Prevalence Bias vs Late-Look Bias

[Ronin786]

Can anyone explain the difference between the two of these?

In Prevalence bias ( a form of selection bias) you're selecting a population that doesn't include everyone because the worst cases have died, or there is a high fatality rate.

On the other hand Late-Look bias means the people responding to your survey don't include the dead people. I also read that Recall Bias is a form of Late-Look bias? Is that true and how so? I especially don't get that part about recall bias, because based on my understanding recall is a procedural bias while late-look is a selectional bias.

Thanks.

---

Members don't see this ad.

 

[Phloston]

Can anyone explain the difference between the two of these?

In Prevalence bias ( a form of selection bias) you're selecting a population that doesn't include everyone because the worst cases have died, or there is a high fatality rate.

On the other hand Late-Look bias means the people responding to your survey don't include the dead people. I also read that Recall Bias is a form of Late-Look bias? Is that true and how so? I especially don't get that part about recall bias, because based on my understanding recall is a procedural bias while late-look is a selectional bias.

Thanks.

I've never heard of "prevalence bias," but if you're SELECTING groups in any way, then the randomness element is eliminated, and there's selection bias. Selection bias also occurs when there's losses due to non-follow-up.

I believe that if you encounter a vignette/question stem that asks for bias in relation to a terminal disease, late-look is correct over selection bias for the reason you've given.

I've never heard of recall bias as a form of late-look bias. Recall bias is one's knowledge of his or her disease influences his or her recollection of the past.

Side note: I've encountered in a practice question that asking patients too early for their opinions is a form of late-look bias (e.g. asking cardiac valve replacement patients their satisfaction of the surgical experience 2 hours after surgery versus 2 months later; the bias is that they are more likely to be unsatisfied in the former group because of the nascent pain induced by the surgery). I was a bit confused as to how that's "late-look," but apparently it is.

 

[Ronin786]

I got prevalence bias from here:

Selection Bias

Prevalence Bias: Occurs when a disease or condition is characterized by early fatalities, as some subjects die before they are diagnosed. Prevalence bias may also result if there is a time gap between exposure / onset of the condition and selection of subjects, and the "worst" cases have died.
Admission Rate Bias (Berkson's Fallacy): Occurs when patient admission rates vary between the experimental and control groups, especially in studies that use hospitalized patients for both groups. For example, this type of bias is present when hospitalized patients with the risk factor (experimental group) are admitted at a higher rate than the control group.
Nonresponse Bias:This type of bias occurs most often with surveys. If there is no follow-up of people who did not respond to the survey, it is difficult to determine if the responses received are typical of the general population.
Membership Bias: Occurs when there are pre-existing groups, because often, one or more of the characteristics that cause the people to belong to the groups are related to the outcome of interest. For example, in studying the clinical effects of smoking on cancer, it may be that smoking is not the actual cause of cancer, but rather a trait that is more common among smokers.
Procedural Selection Bias: Frequently occurs in studies that are not randomized or in studies using historical or external controls. Treatment assignments are made on the basis of certain patient characteristics, which results in dissimilarity between treatment groups.
Procedural Bias occurs when study groups are not treated similarly. This may happen when one group receives more attention or care, which may change the attitude of one group. This causes an imbalance in the study results.

Recall Bias: Occurs when subjects are asked to recall certain events. Subjects in one group may be more likely to remember the event than those in the other group.
Detection Bias: Occurs with the advent of new technology and equipment for detection of disease. If new technology allows for earlier detection, survival for these patients will appear to be longer because the condition was diagnosed earlier.
Compliance Bias: Occurs when patients find it easier to comply with one treatment as compared to another. This may cause data to be skewed towards the favorable treatment, even it is not as effective.

http://www.amfoundation.org/medresearch.htm

I'm asking because of this question (Disclaimer, I found it online dated back to 2010 but I don't know if its from UWorld or NBME or anything):

In a population with a high incidence of cardiovascular disease, diabetics are at least twice as likely to die from myocardial infarction as are non-diabetics. A case-control study conducted in the community identifies 1,000 people with sustained myocardial infarction and 1,000 people without sustained myocardial infarction. The subjects are asked whether they have a history of diabetes mellitus. According to the study results, diabetes has a protective effect against myocardial infarction. Which of the following best explains the observed study results?
A. Latent period
B. Selection bias
C. Observer bias
D. Hawthorne effect
E. Recall bias

In this case the answer was Selection Bias, because the group you selected did no include anybody that had diabetes, an MI and managed to live (due to the high mortality of MI in diabetes). However, I don't see how this is any different than Late-Look bias, where you would only be asking questions in people that aren't dead.

 

[thehundredthone]

Selection bias has got to do with the assignment of patients to study groups. Assigning people who've sustained MI to your study group inherently biases it because you've likely only chosen people who've survived (i.e. less likely to have had comorbid conditions that increased chances of mortality).

Late look/prevalence-incidence/Neyman bias is information retrieval from either group at the wrong time. If a disease is rapidly fatal you can only get information/select subjects from the more indolent cases of that disease.

So yes, the late look bias is a type of selection bias. Remember that the term late look bias applies mainly to cross sectional studies.

In a more simplistic sense, selection bias indicates that the two groups are not comparable. Late look bias accounts for selection bias in certain situations (specifically cross sectional studies).

This might help
http://forums.studentdoctor.net/showthread.php?t=745483
http://canarydatabase.org/about/study_methodologies?meth=cross_sectional
http://en.wikipedia.org/wiki/Experimenter's_bias
http://medicine.ucsf.edu/education/resed/articles/Lancet11_bias.pdf

EDIT: If you want to get really nitpicky, Neyman/prevalence bias can occur if a disease is rapidly fatal (causing late look bias), transient or subclinical (length bias).

 

[Phloston]

Selection bias occurs not only with non-randomization, but also with losses due to non-follow-up, so if patients die before the end of the study and their info is not included in the final results, that would create a form of selection bias. If patients/participants leave a study before it is over, they must still be included in the results, or else selection bias occurs (or if they die and they are not included: prevalence bias).

The difference between prevalence and late-look bias, therefore, is that the latter occurs even when numbers are fixed; the former relies on the participant #s changing.

 

[thehundredthone]

The difference between prevalence and late-look bias, therefore, is that the latter occurs even when numbers are fixed; the former relies on the participant #s changing.

There is no difference between prevalence and late look bias, late look is merely a type of prevalence bias. The articles I have posted clearly state the same. Here's another.
http://www.epidemiology.ch/history/PDF bg/Sackett DL 1979 bias in analytic research.pdf

Prevalence bias leads to selection bias in a study. The late look is more applicable to a cross sectional study simply because in a longitudinal study one can still account for fatalities.