BIAS AND CONFOUNDING IN RESEARCH
Module 5 – Home
BIAS AND CONFOUNDING IN RESEARCH
Bias and confounding are two important concepts in research that can influence the validity and reliability of research findings.
Bias refers to any systematic error or deviation in the design, conduct, or analysis of a research study that can affect the results. There are several types of bias that can occur in research, including:
Sampling bias: This occurs when the sample of participants in a study is not representative of the population being studied. For example, if a study only includes participants of a certain age or gender, the results may not be applicable to other age or gender groups.
Selection bias: This occurs when the selection of participants or the allocation of treatments in a study is not random, and is influenced by some other factor. For example, if a study only includes participants who have a certain condition or characteristic, the results may not be applicable to those who do not have that condition or characteristic.
Measurement bias: This occurs when the measurement of a variable is not accurate or reliable. For example, if a study uses a self-report questionnaire to measure a variable, the results may be influenced by participants’ willingness or ability to accurately report their experiences or behaviors.
Confounding refers to the presence of a third variable that is related to both the independent variable (the variable being manipulated in the study) and the dependent variable (the variable being measured in the study). Confounding can occur when the relationship between the independent and dependent variables is obscured or distorted by the presence of the confounding variable. For example, if a study is examining the relationship between a new medication and blood pressure, and the participants also happen to be exercising more or eating a healthier diet, this could confound the relationship between the medication and blood pressure, making it difficult to determine the true effect of the medication.
Bias and confounding can have significant impacts on the validity and reliability of research findings. It is important for researchers to be aware of these issues and to take steps to minimize their influence, such as using appropriate sampling and selection methods, using reliable measurement tools, and controlling for confounding variables.
Modular Learning Outcomes
Upon successful completion of this module, the student will be able to satisfy the following outcomes:
Case
Identify possible sources of bias and confounding in research.
Apply knowledge about sources of error to peer-reviewed articles on foodborne disease.
SLP
Identify possible sources of bias and confounding in research.
Discussion
Describe the effect of non-differential misclassification on studies.
Module Overview
When a statistically significant association between a factor and health outcome has been demonstrated, the association may be of three types: 1) artifactual, 2) indirect, or 3) causal. The focus of Module 3 was the assessment of a causal relationship. This module will focus on artifactual association, indirect association, and effect modification.
Artifactual Association
An artifactual association is a false association that can result from chance occurrence or bias in study methods. In a certain proportion of studies, an outcome may be found statistically significant even though it results from random fluctuation (Type I error). Bias can also result in an artifactual association, through flaws in study design, methods used to collect data, or the way that a study group is selected. Bias is defined as a systematic error which results in an incorrect or invalid estimate of the measure of association. Some of the common types of bias are selection bias and information bias.
Selection Bias
Selection bias arises when participants of a study group are systematically excluded due to a particular attribute. This exclusion may influence the statistical significance of the test. For example, selecting individuals from a treatment clinic may result in selection bias, since the characteristics from the control group are not representative of the population. Also, allowing individuals to refer themselves to the study may result in selection bias, since the reasons for the referral may be associated with the outcome.
Information Bias
Information bias arises from errors in measuring exposure or disease. Some common types of information bias include:
Interviewer bias (i.e., when the interviewer influences questions)
Observer bias (i.e., when an observer has preconceived expectations)
Reporting bias (i.e., when individuals with more severe disease are more likely to report than those with less severe disease)
Recall bias (i.e., when those who are ill are more likely to recall an exposure than those who were are not ill)
Follow-up bias (i.e., those who are lost to follow-up are different from those who are retained)
Measurement error can also be a bias in studies. Therefore, it is important to assess the quality of measurements, which involves examining validity and reliability. For more information, click on Measurement Error.
Controlling for Bias
There are a number of methods that can help control for bias. Selecting a random sample reduces the likelihood of selection bias. Strategies for controlling for information bias include:
Blinding of subjects, interviewers, and/or analyst
Including similar questions on a questionnaire to verify validity
Training of observers and/or data collectors
Limit the number of observers and/or data collectors
Random assignments for interviewers and/or data collectors
Minimize loss to follow-up
Utilizing case definitions
Additionally, replication of studies minimizes the likelihood of an artifactual association.
Indirect Association
An indirect association refers to when an exposure appears to be associated with an outcome; however, there is a third variable that is related to both the factor and outcome. The relationship between the exposure and outcome is confounded by this third factor.
(Howell, 2019)
Confounding can be controlled in study design by using random selection, matching, and restricting subjects according to confounder. In data analysis, confounding may be controlled through methods such as stratification and model fitting.
Effect Modification
Effect modification (interaction) refers to when the effect of an exposure on the outcome differs depending if the effect modifier is present. If the presence of an effect modifier strengthens the effect, the interaction is synergistic (positive interaction). If the presence of an effect modifier weakens or eliminates the effect, the interaction is antagonistic (negative interaction)
Confounding vs. Effect Modification
When a statistically significant association between a factor and health outcome has been demonstrated, the crude odds ratio should be compared with the adjusted odds ratio. A considerable difference in the odds ratios suggests that there is either confounding or interaction. significantly, then there is confounding or interaction. Stratification can be used to distinguish between whether there is confounding or interaction. If there is is different odds ratio between strata, then an interaction is occurring. If the odds ratio is similar between strata, then confounding is occurring. For more information on confounding and interaction, refer to the Background Reading.
Module 5 – Measurement Error
BIAS AND CONFOUNDING IN RESEARCH
Introduction
Three important goals of data collection and analysis are the promotion of accuracy and precision, the reduction of differential and non-differential errors (that is, nonrandom and random errors), and the reduction in inter-observer and intra-observer variability (that is, variability between the findings of two observers or between the findings of one observer on two different occasions).
Various statistical methods are available to study the accuracy and usefulness of screening tests and diagnostic tests in clinical medicine. In general, tests with a high degree of sensitivity and a corresponding low false-negative error rate are helpful for screening patients, while tests with a high degree of specificity and a corresponding low false-positive error rate are useful for confirming the diagnosis in patients suspected of having a particular disease.
Promoting Accuracy and Precision
Accuracy refers to the ability of a measurement to be correct on the average. If a measure is not accurate, it is biased. Precision, sometimes known as reproducibility or reliability, is the ability of a measurement to give the same result or a very similar result with repeated measurements of the same thing. Random error alone, if large, will result in lack of precision.
Reducting Differential and Non-Differential Errors
Bias is a non-random, systematic, or consistent error in which the values tend to be inaccurate in a particular direction. Bias results, for example, from measuring the heights of patients with their shoes on or from measuring blood pressures with an arm cuff that reads too high or too low. Statistical analysis cannot correct for bias unless the amount of bias in each individual measurement is known. In the example of the patients’ height measurements, bias could be corrected only if the height of each patient’s heel was known and subtracted from that patient’s reported measurement.
While measuring patients in their bare feet could eliminate bias, it would not necessarily eliminate random errors or non-differential errors. When data have only random errors, some observations will be too high and some will be too low. It is possible for random errors to produce biased results. However, if there are enough observations, data with random errors may produce a correct estimate of the mean.
Reducing Intra-Observer and Inter-Observer Variability
A goal of data collection and analysis is to reduce the amount of intra-observer (within observer) and inter-observer (between observers) variability. If the same physician takes successive measurements of the blood pressure and height of the same person or if the same physician examines the same x-ray several times without knowing that it is the same x-ray, there will usually be some differences in the measurements or interpretations obtained. This is known as intra-observer variability. If two different physicians measure the same blood pressure or examine the same x-ray independently, there will usually be some differences. This is called inter-observer variability.
Characteristics of Screening Tests
Screening tests are the basic tool in the early detection of disease. Important characteristics include validity, predictive value, reliability, and yield.
Validity
Screening tests should provide a good preliminary indication of which individuals have the disease and which do not. This is referred to as validity. Validity has two components: sensitivity and specificity. Sensitivity is defined as the ability of a test to identify correctly those who have a disease. Those who have a disease may either correctly test positive for a disease (true positive) or incorrectly test negative for a disease (false negative). Specificity is defined as the ability of a test to identify correctly those who do not have a disease. Those who do not have a disease may either correctly test negative for a disease (true negative) or incorrectly test positive for a disease (false positive). The formulas for sensitivity and specificity are:
Sensitivity = True Positives/All With Disease = True Pos/(True pos + False neg)
Specificity = True Negatives/All Without Disease = True Neg/(True neg + False pos)
An ideal screening test would be 100% sensitive and 100% specific. In practice, this does not occur because they are usually inversely related.
Predictive Value
The ability to predict the presence or absence of a disease from test results is referred to as predictive value. It is dependent on the prevalence of a disease in the population tested, as well as the sensitivity and specificity of the test. The higher the prevalence, the more likely it is that a positive test is predictive of a disease. The predictive value of a positive and negative test can be calculated as follows:
PV (pos) = True positives/All positives = True pos/(True pos + False pos)
PV (neg) = True negatives/All negatives = True neg/(True neg + False neg)
For example, if Disease X has a prevalence of 10% in a population of 1000 people, and the sensitivity of the test is 90% and the specificity is 60%, the predictive value would be calculated as follows:
Write down all formulas containing the given variables and plug in the values above.
PV (pos) = True pos/(True pos + False pos)
PV (neg) = True neg/(True neg + False neg)
Sensitivity = .90 = True pos/All with disease = True pos/(True pos + False neg)
Specificity = .60 = True neg/All without disease = True neg/(True neg + False pos)
Prevalence = .10 = All with disease/Total pop = #cases/1000
Calculate the variables needed for the predictive value by rearranging the formulas and plugging in the calculated values.
All with disease = Prevalence x Total pop = (.10)(1000) = 100
True pos = sensitivity x All with disease = (.90)( 100) = 90
False neg = All with disease – True pos = 100 – 90 = 10
All without disease = Population – All with disease = 1000 – 100 = 900
True neg = specificity x All without disease = (.60)(900) = 540
False pos = All without disease – True neg = 900 – 540 = 360
Plug in calculated values for the predictive value.
PV (pos) = 90/(90 + 360) = 20%
PV (neg) = 540/(540 + 10) = 98%
Reliability
Reliability is defined as consistency in screening test results when the test is performed more than once on the same individual under the same conditions. Two major factors affect reliability: variability in the method (e.g. stability of reagents) and observer variation. These variations can be reduced by standardization of procedures, training of observers, and periodic checks on their work.
Yield
Yield is defined as the amount of previously unrecognized disease that is diagnosed and brought to treatment as a result of screening. The following factors affect the yield of a screening program:
Sensitivity of the test
Prevalence of unrecognized disease
Frequency of screening
Participation in screening and follow-up
Assessment of Screening Programs
Because of the high costs and risks, screening programs should be conducted only when it has the potential to lead to a significant decrease in the rates of death or disability. Outcome measures such as the disease-specific death rate, physiological variables (e.g. blood cholesterol level), and case-fatality rate should be compared between screened and unscreened groups. The least biased method for evaluating a screening program would be a random, experimental study (i.e. randomly assigning individuals in a study sample to either a group that provides free screening or a control group).
Sources of Bias in Screening Programs
Unfortunately, randomized, experimental studies are rarely done. Instead, observational studies are often conducted, which are subject to several different sources of bias:
Lead Time Bias. Lead time is defined as the interval between a time a condition is detected through screening and the time it would normally have been detected by the reporting of symptoms. A screening test may lead to early detection without delaying the time of death, and as a result, the screened case group may appear to have a higher survival rate than the control group.
Patient Self-Selection. Individuals who choose to participate in early detection programs may differ from those who do not in characteristics that may be related to survival. For example, participants may be more health-conscious, more likely to be compliant with prescribed therapy, etc.
Module 5 – Background
BIAS AND CONFOUNDING IN RESEARCH
Required Reading
Stratified Analysis: Introduction to Confounding and Interaction Retrieved February 21, 2012, at:
http://www.sjsu.edu/faculty/gerstman/StatPrimer/stratified.PDF
Cruz, M. A., Katz, D.J ., Suarez, J. A. (2001). An assessment of the ability of routine restaurant inspections to predict food-borne outbreaks in Miami-Dade County, Florida. American Journal of Public Health, 91(5). Retrieved February 21, 2012, at ProQuest
Shrier, I. & Pang, M. (2015). Confounding, effect modification and the odds ratio: Common misinterpretation. Journal of Clinical Epidemiology, 68(4), 470-474. Retrieved August 24, 2019 from https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4882164/
Simon, P., Leslie, P., Run, G., Jin, G.Z. et al (2005). Impact of Restaurant Hygiene Grade Cards on Foodborne-Disease Hospitalizations in Los Angeles County. Journal of Environmental Health, 67(7). Retrieved February 21, 2012, at ProQuest
M. Tevfik DORAK Paediatric & Lifecourse Epidemiology Research Group School of Clinical Medical Sciences (Child Health) Newcastle University England, U.K. Presentation Retrieved February 21, 2012, at:
http://www.dorak.info/epi/bc.ppt
Additional Reading
Bernstein, J., Chollet, D., & Peterson, G. (2010) Encouraging Appropriate Use of Preventive Health Services. Mathematica Policy Research, Inc., 2, 1-5. Retrieved August 24, 2019 from https://www.mathematica-mpr.com/our-publications-and-findings/publications/encouraging-appropriate-use-of-preventive-health-services
Schneeweiss, S., Seeger, J., Maclure, M., Wang, P., Avorn, J., & Glynn, R. (2001). Performance of comorbidity scores to control for confounding in epidemiologic studies using claims data. American Journal of Epidemiology, 154(9), 854-864. Retrieved August 24, 2019 from https://academic.oup.com/aje/article/154/9/854/124960
Confounding. Super Course. Retrieved February 21, 2012, at: http://www.pitt.edu/~super1/lecture/lec7091/index.htm
Module 5 – Measurement Error
BIAS AND CONFOUNDING IN RESEARCH
Introduction
Three important goals of data collection and analysis are the promotion of accuracy and precision, the reduction of differential and non-differential errors (that is, nonrandom and random errors), and the reduction in inter-observer and intra-observer variability (that is, variability between the findings of two observers or between the findings of one observer on two different occasions).
Various statistical methods are available to study the accuracy and usefulness of screening tests and diagnostic tests in clinical medicine. In general, tests with a high degree of sensitivity and a corresponding low false-negative error rate are helpful for screening patients, while tests with a high degree of specificity and a corresponding low false-positive error rate are useful for confirming the diagnosis in patients suspected of having a particular disease.
Promoting Accuracy and Precision
Accuracy refers to the ability of a measurement to be correct on the average. If a measure is not accurate, it is biased. Precision, sometimes known as reproducibility or reliability, is the ability of a measurement to give the same result or a very similar result with repeated measurements of the same thing. Random error alone, if large, will result in lack of precision.
Reducting Differential and Non-Differential Errors
Bias is a non-random, systematic, or consistent error in which the values tend to be inaccurate in a particular direction. Bias results, for example, from measuring the heights of patients with their shoes on or from measuring blood pressures with an arm cuff that reads too high or too low. Statistical analysis cannot correct for bias unless the amount of bias in each individual measurement is known. In the example of the patients’ height measurements, bias could be corrected only if the height of each patient’s heel was known and subtracted from that patient’s reported measurement.
While measuring patients in their bare feet could eliminate bias, it would not necessarily eliminate random errors or non-differential errors. When data have only random errors, some observations will be too high and some will be too low. It is possible for random errors to produce biased results. However, if there are enough observations, data with random errors may produce a correct estimate of the mean.
Reducing Intra-Observer and Inter-Observer Variability
A goal of data collection and analysis is to reduce the amount of intra-observer (within observer) and inter-observer (between observers) variability. If the same physician takes successive measurements of the blood pressure and height of the same person or if the same physician examines the same x-ray several times without knowing that it is the same x-ray, there will usually be some differences in the measurements or interpretations obtained. This is known as intra-observer variability. If two different physicians measure the same blood pressure or examine the same x-ray independently, there will usually be some differences. This is called inter-observer variability.
Characteristics of Screening Tests
Screening tests are the basic tool in the early detection of disease. Important characteristics include validity, predictive value, reliability, and yield.
Validity
Screening tests should provide a good preliminary indication of which individuals have the disease and which do not. This is referred to as validity. Validity has two components: sensitivity and specificity. Sensitivity is defined as the ability of a test to identify correctly those who have a disease. Those who have a disease may either correctly test positive for a disease (true positive) or incorrectly test negative for a disease (false negative). Specificity is defined as the ability of a test to identify correctly those who do not have a disease. Those who do not have a disease may either correctly test negative for a disease (true negative) or incorrectly test positive for a disease (false positive). The formulas for sensitivity and specificity are:
Sensitivity = True Positives/All With Disease = True Pos/(True pos + False neg)
Specificity = True Negatives/All Without Disease = True Neg/(True neg + False pos)
An ideal screening test would be 100% sensitive and 100% specific. In practice, this does not occur because they are usually inversely related.
Predictive Value
The ability to predict the presence or absence of a disease from test results is referred to as predictive value. It is dependent on the prevalence of a disease in the population tested, as well as the sensitivity and specificity of the test. The higher the prevalence, the more likely it is that a positive test is predictive of a disease. The predictive value of a positive and negative test can be calculated as follows:
PV (pos) = True positives/All positives = True pos/(True pos + False pos)
PV (neg) = True negatives/All negatives = True neg/(True neg + False neg)
For example, if Disease X has a prevalence of 10% in a population of 1000 people, and the sensitivity of the test is 90% and the specificity is 60%, the predictive value would be calculated as follows:
Write down all formulas containing the given variables and plug in the values above.
PV (pos) = True pos/(True pos + False pos)
PV (neg) = True neg/(True neg + False neg)
Sensitivity = .90 = True pos/All with disease = True pos/(True pos + False neg)
Specificity = .60 = True neg/All without disease = True neg/(True neg + False pos)
Prevalence = .10 = All with disease/Total pop = #cases/1000
Calculate the variables needed for the predictive value by rearranging the formulas and plugging in the calculated values.
All with disease = Prevalence x Total pop = (.10)(1000) = 100
True pos = sensitivity x All with disease = (.90)( 100) = 90
False neg = All with disease – True pos = 100 – 90 = 10
All without disease = Population – All with disease = 1000 – 100 = 900
True neg = specificity x All without disease = (.60)(900) = 540
False pos = All without disease – True neg = 900 – 540 = 360
Plug in calculated values for the predictive value.
PV (pos) = 90/(90 + 360) = 20%
PV (neg) = 540/(540 + 10) = 98%
Module 5 – Case
BIAS AND CONFOUNDING IN RESEARCH
Case Assignment
Please answer the following questions in a paper of at least 6 pages:
A researcher is studying the effect of factors that may contribute to a birth defect. The following is a table summarizing the results of bivariate and multivariate analysis.
Table 1. Bivariate versus Multivariate
Variable
Crude OR
95% CI
Adjusted OR
95% CI
Gene X
0.986
0.974–0.998
0.985
0.971–0.999
Gene Y
4.317
1.916–9.727
3.430
1.377–8.547
Gene Z
2.02
1.081–3.891
1.34
0.854–2.115
Gene T
3.365
1.021–11.088
6.236
1.562–24.892
Identify variables from Table 1 that should be further analyzed (to rule out whether they are confounders or effect modifiers). Briefly provide rationale.
Stratified analysis was conducted on one of the genes in Table 1 and odds ratios were obtained. What do these results suggest:
Drug A=0: OR=2.2 for birth defect given presence of gene
Drug A=1: OR=1.4 for birth defect given presence of gene
Critically review the following controversial paper published (and later retracted) by Wakefield et al, which implicated MMR vaccine as a possible etiologic contributor to Autism, etc. You may use either version for your review. Describe potential sources of bias and other limitations in the study:
Ileal-lymphoid-nodular hyperplasia, non-specific colitis, and pervasive developmental disorder in children A J Wakefield, S H Murch, A Anthony, J Linnell, D M Casson, M Malik, M Berelowitz, A P Dhillon,
M A Thomson, P Harvey, A Valentine, S E Davies, J A Walker-Smith
The Lancet, Volume 351, Number 9103 28 February 1998
Pro 7 Con Arguments: “Should any vaccines be required for children? Retrieved from http://vaccines.procon.org/ on November 5, 2013.
Assignment Expectations
Length: Case Assignments should be at least 6 pages (1,300 words) in length.
References: Required readings should be included. All references need to be cited using APA format.
Organization: Subheadings should be used to organize your paper according to question.
Format: APA format is required for all assignments at the Ph.D. level. See Syllabus page for more information on APA format.
Grammar and Spelling: While no points are deducted for minor errors, assignments are expected to adhere to standards guidelines of grammar, spelling, punctuation, and sentence syntax. Points may be deducted if grammar and spelling impact clarity.
The following items will be assessed in particular:
Relevance—all content is connected to the question.
Precision—specific question is addressed. Statements, facts, and statistics are specific and accurate.
Depth of discussion—present and integrate points that lead to deeper issues.
Breadth—multiple perspectives and references, multiple issues/factors considered.
Evidence—points are well-supported with facts, statistics, and references.
Logic—presented discussion makes sense; conclusions are logically supported by premises, statements, or factual information.
Clarity—writing is concise, understandable, and contains sufficient detail or examples.
Objectivity—avoid use of first person and subjective bias.
Additional expectations for the assignment:
Your references and citations should be consistent with a particular formatting style, such as APA.
Your response should be based on scholarly material, such as peer-reviewed articles, white papers, technical papers, etc. Do not include information from non-scholarly materials such as Wikis, encyclopedias, www.freearticles.com (or similar websites).
Your response should incorporate the objectives of the module with the requirements of this assignment.
Your Case paper will be further evaluated based on the following criteria:
Precision (Excellent; Good; Average; Poor)
Each question and/or assignment requirement is addressed in the paper.
Accuracy of your answers, key points, and supporting discussion.
Clarity (Excellent; Good; Average; Poor)
The paper is well organized, concise, reads clearly, and it is not confusing.
Breadth (Excellent; Good; Average; Poor)
The paper presents appropriate breadth covering the assignment questions/requirements.
Depth (Excellent; Good; Average; Poor)
Presents key points that lead to deeper matters and issues.
Integrates several points into coherent conclusions.
Critical Thinking (Excellent; Good; Average; Poor)
The paper demonstrates good understanding and synthesis of the module background material.
Logically incorporates key concepts presented in the background material into the overall analyses, key points, and supporting discussions.
Presents well-reasoned conclusions and positions as well as convincing arguments in support of the same.
Writing Mechanics
Grammar (Excellent; Good; Average; Poor)
Spelling (Excellent; Good; Average; Poor)
Vocabulary (Excellent; Good; Average; Poor)
Referencing (Excellent; Good; Average; Poor)
Additional considerations to keep in mind while working on your Case Assignments:
Originality and Use of directly quoted material
The purpose of each assignment is for you to present your understanding and synthesis of the background material and key concepts. Accordingly, this must be accomplished substantially in your own words.
Use directly quoted material sparingly and only when preserving the exact words of an author is necessary. For purposes of this course, rarely should directly quoted material represent more than 5–10 percent of your entire case paper’s content. Composing a paper patched together from mostly quoted material is not acceptable.
Be sure to properly reference all directly quoted material and include in-text citations.
Grades
A paper that answers all assignment questions with good key points, well-supported discussions, and quality references will earn a solid “B” grade.
“A” papers are exceptional works that go beyond just answering the questions by providing insightful answers.
Limit response to a minimum of 6 pages
Module 5 – SLP
BIAS AND CONFOUNDING IN RESEARCH
Limitations of Studies
This assignment involves identifying limitations of your proposed study. In approximately 2–3 pages, describe the limitations of the study you proposed in Module 4. Your discussion of limitations should include the following (use subheadings for each section):
Limitations of Proposed Study Design. Discuss potential bias and limitations.
Limitations of Proposed Sampling Method. Discuss potential bias and limitations.
Limitations of Proposed Data Sources/Data Collection Methods. Discuss potential bias and limitations.
Limitations of Proposed Analytic Methods. Discuss potential bias and limitations.
Discuss potential confounding factors in your proposed study.
Please upload this SLP component at the end of Module 5. (Your other SLP components should be uploaded under the previous corresponding modules, rather than under Module 5).
SLP Assignment Expectations
Length: This SLP assignment should be at least 2 pages (500 words) in length.
References: References should be from academic sources and cited using APA format.
Organization: Subheadings should be used to organize your paper as indicated above
Format: APA format is required for all assignments at the Ph.D. level. See Syllabus page for more information on APA format.
Grammar and Spelling: While no points are deducted for minor errors, assignments are expected to adhere to standards guidelines of grammar, spelling, punctuation, and sentence syntax. Points may be deducted if grammar and spelling impact clarity.
The following items will be assessed in particular:
Relevance—all content is connected to the question.
Precision—specific question is addressed. Statements, facts, and statistics are specific and accurate.
Depth of discussion—present and integrate points.
Breadth—range of limitations included.
Evidence—points are well-supported with facts, statistics, or references.
Logic—presented discussion makes sense; conclusions are logically supported by premises, statements, or factual information.
Clarity—writing is concise, understandable, and contains sufficient detail or examples.
Objectivity—avoid use of first person and subjective bias.
Reliability
Reliability is defined as consistency in screening test results when the test is performed more than once on the same individual under the same conditions. Two major factors affect reliability: variability in the method (e.g. stability of reagents) and observer variation. These variations can be reduced by standardization of procedures, training of observers, and periodic checks on their work.
Yield
Yield is defined as the amount of previously unrecognized disease that is diagnosed and brought to treatment as a result of screening. The following factors affect the yield of a screening program:
Sensitivity of the test
Prevalence of unrecognized disease
Frequency of screening
Participation in screening and follow-up
Assessment of Screening Programs
Because of the high costs and risks, screening programs should be conducted only when it has the potential to lead to a significant decrease in the rates of death or disability. Outcome measures such as the disease-specific death rate, physiological variables (e.g. blood cholesterol level), and case-fatality rate should be compared between screened and unscreened groups. The least biased method for evaluating a screening program would be a random, experimental study (i.e. randomly assigning individuals in a study sample to either a group that provides free screening or a control group).
Sources of Bias in Screening Programs
Unfortunately, randomized, experimental studies are rarely done. Instead, observational studies are often conducted, which are subject to several different sources of bias:
Lead Time Bias. Lead time is defined as the interval between a time a condition is detected through screening and the time it would normally have been detected by the reporting of symptoms. A screening test may lead to early detection without delaying the time of death, and as a result, the screened case group may appear to have a higher survival rate than the control group.
Patient Self-Selection. Individuals who choose to participate in early detection programs may differ from those who do not in characteristics that may be related to survival. For example, participants may be more health-conscious, more likely to be compliant with prescribed therapy, etc.
Module 5 – Outcomes
BIAS AND CONFOUNDING IN RESEARCH
Module
Identify possible sources of bias and confounding in clinical research.
Case
Identify possible sources of bias and confounding in research.
Apply knowledge about sources of error to peer-reviewed articles on foodborne disease.
SLP
Identify possible sources of bias and confounding in research.
Discussion
Describe the effect of non-differential misclassification on studies.
Discussion: Non-Differential Misclassification
Discussion Topic
Task: Reply to this topic
“Many investigators write as if non-differential exposure misclassification inevitably leads to a reduction in the strength of an estimated exposure–disease association. Unfortunately, non-differentiality alone is insufficient to guarantee bias towards the null. Furthermore, because bias refers to the average estimate across study repetitions rather than the result of a single study, bias towards the null is insufficient to guarantee that an observed estimate will be an underestimate. Thus, as noted before, exposure misclassification can spuriously increase the observed strength of an association even when the misclassification process is non-differential and the bias it produced is towards the null.”
Jurek, A., Greenland, S., Maldonado, G., Church, T. (2005). Proper interpretation of non-differential misclassification effects: expectations vs observations. International Journal of Epidemiology, 34(3). Retrieved from http://ije.oxfordjournals.org/content/34/3/680.full.
What effect does non-differential exposure misclassification have on studies?
Do you agree with these authors that it does not necessarily bias towards the null or do you agree with other investigators that it reduces the strength of an association? Why?
Please post your response and then you will be able to comment on student’s responses.
As required for all discussion questions, your response must be referenced and supported.
