Epidemiology #3
Bias in estimating frequency
-Due to: -The misrepresentation of the population in the study sample (Sampling methods are important-Selection) -An error in the measurement of the true state of individual characteristics. (Validating your measurement tools is important- Information bias)
Selection Bias
-Errors due to the systematic differences in relevant characteristics between those who are included in the study and those who are not. -Introduced by: - The investigator (how you choose subjects) - Personal choices of individual (self-selection bias) - How persons with disease are diagnosed in the community (exposure suspicion bias)
Stratification
-Evaluating the association between exposure and disease within homogenous categories (strata) of the confounding variable. -Advantages: -Allows for clear understanding of interrelationship among exposure, disease and confounding variables. -Direct and logical strategy; computations easy to carry out -Minimal assumptions required for analysis to be appropriate -Permits the evaluation of effect modification -Disadvantages: -Inability to control simultaneously for several confounders -Continuous variables must be categorized for analysis
Length Biased Sampling
-Less aggressive forms a of a disease are more likely to be picked up by screening because it has a longer per-clinical phase -Less aggressive forms of disease usually have better survival -thus screening detected cases appear to have better survival
Validity
-The ability to distinguish between who has the disease and who does not
Hill's Criteria 3. Consistency of the association
- The association is consistent when replicated in other studies with different population and settings
Hill's Criteria 1. Temporal sequence
- The exposure must proceed the disease (by a reasonable amount of time) -Complete agreement by epidemiologists as important step -Easiest to establish in a prospective cohort study
Hill's Criteria 5. Biological plausibility
- The hypothesized effect makes sense in the context of current biological knowledge
Calculating Specificity
- True negatives/all non-diseased (d/b+d)
Confounding
-A distortion of the exposure-disease association due to the influence of a third factor. -A confounder may fully or partially account for the observed effect of the study exposure or mask or hide an underlying true association. -Results when another factor is unevenly distributed between comparison groups and may account for the observed association.
Bias
-A systematic departure from the truth (Systematic Error) -any systematic error in the design, conduct, or analysis of a study that results in a mistaken estimate of an exposure's association with disease
Reliability of a screening test
-Ability of test to give consistent results when the test is replicated.
Predictive Value
-Ability to predict presence/absence of disease dependent on: SENSITIVITY, SPECIFICITY, PREVALENCE (Lower prevalence=lower PPV and higher prevalence=higher PPV)
Types of Association
-Artifactual (Spurious) -Non-Causal -Causal-Direct -Causal-Indirect
Causation
-Association does NOT equal causation -Cannot be proven by one study -It's INFERRED using all available information
Indirect causal association
-Association of a factor (C) with disease (A) by means of an intermediate or intervening factor (B) Factor C -> Factor B -> Disease A
Association vs. Causation
-Association: observed -Causation: Inferred
Non-causal association
-Both risk factor and disease are associated with some other factor
Non-Response Bias
-Cannot be identified by comparing case and control response proportions alone. (could be equal with bias or unequal with no bias) -Bias occurs when participant cases have a different exposure frequency than non participant cases or participant controls have a different exposure freq. then non-participant controls. -Reduce impact by achieving high response proportions -Assess by comparing respondents to non-respondents separately in cases and controls on available, relevant characteristics.
Random error
-Chance, sampling variability, sample size -p-values -confidence intervals
Differential Misclassification
-Classification error DOES depend on disease status or exposure status. -Can lead to an association when there is none or to an apparent lack of an association when one does exist. -If the error of exposure classification is NOT the same for cases and controls, the bias can go either towards or away from the null depending on the nature of the misclassification.
Non-Differential Misclassification
-Classification error does not depend on disease status or exposure status. -If the error of exposure classification is the same for cases and controls, the bias will usually be TOWARDS the null -Exceptions: If there are mis-measured confounding factors or if > 2 categories of the variable.
Evaluation of Screening programs 2 -Potential sources for bias.
-Compare cause specific mortality rates between screened and unscreened populations. -Sources: -Volunteer bias -Lead time bias - Length biased sampling
Components of a published paper: Discussion
-Concise review of main findings -Places the results in the context of previous research and existing theoretical framework. -Weaknesses and limitations presented -Discuss implications for research and practice
Key Points: Confounding and Effect Mod.
-Confounding and effect modification are different concepts -Effect modification is present if the estimate of the measure of association differs according to the level of another variable -Confounding is present when the observed effect is distorted by the influence of another variable -When assessing interaction and confounding in the same study, it is possible to have one with or without the other.
Multivariate Modeling
-Constructing mathematical models to describe the association between exposure and outcome variables. -Advantages: -Allows for control of a number of confounding factors simultaneously. -Useful in situations where stratification would fail because of insufficient numbers -Disadvantages: -The choice of the appropriate model is complex -Efficient mathematical modeling can often occur at the expense of a clear understanding of the data. -Great potential for misuse
Components of a published paper: Methods 3
-Data Analysis: --How were the associations assessed or tested? --Was confounding assessed and controlled for? --What statistical techniques were used?
Components of a published paper: Results
-Describe characteristics of of study population and study findings -Do the study findings answer research question? -Would you come to the same conclusion based on the data presented in the tables or graphs? -Are point estimates and confidence intervals presented?
Recall Bias (information bias)
-Differences in cases and controls in accuracy of completeness of recall of exposure. -Results in OR away from the null when cases more accurately remember than controls or when cases over-report exposure history. -Avoid by validating exposure history from independent source or objective measures -Use of "sick" controls help equalize this bias
Self-Selection/Membership Bias 2
-Difficult to avoid; choose comparison groups with same selection pressures, try to equalize from beginning -Cannot be completely avoided in observation studies (only controlled if we are aware of all self-selection factors -Can only be controlled by randomization (in experimental studies) because individuals do not choose their exposure status-it's assigned.
Sufficient NOT necessary
-Disease Y always results when X is present, but Disease Y has other causes and can also occur without X -Disease will always occur in exposed individuals X --> Disease Y and Z --> Disease Y
Neither necessary nor sufficient
-Disease is multi-causal - There are many pathways that can lead to disease -Several risk factors can come together to cause disease
Evaluation of a Screening Program
-Does early detection of disease result in benefits to the individuals being screened? -Is the screening program effective in reducing morbidity and mortality from disease?
PPV NPV Interpretation
-Ex.- PPV=.44 44% of those who test positive actually have the disease -Ex.-NPV= .98 98% of those who test negative do not have the disease -Conclusion: The screening program may not be satisfactory because 56% of those who test positive do not have the disease.
Necessary but not Sufficient
-Factor X must always be present when disease Y is present , but disease Y is not always a result when X is present. -Some additional factor must be present also Factor X + Z --> Disease Y
Reason for screening
-Goal: Improve prognosis to reduce mortality -Assumption: Early Detection will lead to decreased morbidity and mortality -Secondary Prevention: Early diagnosis and intervention
The Screening Test: Sensitivity vs. Specificity
-Ideally want 100% for both -In practice, one is increased at the expense of the other. -Possible to vary sensitivity and specificity by changing the level at which test is interpreted as positive.
Loss to follow-up/Withdrawal Bias - Cohort Studies (Selection Bias) 2
-If incidence of outcome is different among exposed who are followed, compared to those to those who are lost, or among unexposed who are followed compared to those who are lost, than biased risk estimate will result. -Reduce by minimizing lose to follow-up -Assess by comparing lost group (exposed and unexposed) to followed group (exposed and unexposed) on baseline characteristics.
Necessary and Sufficient
-In the presence of the factor, the disease ALWAYS develops; without the factor, disease will NEVER develop Factor X --> Disease Y
Types of variation or error
-Intraperson -Intraobserver -Interobserver -Instrument or method
Lead time Bias
-Lead time: Amount of time diagnosis advanced as a result of screening -Lead time bias: Survival may falsely appear to be increased among screened group. (Simply because the diagnosis was made earlier in the course of the disease)
Restriction
-Limiting inclusion criteria to prohibit variation (Restricting subjects to those within a narrow age range) -Advantages: -Effective in providing complete control of known confounders -Convenient, inexpensive, straightforward data analysis -Disadvantages: -Shrinks pool of available study subjects -Residual confounding if restriction not sufficiently narrow -Does not permit evaluation of exposure-disease across levels of the restricted factor -may limit generalizability
Components of a published paper: Methods 2
-Measures: --How were the outcome, exposure and covariates measured? --Reliability and validity of measures and/or instruments used.
Validity- To evaluate a screening test...
-Need to know the "truth" - Need to know test results -Compare test results to "gold standard"
Non-response bias in case-control
-Never get 100% participation in studies -People who agree to participate may be different in terms of exposure or other important characteristics from those who do not agree to participate in study. -May increase or decrease estimate depending on factors related to non-response.
Hill's Criteria 8. Experiment
-Not a guideline--way of testing hypothesis - if all variables are held constant, a change in the factor leads to a change in the outcome or disease than the relationship is causal. -Most epi studies are observational
Artifactual (Spurious)
-Observed association may be the result of bias
Family Information Bias
-Occurs when cases are more aware of their family because of the occurrence of their disease has stimulated discussion or investigation in past disease history in family. -Results in an increased (odds) estimate, since cases may be more aware of a positive family history than are controls. Avoid by validating disease status of family members.
Quantifying reliability
-Percent agreement between observers or instruments PA= # of tests in which observers agree/total # of tests read X 100 a+d/total X 100
Direction of confounding
-Positive Confounding (Away from the null) -the confounding factor produces an observed estimate of the association between exposure and disease that is and OVERESTIMATE than the true association -Negative Confounding (Toward the null) -The confounding factor produces an observed estimate of the association between exposure and disease that is an UNDERESTIMATE of the true association.
Interpreting Diagnostic and Screening Tests (PPV NPV)
-Positive Predictive Value: What proportion of of patients who test positive truly have disease? -Negative Predictive Value: What proportion of those who test negative do not have disease?
Randomization
-Randomly assigning exposure to ensure equal distribution of confounders in each exposure category -Advantages: -Provides control of known and unknown factors -Convenient, inexpensive, straightforward data analysis -Disadvantages: -Can be applied only to intervention studies -Works well only for large sample sizes
Components of a published paper: Conclusion and references
-Recommendations for future research -Cited works for all scientific information referred to in article.
Self-Selection/Membership Bias
-Refers to characteristics of an individual that may consciously or unconsciously affect membership in a certain group -Individuals may select themselves into certain occupations or choose behaviors because of certain personal characteristics. -Healthy Worker effect (Morbidity and mortality is lower in workers than in the general population.) -May increase or decrease risk estimate.
Berkson's Bias (Selection bias)
-Refers to the selection factors that lead hospital cases or controls to be systematically different from all cases or controls in the population they represent. -Can occurr, for example, when a combination of exposure and disease increases the probability of hospital admission, leading to a systematically higher exposure proportion among hospital cases than among all cases. -Can occur in controls also. -Difficult to manage unless you know it is present.
Prevalence/Incidence Bias (Selection)
-Relevant to case-control studies and affects inferences from cross-sectional studies - Results when prevalent cases, rather than incidence cases, are used -Prevalence cases over-represent cases of long duration because those who die or are rapidly cured are less likely to be included.
Components of a published paper: Introduction
-Review of the relevant literature -Describe the problem being studied -include the specific aims of the study -Clearly state the research question and/or hypothesis
Direct Causal Association
-Risk factor directly associated with disease without an intermediate factor -Direct causes are not more important that indirect ones
Hill's Criteria 4. Dose-response effect
-Risk of disease increases with increasing levels of exposure
Matching
-Selecting subjects according to the value of suspected confounders to ensure equal distributions among study groups -Advantages: - Smaller sample size requirements for cohorts -Useful when there would not be a sufficient subjects alike to control for these factors in the analysis. -When case series is small, allows enough subjects in each strata to control for these factors in the analysis -When number of cases is small, R:1 matching can increase statistical power -Disadvantages: -Can be costly and time-consuming, requiring extensive searching and record keeping to find matches -May introduce confounding in case-control studies -Matching factor can no longer be evaluated as a risk factor
Types of Systematic Error
-Selection Bias: how the study group is chosen -Information Bias: Inaccuracy in measurement or classification of exposure, outcome, or co-variates (results in measurement error./mis-classification)
Prevalence/Incidence Bias (Selection) 2
-Selective mortality in the exposed cases would result in the observed OR being towards the null -Selective survival in the exposed cases would result in the observed OR being away from the null -Avoid by using incidence cases -If prevalent cases must be included: --Use cases with short interval between onset and diagnosis --Count and obtain exposure information from deceased cases --Include cases from entire spectrum of disease
Components of a published paper: Abstract
-Short summary of: --Objectives --Methods --Main Findings --Conclusion
Methods for controlling confounding.
-Study Design: 1.Randomization 2.Restriction 3.Matching -Analyzing Data 1. Stratification 2. Multivariate analysis
Other types of bias
-Surrogate interviewer: asking spouse or parents instead of person -Medical Surveillance: Looking for exposure more freq. than in general population
Types of interaction
-Synergistic effect (positive) = The effect modifier potentates or accentuates the effevct of the exposure if interest -Antagonistic Effect (Negative interaction)= The presence of the effect modifier diminishes or eliminates the effect if the exposure of interest.
Summary of Bias
-Synthetic, non-random errors and can occur at any stage or research. -Includes selection and information -Results in invalid findings. (over or under estimates) -In contrast, a confounded association, not causal, but real.
Information Bias
-Systematic error in the exposure assessment or outcome assessment. -Information bias results in the misclassifications of study subjects with respect to disease or exposure status. -an error in the measurement of categorical variables is usually called misclassification. -An error in the measurement of continuous variables is called measurement error.
Avoiding and Controlling for Confounding
-The aim is to control confounding and eliminate its effects.
Effect modification Wrap up
-The aim is to describe and report it, NOT control it -Assessed by comparing the magnitude (and direction) of stratum-specific estimates -Use stratification to evaluate and describe.
Loss to follow-up/Withdrawal Bias - Cohort Studies (Selection Bias)
-The analogue of non-response bias in c/c studies -Cannot determine whether bias is present based on follow-up proportions in exposed and unexposed alone. Bias can be present if follow-up proportions are equal or absent if the are unequal.
Hill's Criteria 9. Analogy
-The association is more likely to be causal if a similar relationship has been observed with another exposure and/or disease -Weak form of evidence
Effect Modification
-The effect of the exposure on the outcome differs depending on the level of another variable, the effect modifier. -Effect Modification= Interaction EX: An association that is stronger in older people than in younger people; age is an effect modifier.
Hill's Criteria 7. Coherence of Explanation
-The findings are compatible with existing theory and knowledge
Screening
-The identification of disease in asymptomatic individuals by application of rapid tests to separate persons who probably have the disease from those who probably do not have the disease -NOT intended to be Diagnostic
Critical Appraisal Definition
-The process where a study is reviewed to assess the= --Validity of data --Completeness of reporting --Methods and procedures --Conclusions --Compliance with ethical standards
False Positive Rate
-The proportion of non-diseased incorrectly classified as diseased by the screening test. -False Positives/ All non-diseased or - 1-specificity
Causality
-The relation of causes to the effects they produce --Necessary and Sufficient Causes
External Validity
-The results from the study can be generalized to some other population
Hill's Criteria 2. Strength of association
-The stronger the association between the exposure and the disease , the less likely the association is due to bias
Hill's Criteria 6. Specificity of Association
-The study factor (exposure) is associated with only one disease or the disease is associated with only one factor. --Weakest criteria
Internal Validity
-The study provides an UNBIASED estimate of what it claims to estimate
Calculating Sensitivity
-True positives/All diseased (a/a+c)
Components of a published paper: Methods 1
-Type of study design -Study population: --study setting --sampling technique --sample size --refusal and/or follow-up proportions --inclusion/exclusion criteria --threats to generalizability
Universal Screening vs. Targeted Screening
-Universal: Screening everyone in the population (Newborn babies hearing test) -Targeted: Screening populations identified as being high risk (Mammography screening for women under 40 who have a family history of Breast Cancer)
Description of Association
-What is the magnitude of the association, is it strong, weak, positive, negative? (RR, OR) -Is the association significant? (P-value, confidence interval)
Sufficient Cause
-When a particular factor inevitably produces an effect -May be a single variable (rarely the case) or a cluster of causes which are sufficient together to produce disease
Necessary Causes
-When a particular factor must always be present for the disease to occur -The disease outcome of interest does not need to be the only outcome resulting from the exposure to this causal factor -may or may not require the presence of other factors in order to lead to disease
Interaction
-When the incidence of disease in the presence of two (or more) risk factors differs from the incidence expected to result from their individual effects. -Differences in the effect of one (or more) factors according to the level of the remaining factors.
Definition of Cause
-an event, condition, or characteristic that precede the disease event and without which the disease event either would not have occurred at all or would not have occurred until some later time.
Screening Characteristics Review -Reliability -Validity -Sensitivity -Specificity --Interpretation-- -PPV -NPV
-get the SAME result each time (Precise) -get the CORRECT result (Accurate) -correctly classifying cases -correctly classifying non-cases -proportion of positive test results identifying cases -proportion of negative test results identifying non-cases *SCREENING does NOT = Diagnosis
Interviewer Bias (information bias)
-systematic error due to interviewer's subconscious or conscious gathering of selective information in different groups. -Results in an overestimate of association when interviewers probe for more information among cases -Avoid by blinding interviewers to case/controls status; train interviewers to use standardized forms. -Asses by re-interviewing a sample of study subjects.
Steps for evaluating Confounding
1. Calculate crude overall estimation of the exposure-disease association 2. Stratify the data by levels of the suspected confounder 3. Calculate the stratum specific estimates 4. Compare the stratum specific estimates to one another and to the crude measure of association.
Using Stratification to identify presence of effect modification
1. Calculate the crude overall estimation of the exposure-disease association 2. Stratify the data by levels of the third factor 3. Calculate the stratum-specific estimates 4. Compare the stratum-specific estimates to one another and to the crude measure of association. 5. Determine whether the magnitude of the stratum-specific estimates is different across strata
Considerations for cut point decisions
1. Cost of falsely classifying healthy persons as diseased (FP) -may choose increased sensitivity when penalty associated with missing a case is high. (Disease is serious, can be spread, minimal costs and risks) 2. Cost of leaving true cases undetected (FN) -May choose increased specificity when cost or risk is associated with further diagnostic testing is substantial.
Two Stage process for assessing causality
1. Determine is observed result is valid by ruling out alternative explanation due to bias, confounding, or random error. 2. If association is real, assess whether the exposure actually caused the outcome. Use Hill's criteria to evaluate the strength of the evidence for causality. (Decided if association is causal or non-causal
Criteria for a confounding Variable
1. It must be a risk factor for the disease, or associated with the disease, but not necessarily causal. 2. It must be associated with the exposure under study, but not a result of it. 3. It must not be an intermediate variable on the causal pathway
Reviewing an epidemiological study
1. Motivation for study 2. Study objective 3. Study Design 4. Study population 5. Exposure measurment 6. Primary outcome of interest 7. Potential source for error 8. Interpretation of data
Properties of a causal factor
1. Must be associated with the outcome (Statistical dependance) 2. Must proceed the outcome (time order) 3. A change in the casual factor must produce a change in the outcome (direction)
Levels of prevention
1. Primary prevention: -Prevention of occurrence of disease -remove the exposure/agent 2. Secondary Prevention: -Goal: reduce the progress of disease -early detection and prompt treatment -Focus for most chronic diseases 3. Tertiary Prevention: -Limit disability through rehabilitation -minimize side effects -disease has already occurred and been treated
Outcome Measures
1. Reduction of mortality in the population being screened. 2. Reduction of case-fatality in the screened persons. 3. Increased in percent of cases detected at earlier stages. 4. Reduction in complications. 5. Prevention of/reduction in recurrences or metastases. 6. Improvement in quality of life in screened individuals.
3 Areas to consider when evaluation Journals
1. Relevance: -topic -intended audience 2. Extrinsic: -Authors -Institutional affiliations -Journal -Funding agency -Conflicts of interest 3. Intrinsic Factors: -Methodology
Two Components of Validity
1. Sensitivity- The ability of test to correctly identify those who have the disease. 2. Specificity: The ability to correctly identify those who do not have the disease.
Interpretation
1. Sensitivity: .80 -The test is able to correctly identify 80% of those with the disease 2. Specificity: .89 -The test is able to correctly identify 89% of those without the disease. Conclusion: The screening test is fairly good at correctly ruling out disease, but fails to pick up 20% of those with the disease.
WHO's 6 Criteria for Effective Screening
1. The condition being screened for must be serious 2. The condition being screened for must be treatable 3. The condition must be detectable while asymptomatic and timely treatment must reduce morbidity and mortality more effectively than treatment after the appearance of the symptoms. 4. The screening test must be accurate 5. The screening test must be acceptable to the patient and inexpensive 6. The condition must be sufficiently prevalent to warrant screening.
Criteria for evaluating Screening Tests
1. Validity (Accuracy): -Sensitivity -Specificity 2. Reliability(Precision) -Percent agreement
Direction of bias
1.Away from the null -Observed measure of association is FARTHER from 1.0 than is the true value. (Bigger effect than true) 2. Toward the null -Observed measure of association is CLOSER to 1.0 than is the true value. (smaller effect than is true.
Bias in estimating Association
A wrong estimate of the measure of association between an exposure and a disease -the exposure or disease outcome NOT measured with a valid, standardized method. (Information bias) -The exposure status or disease status influence the selection process. (selection bias)
Other methods of Quantifying reliability -Kappa Statistic-Agreement beyond chance
Kappa= (% observed agreement) - (% agreement expected by chance alone) / 100% - (% agreement expected by chance alone)
Volunteer Bias
Screening population made up of: -Self-selected volunteers -"Worried Well" -May be healthier or at a higher risk of developing the disease than those that don't participate.
Selection for potential confounders
Selection based on: -Existing knowledge -Evaluation -Best Judgement of the Investigator *Selection is not based on statistical significance
Calculating NPV
True Negatives/ all negatives d/c+d
Calculating PPV
True Positives/all positives a/a+b
Association
a.k.a. correlation, relationship, statistical dependance -Relationship between two or more events or variables -Events may occur more frequently together than one would expect by chance -statistical dependance between the causal factor and the effect
