The flashcards below were created by user
on FreezingBlue Flashcards.
1. What does case-control study identify?
2. How many points of observation?
3. How are groups determined?
4. How is exposure determined?
5. Does it provide incidence data?
6. Draw design of case-control study
1. Possible causes of disease by determining how the cases differ from controls based on exposure to some factor
3. Based on presence of absence of disease outcome
4. Exposure is determined retrospectively
5. Not directly.
What are advantages of case-control studies? (3)
What are limitations? (5)
- 1. Shorter time to complete vs. cohort studies
- 2. Smaller sample size required (cheaper)
- 3. For rare disease, may be only practical way to determine risk factors
- 4. Can study many exposures
- 1. Not good for rare exposures
- 2. Info on potential risk factor/important confounders may not be available
- 3. Recall bias (cases may search for cause of disease and be more likely to report exposure)
- 4. Cannot determine causality
- 5. Difficult to identify representative case and control groups
1. What is case-control study often followed up by?
2. What type of disease is it best suited for? (2)
3. When would you use multiple controls? After what point is it not effective anymore?
4. How do you measure effect? Formula? What is small when the disease is rare? What if the disease isn't rare?
5. What would the table look like?
6. What does an odds ratio of 2.2 mean in English? (2) using hip fractures and vitamin A as an example.
1. Provides potential leads for more definitive cohort study
2. One with rapid onset (so you know exactly which exposures led to the disease) and one that requires medical attention.
3. To increase power to find significant finding. Best to use multiple controls of DIFFERENT TYPES (i.e., not just other hospital patients, etc). After 3 controls per case.
4. Odds ratio - ad/bc (it's really (a/a+b)/(c/c+d) --> but when a and c are small, it becomes (a/b)/(c/d) --> ad/bc. If the disease is NOT rare, then you wouldn't use case-control
6. The odds of hip fracture are 2.2x greater in those exposed to vitamin A than in the unexposed OR the odds of using vitamin A are 2.2x greater in hip fracture cases than those in the neighborhood controls.
1. In case-control, why do we calculate odds ratio instead of risk? When do you use relative risk?
2. What does an OR = 1 mean? Greater than 1? Less than 1?
3. What does it provide a good estimate for? Under what conditions? (3)
4. For the following table, calculate how you would do relative risk vs. odds ratio?
5. Why is case-control cheaper than cohort?
6. What would we expect if exposure was associated with disease?
7. Is cancer a rare event?
1. We don't calculate risk because we must have incidence (separate timepoints of data collection) to have risk. Cohort studies (both prospective and retrospective)
2. OR = 1 means no association. OR > 1 means cases were twice as likely as controls to be exposed. OR <1 means cases were less as likely to be exposed as controls.
3. Relative risk, if (1) controls are representative of target population, (2) cases are representative of ALL cases, and (3) frequency of disease in population is small & disease is rare.
5. Because it requires less participants.
6. That proportion of cases who were exposed (a/a+c) to be greater than the proportion of controls who were exposed (b/b+d)
1. What do you need for a case?
2. What would ideally be the protocol for getting cases?
3. What are possible sources of cases? 4
4. How are cases selected? 2
1. Must be precisely defined
2. Identify and enroll ALL incident cases in a defined population in a specified time period.
3. Medical facilities (not always incident), tumor registry, vital statistics bureau, hospital patients
4. Should be incident vs. prevalent cases and should be aware of referral patterns when taking hospital patients
1. What are possible sources for controls? 3
2. What is the most important factor in selecting a control? From a " "
3. What is a problem in control selection? How can you deal with this problem?
4. What question must you ask of the controls?
5. How else do you control for confounding?
1. Patients from same hospital, relatives/friends of cases (SES control), non-hospitalized controls from neighborhood, community.
2. The control should be similar to the case in every way except for having the disease (from a "reference population"
3. Confounders - i.e., if cases are more likely to smoke, etc. By stratifying the results.
4. If their exposure is the same as the general population - perhaps we chose the wrong control group.
5. Matching cases and controls for potential confounders like age, race, sex, SES and occupation
1. In case-control studies, what are the two methods of matching? Which is easiest?
2. What are the types of analysis? (2) First stratify by confounders and then combine a+a, b+b, etc for each square in the table.
1. Group matching - selecting the controls so that a proportion of controls have the same characteristic as a proportion of the cases. This is easiest bc you don't have to turn away as many participants.
2. Individual matching - for each selected case, a control is selected who is similar to the case in terms of variables of concern.
1. Unmatched analysis (prone to confounding) and 2. Matched analysis - where you analyze pairs instead of individuals (that's what's represented by the squares). Odds ratio can be calculated the same way. (ad/bc).
1. What is protocol for recruiting and collecting data for case-control studies?
2. What is important in collecting info on exposure?
2. (1) Blinding of data collectors, (2) routine recording of variable of interest (same time, same way, etc), (3) physical measures are better than subjective ones, and (4) wording of questions should ensure that exposure preceded disease.
1. What is a nested case control study? Which are cases and which are controls?
2. Advantages? (3)
1. A case control study nested in a cohort. Cases are those that developed disease and controls are those who did not develop disease over time.
2. (1) data collected before disease develops --> no recall bias; (2) " " ---> likely pre-morbid factors (3) cheaper than analyzing ALL samples for all cohort members for the risk factor
3. None listed
1. What are case-cohort studies?
2. Advantage of this?
3. How does this differ from nested case-control?
1. Cases are those that developed disease (like nested case-control) but controls are randomly selected from cohort.
2. Because controls are not individually matched to each case, it is possible to study different diseases in the same case-cohort study using the same cohort for controls.
3. (1) cases and controls are NOT matched on calendar time and length of follow-up (2) controls are randomly selected from cohort
1. How are case-control studies analyzed? 2
2. What does it mean for an OR of 1.6 for history of benign breast disease?
1. Using logistical regression and then getting odds ratio from logistic regression.
2. The odds of being a case are elevated by 60% in women with a history of benign breast disease.
1. What is the purpose of a cohort study?
2. What type of individuals do you begin with? Then how do you categorize them? Then what?
3. When should you perform a cohort study? (3)
1. To examine the possible relationship bt an exposure and a disease.
2. Disease-free individuals and categorize them based on exposure. Then, follow them over time to determine INCIDENCE of disease in the exposed vs. non-exposed.
(1) Evidence of potential risk factors to be studied from other data (i.e., case control, cross-sectional)
(2) New agent that may alter risk for several diseases is introduced
(3) We can minimize potential for loss to follow-up --> short interval bt exposure & disease.
1. What can exposure be? (2) Define
2. What are the difficulties of establishing temporality in case-control and cross-sectinoal studies?
3. Are groups randomly assigned in cohort studies? By what?
1. (1) Common exposure associated with specific setting - i.e., secondhand smoke at occupational/school setting or (2) non-specific exposure (born in same year or living in same area).
2. Case-control: recall bias
cross-sectional: difficult to distinguish exposures from disease outcomes, esp if outcome marker is a biological or physiological parameter.
3. No - they are assigned based on exposure status!!!! It is unclear whether the exposure REALLY caused the disease.
1. What are the types of cohort studies? Define each.
2. Advantages of prospective 4 vs. retrospective? 4
3. Disadvantages of prospective 5 vs. retrospective? 3
4. Define historical prospective
1. Prospective - subjects are identified by exposure status and are disease-free --> disease status becomes apparent during follow-up period.
Retrospective - subjects are selected & analyzed based on exposure status, but follow-up has already taken place and disease status is known.
2. Advantages: Prospective --> exposure to risk factor is assessed before disease occurs, can estimate all measures of incidence and effect, possible to study several outcomes to one cause, and recall bias is not a problem
Retrospective --> can study rare exposures, less expensive than prospective, can be quick.
Prospective --> large investiment of time, resources, and finances, requires large sample size, dropout is a huge problem for long-term nature of study, cannot use for rare diseases, not easy to reproduce findings (difficult to look at consistency of effect).
Retrospective --> not good for rare diseases
, problems in ascertaining disease outcome
(duration & intensity)
4. Combo of 2 designs - exposure is from past, objective records but may gather more info over time. Cohort is followed into future.
Cohort study basics
1. Who is recruited? (2) What should this be limited to?
2. Observation points #? For what? How many are there usually?
3. Who should you collaborate with? Why? (2)
4. What are ways to determine exposure status? (4)
5. What is the Hawthorne Effect?
1. At risk individuals who are disease-free. It should be limited to group at HIGHER risk for disease
2. At least 2 - (1) to determine exposure status and eligibility and (2) determine # of incident cases, but there are usually more.
3. Other geographic areas to (1) increase sample size and (2) improve generalizability
4. Lab tests, physical measurements, questionnaires, special proceudres.
5. Hawthorne effect - be careful asking about exposure at every visit bc the study may influence the behavior.
1. What are the types of comparison groups? (2) Which is most likely? When would you use the other? What does it consist of?
2. Do cohort studies require active follow-up?
3. How should disease be measured?
1. Internal and external comparison group. Internal. When all are exposed (i.e, occupational). Comparison groups consist of general population from which the exposed population comes from.
2. Yes in prospective studies.
3. By blinded person unaware of exposure status, must be comparable for exposed and unexposed, predetermined intervals is ideal, pre-determined diagnostic criteria.
1. What is the primary measurement of disease? What does it represent?
2. How do you calculate this? Equation?
3. What is another measurement of risk/disease? Equation? What does this represent?
1. Cumulative incidence - the proportion of the subjects who develop the disease over the follow-up period of the study.
Cumulative incidence represents avg probability that an individual in the cohort will develop the disease over the follow-up period.
- Exposed Y a b a+b
- N c d c+d
- C Incidence in exposed: a/a+b
- C Incidence in unexposed: c/c+d
- Relative Risk (Risk Ratio) = (a/a+b)/(c/c+d)
- 3. Attributable Risk; Cumulative Incidence(exposed) - Cumulative Incidence (unexposed)
Attributable risk represents the total increase in the incidence of disease attributable to the exposure.
1. In cohort studies, what is more important for public health - absolute or relative measures?
2. Which is more important for causality?
3. Diff bt study and target population?
4. What is external validity/generalizability?
5. What is internal validity? What is it influenced by? (3)
1. Absolute measures bc they tell us how many people are prevented from getting disease if exposure is removed.
2. Relative measures - helps us determine strength of association.
3. Study population - actual members of study, target population - population to whom you'd like to generalize your results
4. External validity/generalizability - a study has good ext validity if it can be generalized to other populations
5. A study is internally valid if association bt exposure and disease is good estimate of "real life" exposure and disease association that exists in population.
Influenced by confounding, bias, and chance.
1. What is exchangeability?
2. How can this be achieved?
3. Why is there a greater risk of confounding in cohort studies?
4. What other types of bias that plague cohort studies? (3) Which is most common in retrospective cohorts?
1. Fulfilling the criteria that unexposed in a cohort represents the disease experience of the exposed group had the exposed group not been exposed.
2. Even distribution of potential confounders bt exposed and unexposed groups - aside from exposure itself, the groups are comparable in every relevant way (age, sex, gender, SES, etc).
3. Bc subjects have "chosen" their exposure so there is a greater chance for confounding/lack of exchangeability.
4. Outcome ascertainment bias (no blinding), information bias (quality of info obtained, esp possible in retrospective cohorts), and analytic bias.
Cohort studies: lost to follow-up
1. What happens if losses from group A are greater?
2. What happens if losses from group C are greater?
3. What does unequal loss from these groups result in?
4. What is the power of a study? How is this influenced by sample size?
5. What is used to analyze results from cohort study?
6. What options do you have if you start with a defined population vs. a cohort study that starts with exposed vs. unexposed groups?
1. Exposure and disease relationship will be underestimated
2. Overestimation of exposure and disease relationship.
3. Bias in relative and attributable risks.
4. The probability of finding a statistically significant association in the data, given that the association really exists in the study population. Increased sample size --> increased power.
5. Survival analysis - product limit analysis --> equals 1-probability of NOT developing the disease in any of hte time intervals.
6. If starting wtih defined population, you can examine multiple exposures and multiple outcomes, whereas if starting with exposed vs nonexposed groups, you can only study one exposure and multiple outcomes.