Lectures 11-13 Flashcards
Purpose of controlling confounder
- To get a more precise (accurate) truer measure of association between exposure and outcome
Ways to control confounder
Study design stage: - Randomization - Restriction - Matching Analysis of data stage: - Stratification (w/ weighting) - Multivariate statistical analysis (regression analysis)
Restriction
o Study participation is restricted to only subjects who do not fall within pre-specified category(-ies) of confounder
• Strength:
o Straight forward, convenient and inexpensive
o Does not negatively impact Internal Validity
• Weakness:
o Sufficiently narrow restriction criteria may negatively impact ability to enroll subjects (reduced sample size (N))
o If restriction criteria is not sufficiently narrow it will allow the introduction of residual confounding effects
o Eliminates researchers ability to evaluate varying levels of the factor being excluded
o Can negatively impact External Validity (Generalizability)
Randomization
o Randomization technique hopefully allocates an equal number of subjects with the known (and unknown) confounders into each intervention group
• Strength:
o With sufficient sample size (N), randomization will likely be successful in serving its purpose (making groups “equal”)
o Stratified version more precisely assures equalness
• Weakness:
o Sample size (N) may not be large enough to control for all known and unknown confounders
o Process doesn’t guarantee successful, equal allocation between all intervention groups for all known and unknown confounders
o Practical only for Interventional studies
Matching
o Study subjects selected in matched-pairs related to the confounding variable to equally distribute confounder among each study group
• Strength:
o Intuitive, some feel it gives greater analytic efficiency
• Weakness:
o Difficult to accomplish, very time consuming, and potentially expensive
o Doesn’t control for any confounders other than those matched on
• Over-matching possible; this will mask findings
Stratification
o Descriptive & Statistical analysis of data evaluating
association between Exposure and Outcome within the
various strata (layers) within the confounding
variable(s) [Young vs. Old; in Smoking & CHD example)
• Strength:
o Intuitive (to some), straight-forward and enhances understanding of data
• Weakness:
o Impractical for simultaneous control of multiple confounders, especially those with multiple strata within each variable being controlled
Multivariate Analysis
o Statistical analysis of data by mathematically factoring
out the effects of the confounding variable(s)
• Strength:
o Can simultaneously control for multiple confounding variables
o In Regressions, beta coefficients can be converted to OR’s
• Weakness:
o Process requires all individuals to clearly understand (interpret) the data (results)
o Can be very time consuming for researcher/biostatistician
Bias
- Systematic (non-random) error in study design or conduct leading to erroneous results
- Nothing can be done to “fix” bias once it has already occurred (study end)
3 elements of bias
o Source/Type (2 main categories)
o Magnitude/Strength
• Bias can account entirely for a weak association (a small RR/OR) but is not likely to account entirely for a very strong association (a large RR/OR)
o Direction
• Bias can over- or under-estimate the true measure of association
o Bias can have a enhancing or minimizing effect on the true measure of association (towards or away from 1.0)
2 main categories of source/type bias
‘Measurement’-related (Information/Observation):
o Any aspect in the way the researcher collects information, or measures/observes subjects which creates a systematic difference between groups
• Errors in measurement can also cause a resultant error in patient classification
‘Selection’-related:
o Any aspect in the way the researcher selects or acquires study subjects which creates a systematic difference between groups
• Commonly seen when comparative groups not coming from same population/group or not being representative of the full population or even differentially selected (processes)
Types of selection bias
Selection Bias:
o They way study subjects are selected generates differences in groups (very commonly encountered)
o Key Examples:
• Healthy-Worker bias
o Can easily be seen in prospective Cohort studies, workers usually exhibit lower death rates compared to chronically ill who do not work because of illness
• Self-Selection/Participant (Responder) bias
o Those that wish to participate (volunteer) may be different in some way to those that don’t volunteer or self-select (refusal/nonresponse) to participate
• Control Selection bias
o Can easily be seen in Case-Control studies
Bias in cross-sectional studies
Cross-Sectional studies are subject to Neyman bias
(a.k.a., selective survival)
o More easily descriptive for longer-lasting and more
indolent diseases
o Not effective for acute or rapidly fatal diseases
Subject related bias
- Recall (reporting bias)
- Contamination bias
- Lost to follow-up bias
Recall (reporting bias)
A differential level of accuracy/detail in provided
information between study groups
• Exposed or diseased subjects may have greater
sensitivity for recalling their history (better memory;
easier to remember if more severe) or amplify (exaggerate) their responses
• Individuals can report their “effects” of exposure,
disease symptoms or treatment differently because
they are part of a study
o “Hawthorne Effect”
Contamination bias
• Members of the control group accidently, or outside of the study protocol, receive the treatment (or similar) or are exposed to the intervention being studied
