epi final bias
what are the types of bias
- selection bias -information bias -analytic bias
what are overall solutions to selection bias?
1. get a high participation rate for cohort study 2. select a study population representative of the reference population 3. use incident cases 4. use nested case-control studies
what are ways to minimize selection bias?
1. provide a clear definition of the study population 2. be explicit about case and control definitions 3. cases and controls come from the same population 4. select the exposed and non-exposed without knowing the disease status (retrospective cohort)
what is berkson's bias?
A form of selection bias that causes hospital cases and controls in a case control study to be systematically different from one another because the combination of exposure to risk and occurrence of disease increases the likelihood of being admitted to the hospital.
Cohort study examining effect of smoking on emphysema Diseases are ascertained by asking subjects for self reports of medical diagnoses, not by any medical exams conducted in the entire cohort. what would this do to our OR?
Emphysema may be more likely to be diagnosed in smokers than non smokers, because physicians will be more likely to look for it in non smokers, and smokers are likely to have more contact with physicians due to other health problems related to smoking. OR will be overestimated, because disease will be missed more in non-smokers.
Case control study of pancreatic cancer and coffee drinking. Controls were selected from gastroenterologist's patients in same hospital. However, GI patients are less likely to drink coffee than the rest of the population because of their disease. how does this affect our OR?
Hence the OR for coffee drinking was artificially increased due to the under-representation of coffee drinkers among controls.
A case-control study of Alzheimer's disease and exposure within the past five years to homecare products that contain aluminum will be conducted. Information about the use of products will be gathered via an in-home survey. what type of bias is present?
Individuals with Alzheimer's disease (cases) have memory problems and may not be able to remember past events, especially recent past events (recall bias). Therefore controls will better report past events than will cases, resulting in an underestimation of association with use of aluminum products
A case-control study of fatal automobile crashes will compare fatalities (cases) with living controls, assessing alcohol use and driving practices. Blood samples are always taken from cases (fatalities) by the forensic service in order to chemically ascertain the blood-alcohol content. However, data will be gathered from controls via interview. what type of bias is present?
Most likely the controls will underreport alcohol use, because drinking and driving is illegal (information bias.) Thus an overestimation of the association between alcohol use and fatal crashes is likely to occur.
cancer cases may try harder to recall prior exposures because they think they might be related to their disease. what type of bias? what will this do to OR?
Recall bias overestimation of a. overestimation of OR
A case-control study was conducted to assess the association of passive smoking and asthma. Newly diagnosed asthmatic individuals (cases) were compared with a random sample of individuals without asthma (controls) in regard to exposure to smoke from smokers at home or in the workplace for the previous 10 years. Which type and mode of bias could be introduced into this study?
Recall bias may occur. Passive smoking is not a factor important enough for asthma free individuals (controls) to remember as accurately (10 years recall) as cases may. Most asthmatics (cases) find troublesome to be exposed to smoking smoke and for a large proportion of them it may even be a trigger of asthmatic exacerbations.
define and contrast selection bias and sampling error
Sampling error is the part of the estimation error of a parameter from a sample caused by the random nature of the sample. Random error can be reduced by increasing the sample size (Last, 2001). Selection bias is a systematic error that causes a distortion in the estimate of a parameter and is caused by the manner in which the subjects are selected from the total population into the sample. Bias cannot be reduced by increasing the sample size.
define and contrast selection and information bias.
Selection bias is a distortion in the estimate of association between risk factor and disease that results from how the subjects are selected for the study. Selection bias could occur because the sampling frame is sufficiently different from the target population, or it could occur because the sampling procedure cannot be expected to deliver a sample that is a mirror image of the sampling frame. Information bias occurs when the information obtained from study subjects is systematically inaccurate regarding the disease or exposure under study
A study to assess the association of diabetes and smoking compared a group of hospitalized individuals with diabetes (cases) with a group of volunteer individuals without diabetes (controls) who were full-time employees of the same hospital where the cases were identified. The results from this study reported, for the first time in the literature, a strong association between diabetes and smoking. What type of bias may be present? Why do you suspect the presence of the bias you have identified? 6. The magnitude of this association is likely to be either over- or underestimated. Which do you think is the case, and what makes you think so?
Selection bias. There is evidence in many studies showing that responders, especially volunteer individuals, usually are healthier and are less exposed to risk factors such as smoking (response bias). Smoking increases the chances to be hospitalized for reasons other than diabetes. When hospitalized, it is more likely to be detected as having diabetes (detection bias) and more likely to be included in the study (Berkson's bias). The magnitude of the association is likely to be overestimated because smokers will be overrepresented among cases and underrepresented among controls as compared to the total population
healthy worker effect could cause selection bias how?
The "health worker" effect is really a special type of selection bias that occurs in cohort studies of occupational exposures when the general population is used as the comparison group. The general population consists of both healthy people and unhealthy people. Those who are not healthy are less likely to be employed, while the employed work force tends to have fewer sick people. Moreover, people with severe illnesses would be most likely to be excluded from employment, but not from the general population. As a result, comparisons of mortality rates between an employed group and the general population will be biased.
what is chance ?
a concern when the sample used to represent the whole population (the smaller the sample more likely opportunity for chance finding. due to random error and non-homogeneity. quantified using pvalue and confidence interval
what is a way to detect information bias?
a validation study
what are the sources of bias
any stage of the study 1. before the study (design of the study) 2. during the study (selecting patients, collecting data) 3. after the study (analyzing data, publishing )
exclusion bias
applying different eligibility criteria to cases / controls
what is analytic bias?
bias can occur in the analysis of the study an investigator may analyze the data in different fashions until a desired result is achieved. it is best to have a predefined analysis plan to prevent this.
how to determine if bias is present ?
bias cannot be quantified effects are difficult to determine important to design and conduct your study so that every possible instance where bias could be introduced is anticipated to minimize the occurrence.
how is bias manifested in your study
bias is a property of the procedure or study as result is not bias - its either close to the truth or its not
what is bias
bias is any systematic error in design, conduct or analysis of a study that results in a mistaken estimate of an exposures effect on the risk of disease. somethings done differently. some bias is hard to avoid.
what is one way to avoid information bias in cohort studies
blinding to exposure status
selection bias in random control trials
compromised randomization compromised blinding withdrawals loss to follow up competing risks ( medicine gives you risk for something else and could still die but not from the exposure. if different in two groups could cause bias)
how can you control bias in data collection?
construction of specific instruments ex questionnaires, physical exams, forms of abstracting data from records. administering of instruments: blinding, standardized training, use of clearly written protocols.
what is differential information bias?
differ across comparison groups. can lead to an apparent association when one actually does not exist or can lead to an apparent lack of association when one does actually exist.
what is surveillance bias?
disease ascertainment may be better in the monitored population than in the general population.
non response bias
exposed cases are less or more likely to participate than exposed controls
selection bias in case-control studies
hospital - based vs. community based prevalent vs. incident participation
problems with hospital-based studies in terms of selection bias
hospital based studies controls may not be representative of all persons without disease or if exposure leads to cases being detected
how does information bias manifest in a cohort study?
in cohort studies the assessment of exposure status is likely to be assess similarly for all persons but if the we follow the exposed and non exposed differently we assess the development of disease differently for exposed and non exposed.
what are types of observer bias
interviewer bias biased follow up
what are the types of information bias
interviewer bias recall bias reporting bias surveillance bias
example: investigators probe cases about potential risk factors. how does this affect our OR and what type of bias will this cause?
interviewer bias. overestimation of a. will lead to overestimation of OR
what are the sources of information bias
interviewers, participants, and measurement error
how to detect selection bias in your study
is present when the probability of being selected is different based on a relevant study characteristic (the exposure or outcome of interest) look for difference (low or unequal ) in repsonse rates between cases and controls
what is the problem with non random systematic error
it is a threat to internal validity
what are the causes of information bias?
knowledge of the exposure status influences the intensity and outcome of the diagnostic process. knowledge of the disease status influences the intensity and outcome of the diagnostic process. cases are more likely to recall exposure differently than the controls.
selection bias in cohort studies
loss of follow up withdrawals health worker effect self-selection
what are the biggest causes of selection bias in cohort studies?
loss to follow up and self selection of exposure
why does loss to follow up lead to selection bias? why do we have loss to follow up in cohort studies?
loss to follow up leads to selection bias because those who remain enrolled in a study might be different from those who are lost to follow up (this is usually the case) reasons for loss to follow up in cohort studies loss of interest migration from area they die from outcome general refusal
what are sources of recall bias in case control studies?
may be a problem because cases may think why did this happen to me. cases might be taking the interview more seriously than controls. cases might more easily remember their exposure status because of their diagnosis.
What are sources of interviewer bias in a case control studies ?
may be a problem if the interviewer knows the case control status they could assess them differently. they are not blinded. although people with the disease can be classified as a control it is more common that non-exposed cases may be classified as exposed.
what types of bias are more common in cohort studies?
misclassification selection bias: loss to follow up - if large doubts about validity of result.
what is publication bias?
more likely to publish favorable results. or if the media is interested in a particular disease.
what are the two categories that information bias falls into?
non differential and differential
what are types of selection bias
non-response bias exclusion bias
what is information bias detailed?
noncomparable information is obtained from different study groups. systematic error in the measurement of information on exposure or outcome. erroneous classification of exposure/disease status for study participants (exposure identification bias and outcome identification bias) inaccuracies in data acquisition lead to systematic tendency for individuals selected for inclusion in the study to erroneously place in different exposure/disease categories from which then belong. we consistently call exposed non-exposed and vice versa. we consistently call cases controls and vice versa
how is random error different from bias?
not systematic. random error may be equivalent to low reliability. due to poor precision from non specific causes and not reproducible. systematic error is due to poor accuracy but with definite causes and is reproducible.
what are the places that information bias occurs
observer bias reporting bias serveillance/diagnostic bias
selection bias (incidence vs. prevalence
prevalent cases are survivors (don't represent all cases of the disease) exposures may change after development of disease prevalent cases may be easier to find solution: ask about behavior changes or when feasible, use incident cases.
selection bias in cross sectional studies
prevalent vs. incident participation
what are the two types of error
random error non random error (systematic)
what is the difference between random error and non random error in epidemiological studies
random error will decrease as sample size increases non random error will stay constant no matter the sample size
what is a type of reporting bias
recall bias
what is the most common information bias?
recall bias
example: mothers with birth defects will remember past exposures better than mother with healthy children. what type of bias is this? how will this effect our OR?
recall bias overestimation of a. will lead to overestimation of OR
simply put what is information bias?
refers to imperfect information on study factors (exposure, outcome, confounder) the method of collecting information on exposure, outcome, or other relevant factors induces errors
what is non differential information bias ?
same across comparison groups. likely to underestimate OR
what are the three broad categories of non random systematic error
selection bias information bias confounding
what is a way to control selection bias?
selection bias can be controlled by minimizing loss to follow up or by using a well defined reference population rather than populations from a particular institution
what is a way to detect selection bias?
selection bias can be detected through follow-up of losses in cohort studies or by evaluating participant selection in case-control studies
in which study design is selection bias the biggest problem? and which design is it not as common
selection bias can occur in any study design it is the biggest problem in case control designs and not as big of an issue in cohort studies because we start with exposed and unexposed the outcome has yet to occur.
what types of bias are more common in case-control studies?
selection bias. exposure and disease occurred. participation rates unequal for cases and controls. information bias: recall bias and misclassification
why does self selection of exposure lead to selection bias in some cases?
self-selection of exposure can be a problem if for example we are looking at if working in a bar causes higher risk of disease. those who work in bars are different from those who choose not to in many ways. drink more, smoke more etc. be aware of differences regarding self selection be sure to match on them or adjust for them the health worker effect would also become a selection bias for a similar reason comparing the working population to those who are not could cause bias.
what are the two types of associations
spurious (not real) by chance by bias (error) real non causal (confounding) causal
what are possible ways to control information bias?
standardized protocol for data collection keep info gathering methods the same for all participants keep all study personnel unaware of participant exposure/outcome status
what is reporting bias?
subject reluctant to report exposure because of various reasons - potential of under reporting of exposure. also known as social desirability bias.
what is validity
the ability of a tests (in this case study) to measure what it purports to measure
where is the prevention of bias key?
the design phase. standardized protocol choice of study population cohort- occupation group or group with centralized source of info. good tools minimize loss to follow up
what is reliability
the extent to which the results obtained by a test are replicated if the test is repeated.
what is selection bias
the method of participant selection can distort the exposure - outcome relationship from that present in the target population. can effect any study design may increase or decrease observed associations
we investigate the risk of myocardial infraction. smokers are more closely followed by their doctors. how does this affect our OR?
this would cause an overestimate. the a or exposed cases would be higher.
We investigate risk of myocardial infarction using prevalent cases (exposure = cigarettes) Smokers may suffer sever MIs (more likely to die How does this affect our OR?
underestimate - using prevalent cases in this case would show an underestimate because. Smokers likely died from the MIs.
a case control study of prostate cancer. looking at the exposure of smoking. however only 65 percent of those who are approached agreed to participate. how would this affect our OR
underestimate. smokers might be less likely to participate. our concern would be that the prevalence of exposure is higher among cases who do not participate.
ways to prevent recall bias in case control studies?
use previously collected records ask questions that you know the answer to (validate)