ep test 2d
Terms
undefined, object
copy deck
- Bias
- systematic error in a study which results in an incorrect estimate of the association between exposure and disease
- Types of Bias
- Two broad classification of bias: Selection bias Observation (information) bias
- Selection Bias
- a systematic error in the process of identifying and recruiting members of the study sample related to either outcome or exposure does NOT relate to sample representativeness or generalizability
- Sources of Selection Bias
- Differential diagnosis, surveillance or referral into the study Differential refusal or non-response rates among subjects by either disease or exposure status
- Differential Diag
- Particularly a problem for facility-based research persons who have a risk factor may be tested and diagnosed at a faster rate, and be in care at a different rate, than those who do not have the risk
- Differential Surveillance
- Particularly a problem when using passively reported data persons who are treated by different care providers may have differential rates of disease reporting
- Differential Referral
- Recruitment of subjects from sources (service providers) that are not accessed equally by those at risk or with disease recruiting cases solely from private physician offices, while controls are from private and public facilities
- Differential Refusal or non-response
- Subject willingness to participate may be influenced by either their disease or exposure status persons asked to be controls in a case-control study of behavioral risks for HIV may be unwilling to participate because they do not wish to reveal risk factors
- Observation (info) Bias
- systematic error in the collection or measurement of data related to either exposure or outcome
- Misclassification
- when the data gathered does not actually reflect the true/real disease or exposure status random (non-differential) is not a serious problem, although it can dilute your measured association non-random (differential) can distort observed relationship and conclusions relative to true situation
- Recall (obs) Bias
- Recall Bias patient characteristics may influence data gathered - age, lifestyle, attention to detail women who are experiencing uterine CA may, because of the disease, spend more time remembering their past reproductive history (not as much hopefully) than those without disease
- Interviewer (obs) Bias
- Interviewer Bias interviewer actions may influence data gathered a chart abstractor may be more aggressive in seeking risk information in the chart for cases than controls, or ask probing questions when they get a negative response in an interview
- Follow-up (obs) Bias
- Follow-up Bias ability to gather data over time varies by exposure and outcome tracking subjects may be influenced by exposure status - workers exposed to asbestos may be more likely to remain in study and be diagnosed than unexposed workers who drop out but still develop disease
- Control of Bias
- Careful study design and study implementation are the only ways to control for selection bias, and for many forms of observation bias
- Methods to Control Bias
- Choice of study subjects thinking of as many problems as possible before diving into the process establishing clear inclusion/exclusion criteria for all subjects defining a strict recruitment protocol Choice of study subjects Consistency in sources of disease and exposure information defining if and how multiple sources will be used which source takes precedence if there is a conflict Mechanisms in data collection objective measures and standard protocols training of study personnel “blinding†of researchers and subjects
- Specific Designs & Their Impt Bias Threats
- Case-control selection and recall bias Retrospective cohort recall and selection bias Prospective cohort interviewer and follow-up bias Randomized controlled trial interviewer and follow-up bias
- Confounding
- The mixing of the effect of the exposure upon disease with the effect of a second factor that is related to both the exposure and the disease Essentially, real relationships that create “noise†in the study
- Confounding details
- Confounders are: predictive, but not necessarily causal for disease associated, but not necessarily causally related, with the measured exposure not an intermediate link in the causal chain or web between the disease and exposure
- Confounding (ex)
- In a study of the contribution of air pollution to occurrence of bronchitis (can be contagious), urban density or crowding acts as a confounder In a New Mexico study of CHD mortality, altitude of residence appeared protective, but was in fact confounded by ethnicity
- ID-ing Confounding
- Potential confounders are identified: NOT by levels of significance of its relationship to either disease or exposure But rather by: knowledge of the disease previous investigations of similar questions and populations by testing for differences in crude and adjusted rates and RR/OR in the study sample
- Confounding effects
- The effect of confounding can be either: Positive - enhancing the apparent association observed; e.g. a true RR of 4.5 is estimated as 6.7 Negative - diluting the association and bringing it closer to 1.0 (no effect)
- Control of Confounding
- Confounding can be controlled at the design phase through: Randomization of subjects in clinical trials Restriction of sample Matching* Confounders are controlled at the analysis phase by: Matched analysis* Stratification Multivariate analysis
- Randomization
- Not possible in observational studies; Only way to control for unknown confounders; Helps ensure that potential confounders (known and unknown) are evenly distributed in your study sub-groups; Sample size must be sufficiently large for randomization to be effective
- Restriction
- reduces size of pool of potential study subjects; controls only known confounders; does not allow assessment of the association at varying levels of the confounding variable; reduces generalizability
- Matching
- Must be addressed at design AND analysis Can control only a limited number of confounders Potentially expensive in $$ and time Matched factor cannot be tested as to its effect on the outcome Useful strategy for environmental / genetic factors Useful when number of cases is small and multiple controls can be matched
- Types of Matching
- Frequency Proportion of sample in each level of a confounder in one study arm (usually exposure or case) is determined, and then the same frequency distribution is sought for the other study arm Individual each case is matched individually with a specific control on the confounder(s) (charts)
- Odds Ratio for matched case-control study
- OR = b/c
- Stratified analysis
- Magnitude of confounding is determined by the difference between crude and adjusted risk measures Calculates unconfounded estimates of risk within each level of potential confounder Evaluates both confounding and effect modification
- Adjusted or Standardized OR & RR
- There are a number of different formulas and formula choice depends upon: study design whether stratum-specific estimates are uniform or non-uniform whether incidence is in CI or ID format
- Effect Modification
- A variable is an effect modifier when the association between the exposure of interest and the disease varies with the level of that variable also called an “interaction†Don’t worry about material in text that focuses in depth on effect modification (charts)
- Multivariate analysis
- Allows for simultaneous control of multiple confounders Choice of model is a complex process, linked to design used and purpose of study Reliance on analysis packages is often detrimental to real understanding of dynamics of relationships