotheses concerning the construct of interest
experimental demonstration that a test is measuring the construct it claims to be measuring
the extent to which a test or procedure appears to measure a higher order, inferred theoretical construct, or train in contrast to measuring a more limited, specific dimensions
strategies for enhancing measurement precision
standaridzation, training and certifying, refining, automating, repitition
strategies for enhancing measurement accuracy
unobtrusive measurements, binding observers and subjects, calibration of the instrument
confounding bias
bias due to an invalid comparison of exposure groups
inherent differences int he risk between exposed and unexposed
the estimate of effect does not equal the true casual parameter in the source population
controlled by several methods
controlling confounders
Randomization – random allocation of subjects into treatment groups.
Restriction – selecting subjects who have the same value for the confounder.
Matching – prohibit the confounder from varying by pairing comparison groups.
Stratified Analysis – analyze data within categories of the confounder.
Mathematical Modeling – fit data into a model (e.g., logistic regression).
Special Advanced Methods – propensity scores and instrumental variables.
residual confounding
Occurs when adjustment does completely remove confounding from a given variable or set of variables:
Improper definition of the categories (typically too broad).
Imperfect surrogate for the confounding characteristic.
Confounder is not included in the study or analysis.
Misclassification of the confounder.
confounding by indication
Often seen in a non-randomized comparison of drugs or treatments.
Those given one intervention are inherently different than those receiving no intervention or another intervention.
Typically, there are differences in disease severity or other risk factors between subjects receiving different interventions.
The indication is a confounder because it is associated with the intervention and is a risk factor (or indicator) for the outcome.
