BIAS

what is the last type of Systematic Error (Bias)?

confounding

Information Bias is Introduced as a result of _____________ in assessment of exposure and/or outcome

measurement error

what are the three types of systematic errors?

selection information and confounding

Selection Bias is an Error due to systematic differences in characteristics between who?

those selected for study and those not

What are three things to do to reduce interviewer bias?

1. Blind interviewers to subject's status 2. Standardized training on study protocol and data collection forms 3. Use clear language in questionnaires

What are three things to do to reduce recall bias?

1. Memory aids 2. Blind participants to study goals 3. Use work or medical records for exposure history

what are two technique used to reduce Nonresponse Bias?

1. Select all cases in an area 2. Provide incentives to have a high participation rate. Entering study Follow-up

Example of Selection Bias: Hospital-based case control study of Risperdine and Breast Cancer Case - newly diagnosed breast cancer Control - elective surgery for a benign condition Authors excluded women with operations such as cholecystectomy, thyroidectomy, renal disease or vascular graft because they frequently took Risperdine for that condition

ISSUE: Did not exclude cases who had a similar condition as the controls-prevalence of Risperdine much lower in controls due to this exclusion and not due to any association with breast cancer

_________-Interviewer probes more thoroughly for exposure information in case than in control

Interviewer Bias EX: X-ray reader might look just a little hard to see if there is a tumor there if exposure status is known

_________-This can occur during recruitment or it can occur during followup. Error would be that somehow those who were Eligible for the study but who chose not to participate are different from those who are eligible and choose to participate.

Nonresponse

__________ - there is variability in the data itself.

Poor precision -So, for example there is some true underlying resting blood pressure for me; however, if I measure it today I'll get one reading and tomorrow I'll get another reading. This is just random variability in the data.

Can random error be prevented?

Precision can be improved by increasing the sample size. Sampling error you can try very hard to prevent but it can occur simply by chance - I would call this very bad luck. Variability in measurement could be reduced by taking multiple measurements, training staff to do the measurement all in the same way, etc.

What are some headlines you can remember that have been alarming or have turned out not to be true?

Problem is that one study gets published in isolation whereas in epidemiology it's not just one study but a collection of studies that gives evidence of an association between an exposure and an outcome. So, often it's a false alarm. Which means when there is a serious message, we may not be heard.

________ - Case may be more likely to recall exposure due to having the disease than the control who does not have the disease (When cases are more likely to recall an exposure than controls)

Recall Bias

what are the two types of information bias?

Recall Bias and Interviewer Bias

You know about a specific information bias that occurs in case-control studies. What is that bias called? Berkson's bias Recall bias Loss to follow-up None of the above

Recall bias

Strategy to reduce Berkson's bias:

Recruit controls from population rather than hospital

__________ - happens when sample is somehow different from the underlying population from which it was drawn (also known as really bad luck)

Sampling error

So what does that mean then in terms of the 2X2 table?

Some people are assigned to the exposed group who are really unexposed or vice versa. Or some people are assigned to the disease group who are really in the disease free group or vice versa. Error can be due to the observer, the responder or the instrument(s) used to make the measurement

what is a Systematic (Bias) error?

Systematic deviation of results from the truth

Iowa Women's Health Study - eligible pool of licensed female drivers, chose 50% random sample. Some women chose not to participate so the investigators had access to the participants and nonparticipants self-reported height and weight (off driver's license). Responders were on average 3 months younger and 0.38 kg/m2 lighter. Also slightly more likely to live in rural less affluent counties.

The SSN was linked to driver's license which was then used to determine cancer incidence/mortality for respondents and nonrespondents. Smoking-associated diseases were more frequent among nonrespondents than respondents, consistent with a lower response rate of smokers to a health survey.

what is a Random error?

Fluctuation around the true value of a parameter (RR) - random and often due to chance

1. This reduces recall bias. For example, if doing a 24 hour recall of foods, structure the interview to refer to particular meals and snack times. Can use food models to indicate portion sizes to help with quantity and posters of commonly eaten snack foods are helpful reminders. OC studies have used pictures of pill dispensers to help users ID brand names and formulations. 2.Even the most well intentioned research staff can introduce bias; best to blind them to status of study subject; case versus non case; exposure versus not exposed. Sometimes bias the subjects to true study goal in order to reduce the possibility of the subject providing answers to please the investigator by anticipating the "correct" answer or producing a socially desirable answer. 3.Take time to adequately train staff

4. Take time to develop these carefully so complete data are collected on all study participants 5. Ethics dictates that subjects be informed of the reason for the study and basis for invitation to participate, the specific hypothesis to be tested need not be revealed. 6. Generally written at a 6th grade level, pretesting done with focus groups to be sure questions are clear, all choices are given, watch for things which subjects are unwilling to answer.

what is one technique used to reduce Berkson's Bias?

Be sure exposure isn't associated with control selection

_________-Occurs when using cases and controls from hospitals

Berkson's Bias -controls' exposure history is more like cases than general population

You know about a specific selection bias that occurs in case-control studies. What is that bias called?

Berkson's bias

Processes lead to such deviation - systematic meaning methodical or orderly, efficient. You actually already know about a couple of forms of systematic error or bias already. Can you think of two biases we have studied?

Berkson's bias and Recall bias YES There are actually two different types of systematic errors we will be talking about in this class.

You know about a specific selection bias that occurs in cohort studies. What is that bias called? A. Berkson's bias B. Recall bias C. Loss to follow-up D. None of the above

C. Loss to follow-up

Difference in exposure/disease relationship between those selected to participate and those eligible but do not participate This can occur when?

Can occur: Before study begins During recruitment During follow-up

So, these are examples of random error. But what if we have error that is not random but the result of something involving the study design, conduct or analysis? term-10

This is call systematic error.

Example of Berkson's bias Study looking at association of smoking and respiratory disease Use hospitalized cases and hospitalized controls Controls are more likely to be smokers than the general population - why?

This occurs during recruitment of controls into the study choosing controls that have an exposure history different from the general population

Think about that - do you think that might sometimes happen? Who might not want to participate? Smokers have been shown to have a lower response rate. Ethnic minorities have a lower response rate. Who might want to participate? Those with higher education. Or who might choose to respond (sicker people looking for answers) as opposed to people who aren't as sick (they might be less interested in the study as time goes on).

Usually difficult to get information on those who don't participate or don't respond...

__________ - This is also variability in the data, but due to how something is measured. So, going back to the blood pressure example, there may be random variability in the way the blood pressure is measured by different people or even the same person a few different times depending on how they were trained.

Variability in measurement So, going back to the blood pressure example, there may be random variability in the way the blood pressure is measured by different people or even the same person a few different times depending on how they were trained.

Danger of getting your info solely from the media. Family Circle magazine - canola oil is out, Coconut oil is in... Very selective - they only report what they want to report, which may not necessarily be what the field believes is the most important, Often focus on the one positive result rather than the several negative results. Why is that? Because they need to publish newspapers. Despite peer review, flawed studies do appear in print (flawed methodologically) and get media attention. Cannot assume that just because it was published in the NEJM, it should not be questioned.

While the results of an epidemiological study may reflect the true effect of an exposure(s) on the development of the outcome under investigation, it should always be considered that the findings may in fact be due to an alternative explanation1. So...how do you figure out what to believe? Read the original article. Critically evaluate the article for selection of study subjects, measurement of exposure/outcome, analysis and interpretation of data. Despite the best laid plans of researchers, error is often present. Can be random or system. What is the difference between the two and can they be prevented?