Enhancing the Robustness of Observational Studies Flashcards
Statistically significant but FALSE association due to:
Systemic error (bias)
Other relationships
- Must consider plausibility
Random error
Statistical fluctuations related to precision of measurements
How to minimize error
Take multiple measurements and average them
Increase the same size
Non-random error
“Any tendency which prevents unprejudiced consideration of a question”
Bias prior to implementation
Design or Selection Bias
Define Design Bias
Selecting extremes leads to regression to the mean
Define Selection bias
Sampling and exposure are linked
Medical surveillance bias (unmasking bias)
Berkson bias (source pop different from general pop)
Channeling bias (prefer use of one treatment over another)
How to minimize Selection bias
Matched case-controls Prospective ascertainment of bases with blinding Stratification Population-based controls Avoid convenience Statistics
Match cases and controls
Match on factors associated with both disease and exposure but NOT with cause
During implementation bias
Interviewer Chronology Recall Transfer Measurement Performance
Define Interviewer bias
Leading or unanswerable questions
Min: standardize interactions, interviewer blinding
Define Responder bias
Recall (can’t remember), overstatement, consistency(how they answered previous questions influence next), and acceptance (fit expectations) bias
Minimize responder bias
Prospective design, use objective measurements
Define Chronology bias
Changes in clinical practice and treatment guidelines over time
Minimize Chronology bias
Prospective design
Avoid historical controls
Define Transfer Bias
Unequal attrition and hence unequal information deficiencies
Minimize transfer bias
A priori plan for pts lost to followup
Minimize misclassification of exposure/outcome
A priori definitions of exposure/outcome
Use objective measures
Prospective design
Define Performance bias
Systematic differences in care/exposure factors between groups
Minimize performance bias
Cluster stratify
After implementation bias
Publication
Citation
Data analysis (confounding factors)
Confounding Factors
Associated with both exposure and outcome
If uncontrolled, can affect validity of results
Control Confounding factors
Matching Restriction Stratification Modelling using multivariate tech Randomization
Internal Validity
Did you measure what you set out to measure
External validity
Is what you measured in your subjects representative of the real population?
Criteria for evaluating causality
Temporal sequence Strength, consistency, specificity of association Biological gradient and plausibility Coherence with existing knowledge base Experimental evidence Analogy
