Flashcards in Epidemiology Deck (86):
When to use case- control study?
1. Rare outcome (rare disease)
2. Multiple potential risk for a single outcome
3. There are association suspicious & hypothesis generating studies are needed
4. Expensive to diagnose or detecting outcome in study individuals
5. Long latent period between exposure & outcome
6. Resources & outcome are limited
Potential bias in case-control study?
1. Selection bias
2. Information bias
Misclassification (including heterogenous outcome)
Differential reporting of exposure data (including recall bias)
When to use cohort study?
1. Risk factor represents a rare event
2. Intent to study the multiple potential outcome of a single exposure
3. To generate incident rate
4. Necessary if limitations make other designs unfeasible
Steps in design a cohort study?
1. Define the hypothesis (-es)
2. Select study population (s) ( exposed & comparison group)
3. Exclude subjects not at risk
4. Ascertain exposure (including confounders)
5. Monitor for and ascertain outcome
Bias in cohort studies?
1. Selection bias
2. Loss to follow-up, esp. unequal in exposure groups
3. Ascertainment of exposure
When to use nested case-control study?
Cohort study provided the opportunity to perform nested case-control studies.
Once sufficient outcome endpoints have accrued, diseased individuals can be compared with those of free of disease, and exposure status can be determined retrospectively.
Most often used when a potential confounder is identified in the analysis as an important determinate of excess disease risk.
1. Essentially the same as cohort study, except the investigator decides who gets the exposure, using random assignment.
2. Strongest study design of all, maximizing internal validity (usually at the expense of external validity ).
Maximizing internal validity by promoting the equal distribution of potential confounders into exposed and unexposed groups
Source of bias in RTC?
1. Errors of allocation
2. Ascertainment of outcomes
3. Loss to follow-up
4. Inclusion of all relevant outcome
5. Cross-over, intent to treat analysis
6. Selection bias (now relates to external validity/generalization )
7. Errors of randomization and confounding
Advantages of RCT?
1. Strongest study design of all
2. Maximizing internal validity, usually at the expense of external validity
- maximizes internal validity by promoting the equal distribution of potential confounders into exposed and unexposed groups
1. Cohort study provides the opportunity to perform nested case control study
2. The cases (those who developed disease in the cohort) are identified
3. The control group is selected by randomly sampling from this source population (cohort)
When is nested case-control study used?
1. Often used in occupational epidemiology
2. Most often used when a potential confounder is identified in the analysis as an important determinate of excess disease risk
3. A few subjects are needed, less expensive than using the entire cohort
Prospective cohort study?
The investigator identifies the cohort to be studied at the beginning of the study and follows the subjects to specific end points, determining whether or not the subjects develop the disease or outcome of interest during the specified period.
In this design, exposure is ascertained as it occurs during the study.
Retrospective cohort study?
1. Also referred to as a historical cohort study.
2. The study compares exposed and unexposed groups; historical data (previously collected) are used and examined.
3. Exposure may be ascertained from past records and the outcome ascertained at the time of the study.
Combination prospective and retrospective cohort study?
1. Exposure is ascertained from objective records in the past, as in the historical cohort study,
2. The outcome is then measured during the time period that the subjects are followed.
1. a retrospective cohort study is based on information collected about events that occurred in the past.
2. A prospective cohort study begins in the present, and the data collection and outcome assessment are conducted over time, as the population is followed.
1. Groups are identified on the basis of the presence or absence of disease or other outcome of interests
2. The search for exposure is retrospective
3. Both cases and controls are classified as either exposed or unexposed
4. The proportion of exposed cases is then compared to the proportion of unexposed cases
5. Odds ratio identifies the odds of exposure among cases, compared to the odds of exposure among controls
1. Compare disease incidence overtime between groups that differ on exposure
2. Prospective cohort - study was initiated before any outcomes occurred
3. Retrospective cohort - cases had already occurred by the time study began
4. Provides absolute incidence rates
1. Exposure and outcome are ascertained at the same point or period of time
2. Provide prevalence
1. Systemic error
- A process at any stage of inference tending to produce results that depart from the truth
- A property of procedure, study or statistic
- Impact internal validity. Efforts should be made to reduce or eliminate bias.
Types of bias
1. Selection bias
2. Information bias
(S.I.C of Bias)
Types of selection bias
1. Non-response bias
2. Exclusion bias
3. Selective survival (surviving differ from those dying)
4. Detection bias (exposure leads to disease detection and subject inclusion)
5. Loss to follow-up
Types of information bias
Systematic differences in the way exposure and /or outcome data is obtained from the study groups
eg.bias From interview format.eg phone, in-person.
Bias from surrogate interviews
2. Measurement bias:
error inherent in tools.
Eg.Instrumentation calibration, lab, data entry, missing data
3. Recall bias:
affected people may respond differently concerning prior exposures
4. Surveillance bias
- the population that is monitored behaves differently compared to those who are not
5. Reporting bias
6. Bias in abstracting records
1. Associated with exposure (risk) and outcome
2. An independent risk factor for the outcome
3. Not in the causal pathway between the risk factor and disease
1. Compare crude and adjusted point estimates
2. A difference of 10% or more indicates confounding
Deciding Cutoff for sensitivity and specificity
1. Determined by the Cost of false positive vs false negative
2. Importance of not missing a case
- chose higher sensitivity
3. For confirmation diagnoses
- chose higher specificity
4. Low prevalence of disease, chose higher specificity, otherwise too many false positives
5. High prevalence of disease, chose higher sensitivity, otherwise too many false negatives
Test positive in all diseased population
Sensitivity = TP /(TP+FN)
Test negative in non-diseased population
specificity = TN/ (FP+TN)
Receiver operating characteristic curve (ROC)
1.Plots sensitivity (true positive rate) against 1- specificity (false positive rate) at various cut-off points across the range of a measurement outcome produced by a test.
2. ROC curve shows sensitivity and specificity for all possible cutoff values.
ROC and test
Area under the curve (AUC) is a single summary measure of test accuracy
- More accurate a test is, the farther toward the upper left its curve falls on the ROC plot
- If more than one test is plotted on the same graph, ROC analysis makes it sole to evaluate which test performs best, usually the one with the greatest area under its curve
Likelihood that person screening positive does have disease (PPV = TP /( TP+FP)
Likelihood that a person screening negative does not have disease (NPV = TN/(TN+FN)
Prevalence and PPV
Increase in prevalence, increase PPV
- screening most efficient if targeted to high-risk population
For infrequent disease, PPV increases more if specificity increases than if sensitivity increases
1. Person considered positive only if tests positive on all tests
2. Use very sensitive first test to pick all cases as well as false positives
3. Use very specific second test to eliminate false positives
Sequential tests enhances specificity since harder to be positive
1. Person considered positive if tests positive on any test
2. Person considered negative only if negative on all tests
3. Order of test not important since all tests required
Simultaneous test enhances sensitivity since easier to be positive, harder to be excluded
1. To make two rates comparable, they need to be standardized.
2. Direct adjustment or indirect adjustment can be used depends what is available
Knowing sample rate, Use a standard population to adjust your rate.
1. If having age-specific rates in the sample, using standardized population to calculate expected cases in each age group, and total expected cases in the standardized population (by summary).
2. Then, you can calculated the age-adjusted rate:
= expected cases in the standardized population/total number of the standardized population
3. Standardize the other group using the same standardized population.
4. Then, you can compare the two adjusted rate.
Knowing observed cases number, Use a standard rate to find out expected cases in the same population.
1. Calculate expected cases by using standard rate: you have number of cases in two groups (observed cases), by using a standardized rate, eg.US age specific death rate and the population in that age group, you could calculated expected cases in that age group, by summarize all age groups, you can get all the observed and expected cases
2. SMR (standardized mortality ratio = summary observed cases / summary expected cases
3. Compare the two SMRs
Comparison of crude rate and adjusted rate
1. Actual summary rate
2. Readily calculable
1. Difficult to interpret because of differences in population structure
1. Provide a summary figure
2. Controls confounders
3. Permits group comparison
1. Fictional rate
2. Magnitude depends on population standard
3. Hides subgroup differences
Lead time bias
Lead time is the period between early detection and the clinical presentation of disease.
The inclusion of this period could result in falsely prolonged survival periods for pts screened compared with those not screened.
Length time bias
Slow growing cancer would be picked up more likely by screening test. It may result in over representation of this type of disease in screening studies.
Rapidly progressing cancers will be less likely to be detected by screen test.
A way of selecting subjects that are comparable with respect to specific variables
Subjects stratified into relatively homogeneous strata. The comparison between groups can occur within each stratum.
Population attributable risk, given prevalence
PAR = AR (attributable risk) x prevalence
Prevalence, what study?
RR, incidence density, cumulative incidence rates
All apply only to cohort studies
Applies to case-control studies and cross-sectional studies
NNT = 1/Attributable risk
(Attributable risk = absolute risk reduction)
Also called absolute risk reduction
Confounder and Effect modifier (EM)
To distinguish, using stratification
1. RR1 and RR2 are different from RR,
1.1. Adjusted RR is also different from crude RR
- Both confounder and effect modifier
1.2. Adjusted RR is the same as the crude RR,
- EM, but not confounder
2. If both RR1, RR2, adjusted RR and crude RR are all the same
- neither a confounder nor an EM
3. If RR1 and RR2 are the same, the adjusted RR is different from the crude RR,
- a confounder, but not a EM
Attributable risk (AR)
AR = incidence in exposed - incidence in unexposed
Attributable risk percent
= Attributable rate percent
= Attributable fraction
= Etiologic fraction
The amount of disease in the exposed group that is due to the exposure
AR% (ARP) = (incidence in exposed - incidence in unexposed)/incidence in exposed x 100%
PAR (Population attributable risk)
The rate of disease in the population is due to the exposure
PAR = total incidence - incidence in unexposed
= (a+c)/(a+b+c+d) - c/(c+d)
Population attributable risk percent
= (total incidence - incidence in unexposed)/total incidence x 100%
Measures of access risk
PAR (population AR)
Incidence (cumulative incidence)
= attack rate
= risk of disease
= probability of getting disease
New cases in people who were initially at risk
New cases /person-time at risk during the time period
Number of existing cases of a disease at specific time divided by the size of base population at that time
Prevalence at a single point in time
The prevalence measured for a specific time interval
Point-source (common-source) outbreak epidemic curve
eg. guest at a wedding reception
- a single hump with rapid rise and slower return to baseline
Propagated-source outbreak epidemic curve
Also known as person-to-person outbreak (eg community outbreak of shigellosis)
- slow progressive rise
Ongoing-source outbreak epidemic curve (eg food continuously contaminated by food handlers)
Rapid rise (at onset of exposure ) to persistent epidemic level.
How to reduce confounding?
1. By design of the study:
1. By randomization
2. By restriction/filtration
3. By matching (in case-control study)
2. By analysis
2. Multivariate statistical techniques to adjust
An ecological study
- Is an epidemiological study in which the unit of analysis is a population rather than an individual.
- It is susceptible to ecology fallacy.
- In an ecological study, there is no information available about the individual members of the population.
Epidemiological curve with multiple humps
- 1st peak: primary attack rate for those exposed to the index case
- subsequent peaks represent 2ndary and beyond attack rates as those in the 1st peak infect others
= 1 - attack rate in vaccinated/attack rate in unvaccinated
= 1 - incidence rate in vaccinated / incidence rate in unvaccinated
- Is a relatively minor and continuous change in the structure of a virus.
- eg. The recurrence of influenza A epidemics every two to three years.
- is a sudden and major change in the antigenic structure of virus, occurring at longer intervals than with antigenic drift
- is typically associated with pandemics.
Occasional cases at irregular interval
eg. Tularemia, Rabies
- low level (usual), expected frequency of disease occurrence; the constant presence of a disease within a given geographical area (usual prevalence)
- A gradual increase in the frequency of disease occurrence above the endemic level
A sudden increase in the frequency of disease occurrence above endemic level (clearly in excess of expected level)
- an epidemic occurrence across continents
Types of infections in a population
Plot of number of cases against time
Types or patterns of epidemiology cure
1. Point source or common source
2. Ongoing source or continuos source
3. Propagated epidemiological curve
4. Epidemiological curve with multiple humps
Secondary attack rate
- measures the rate of spread of disease within an exposed group
- Secondary attack rate
= (# new cases - # initial cases)/(# susceptible - # initial cases)
= # of people dying in a specific time period/ average # of people alive during that period of time
1. Chance is a random error.
2. Come into play when using samples to represent a population
3. Random variation/error
- divergence of an observation on a sample from the true population due to chance alone
- statistics can be used to estimate and control the probability of chance
- cannot be totally eliminated
- Nuisance effect which one wishes to eliminate
- depends solely on whether the factor is evenly distributed between the study groups
Interaction (effect modification)
- characteristic of nature that exists independent of any study design or subjects
- it is to be described, reported, not controlled
Information bias can lead to
- misclassification of exposure status (exposed v non-exposed)
- misclassification of disease status (case v control)
Misclassification may occur in two forms
1. Differential misclassification
2. Non-differential misclassification
Differential misclassification bias
Can lead to either
1. an apparent association even if one does not really exist or
2. to an apparent lack of association when one does in fact exist
Non-differential misclassification bias
RR or OR tends to be diluted with Non-differential misclassification bias. We are less likely to detect an association even if one really exists.