Use LEFT and RIGHT arrow keys to navigate between flashcards;
Use UP and DOWN arrow keys to flip the card;
H to show hint;
A reads text to speech;
32 Cards in this Set
- Front
- Back
- 3rd side (hint)
What is inference in epidemiologic studies?
|
Inference involves making generalizations about a larger group of individuals on the basis on a subset or a sample.
There is a possibility that the inference will be inaccurate or imprecise, simply because of chance or variability. -relative comparisons describe te strength of the association between the determinant and an exposure. |
|
|
When studying a sample of the population, the observed associations can be:
|
-Bias
-confounding -random error -a true association |
|
|
Remember, epidemiologists compare disease frequency in two ways:
|
absolute comparisons such as risk or rate differences and relative comparisons such as rate ratios and odds ratios.
|
|
|
Relative comparisons:
|
describe the strength of the association between a determinant and an exposure.
|
|
|
Validity:
|
The primary objective of most epidemiologic research is to obtain a valid estimate of an effect measure of interest
|
Whatever we are measuring, we want to know if it’s true. Bias can effect validity. Confounding can effect your validity. Random error is a precision thing.
|
|
Validity
|
The validity of a study is defined as "The degree to which the inference drawn from a study, especially generalizations extending beyond the study sample, are warranted when account is taken of the study methods, the representativeness of the study sample, and the nature of the population from which it is drawn" (Last, 1995)
|
Study design
Data collection Data analysis (Exposure to outcome) |
|
How to determine if something is valid:
|
there are no such imperfections, we say that the study is valid
If there are imperfections, then the extent of the distortion of the results from the correct conclusions is called bias Validity of a study is what we strive for, bias is what prevents us from obtaining valid results Internal validity must be established before the study can be generalized to the population |
If I get an OR of 10 and I have an OR of 1.3 – the ten is probably more accurate because even if its bias, we know that it is significant.
|
|
What is Bias?
|
“A systematic error in the design or conduct of study that leads to an erroneous association between the exposure and disease.”
Results in an incorrect or invalid estimate for the measure of association |
|
|
Three types of validity issues:
|
Selection bias
Observation bias (information bias) Confounding – next lecture |
Observation bias- after the person is in the study (also called information bias)
|
|
Validity versus precision:
|
Validity and precision concern two different sources of inaccuracy that can occur when estimating an exposure-disease relationship:
systematic error (a validity problem) and random error (a precision problem) systematic and random error can be distinguished in terms of shots at a target |
When we are talking about bias’ in epi, we are talking about something that is inherent to the data. But not something that is usually intentional.
|
|
Study validity encompasses 2 components:
|
Internal validity
A study is said to have internal validity when there have been proper selection of study groups and lack of error in measurement. External validity Implies the ability to generalize beyond the set of observations to some universal statement |
|
|
Selection Bias
|
Selection bias is systematic error that results from the way subjects are selected into the study or, because there are selective losses of subjects prior to data analysis.
|
Volunteer vs. Nonvolunteer
Form of cohersion? For case/control- our controls have to come from the same population as the case. Cohort and Experimental designs- loss to follow-up is selection bias. |
|
When can selection bias occur?
|
Can be introduced at any stage of a research study: design, implementation, or the analysis.
In the design phase, it may be introduced by the process of subject identification, including the definition of the population sampled In the implementation phase, problems of nonrandom cohort attrition (refusals, loss to follow-up) or missing data (due to protocol failure, design flaws) may give us data on a skewed sample In the analysis phase, approaches to analysis in the face of imperfect protocol compliance or methods of imputed data analysis when you have missing data may affect one select group or subgroup |
We also lose power when we have loss to follow up
|
|
Where does selection Bias most often occur?
|
Case-control studies
the primary source of selection bias is the manner in which cases, controls or both are selected and the extent to which exposure history influences such selection Control selection bias Self selection bias Differential surveillance, diagnosis, or referral bias |
Information bias is the more common form of bias.
|
|
selection bias, Control selection bias:
|
Controls are not selected from the same population that gave rise to the cases
Ex. Case control study for role of PAP smears in prevention of cervical cancer Cases selected from hospital medical records Controls selected from neighborhoods during day hours Exposed if had PAP smear in last we months In this case would bias towards the null Controlled by using same selection criteria for cases and controls |
Example is in book-
Biased toward the null- more likely to have no association. Null is 1 for RR and OR You can bias toward the null or away from the null We should be “12”…Months Selection bias- only during the day. Less likely to have insurance because less likely to have a job. The population is not representative of the group that gave rise to the cases. Bias is toward the null- if in my controls, fewer of them have gotten pap smears, then it is going to mask the association. |
|
Selection bias: self-selection bias:
|
Caused by
refusal or nonresponse of participants related to both exposure and disease Agreement to participate related to both exposure and disease Ex. If exposed cases are more likely to participate than nonexposed cases Controlled by: ensuring high participation rates |
You might have a study where someone has a disease that is more high risk so they are going to participate because they might see a benefit from it.
Every time you do a study design, you should look at inclusion/exclusion to look for bias. We can’t generalize if we have bias. |
|
Differential surveillance, diagnosis, or referral
|
Related to exposure
Ex. Case control study on venous thromboembolism and use of oral contraceptives Cases - hospitalized for thromboembolism Controls - hospitalized for acute illness or elective surgery Bias – link between OC use and venous thromboembolism already known Providers more likely to hospitalize symptomatic women on OC |
|
|
Selection Bias: Cohort studies and clinical trials
|
Cohort studies and clinical trials
the primary sources of selection bias are loss-to-follow-up, withdrawal from the study, or non-response Results when background frequency of the outcome in the groups is not the same, or if one group is more likely to develop the outcome Ex occupational cohorts – the exposed group is typically healthier (workers) than the control group (general population) Loss to follow-up bias can be differential or non-differential |
All three of the bias in cohort are going to cause the same problems.
Example: occupational cohort studies- healthy worker effect- if we use a group that works at lockheed martin and compare them to the general population. They work, which means they are probably going to be healthier. Differential means it differs by something Nondifferential- maybe its by the exposure |
|
Selection Bias: Non differential vs. Differential
|
Loss to follow-up
Nondifferential Loss in the disease or exposure status that is not related to the other “axis” Ex. Loss of diseased individuals that is equal among exposed and unexposed Results in decreased incidence in both exposed and unexposed groups – absolute measure Relative measure will not be affected |
IMPORTANT!
Loss to followup and have the same # of ppl that would have developed the disease If you lose the same amt ppl in both groups at the same time (same rate), it is non-differential and incidence will go down, because I’ve lost ppl that can develop the disease If I lose them at different rates, it is differential |
|
Selection Bias: Loss to follow up, differential
|
Differential
Losses are related to both disease and exposure Can bias absolute and relative measures towards or away from the null Ex. Loss of diseased individuals among exposed only Relative and absolute measures will be biased downwards Control by ensuring high follow-up rates for all groups |
It can bias both absolute and relative measures
Decreasing the measure- biased toward the null, if you lose ppl in your exposed group but not in the unexposed. If the opposite happens, then it will be biased away from the null. The association will be more significant. Applies to anything prospective. |
|
Selection Bias in Cross-Sectional studies
|
the primary source of selection bias is what is called selective survival. Only survivors can be included in cross-sectional studies. If exposed cases are more likely to survive longer than unexposed cases, or vice versa, the conclusions obtained from a cross-sectional study might be different than from an appropriate cohort study
|
|
|
Selection Bias: Berkson's Bias:
|
Type of bias involving hospital patients as cases and/or controls
Bias results when participants are more likely to have multiple diseases or risky health behavior Also called hospital admission bias Nonresponse bias Withdrawal bias Lost to follow-up bias Healthy worker effect Publication bias |
Studies where you don’t find anything are very unlikely to get published. In general only positive studies get published. Go to clinicaltrials.gov Any clinical trial that is done has to be published. You can find the negative result studies. |
|
Oberservation (information bias)
|
Results from systematic differences in which data on exposure or disease (outcome) are obtained.
Ex. Different techniques for interviewing cases and controls After subject enters the study Relates to how data are collected Results in incorrect classification of subjects (exposure or disease status) Recall bias, interviewer bias, misclassification bias |
Method should be the same- questionnaire vs. phone call. Might respond differently.
|
|
Obeservation (information Bias)-Recall Bias:
|
Occurs when individuals with a particular adverse health outcome remember and report their previous exposure experience differently from those who are not similarly affected, or when those who have been exposed to a potential hazard report subsequent events with a different degree of completeness or accuracy than those non exposed
Can bias towards or away from the null Examples? Ways to control for it? |
Those who are sick are going to recall their exposures better than someone who is not sick.
If my sick cases are more likely to remember their exposure than my non sick controls. I’m not going to find as much of an exposure in my controls. Biasing away from the null because I’m not going to remember as much exposure in my controls. Birth defects- someone who had a child with birth defects and without birth defects- and then ask women about illegal drug use. They will probably not tell you so there is going to be a bias toward the null. Less of a difference between them. Can control for the recall bias with blinding? What over-the-counter drugs have you taken with the past 6 months? IF you ask a more specific question (s) you are going to control a little more for bias. |
|
Observation bias/information bias:Interviewer Bias
|
any systematic difference in soliciting, recording, or interpreting information from study participants.
Ex. Interviewer is not blinded to disease status and questions subjects differently Ways to control for it? |
Control with blinding, control with scripts, not always possible to blind.
Training… if you ask them this question, this is how you are supposed to control. |
|
oberservation Bias: misclassification:
|
occurs when subjects are erroneously categorized with respect to the exposure or disease status
also called measurement error may be differential or nondifferential with respect to disease status Ex. Exposed person classified as unexposed May occur from not remembering exposures, using a broad exposure definition |
Recalling exposure can help misclassify you as well.
|
|
Differential Misclassification:
|
rate of misclassification differs in the different study groups (For example: women with a child with a malformation tends to remember more mild infections during their pregnancies than did mothers of normal infants resulting in an apparent association when there is none. This results in recall that is different in the diseased group than the non-diseased group.)
Differential misclassification can lead to an association when there is none or to an apparent lack of an association when one does exists |
If my diseased group does not get classified as exposed and my non-diseased group is classified toward the null.
Read the chapter twice through. |
|
Non-Differential Misclassification:
|
Results from the degree of inaccuracy that characterizes how information is obtained from any study group (cases and noncases or exposed and nonexposed persons).
Such misclassification is not related to exposure status or to case/control status. Nondifferential misclassification usual leads to a diluted effect of the RR or OR, and it is shifted toward the null (1.0). In other words, we are less likely to find an association when one exists. |
|
|
What possible biases have influenced the observed results?
|
What is the direction of the likely effect?
Is the true association masked? Is the association a spurious association where there is none? How great is the distortion? |
need to ask these questions if we think we have bias
|
|
Control of Bias:
|
Careful study design and implementation of the study protocol
Careful assessment of implications in selection of study population(s) cases and comparisons should come from same population base |
Assure complete follow-up of the population
Consider using multiple control groups Choose study groups that are representative of the target groups Prepare detailed manuals of procedures that covers all aspects of data collection |
|
More ways to control Bias:
|
In multi-center projects, use central facilities for interpreting and analyzing data. Use central laboratories for analysis of samples
Maintain tight quality control procedures (send replicates for analysis, blinding technicians, review of certain cases) Blind data collectors and abstractor where possible Train and certify all data collectors Retrain data staff periodically |
You might have different procedures, references, if you send it to different labs, that might introduce a bias because of the lab.
|
|
Even more ways to control Bias:
|
Establish incentives for high participation rates
Sources of data collection on all subjects should be similar Participants should be unaware of specific hypotheses under investigation Outcome data should be obtained without knowledge of exposure status Use of preexisting records Use standardized definitions of exposure and disease status Enroll all cases in a defined time and region |
You don’t want to coerce/do something unethical. But do have incentives.
|