Block 6 Flashcards
T/F: Measures of association take into account random error
False
Statistical inference
The process of drawing conclusions about a population based on data from a sample of that population.
It allows us to deal with random error.
What allows us to deal with random error
Statistical inference.
Two methods of statistical inference in analytical epidemiology
1) Interval estimation (confidence intervals)
2) Hypothesis testing
Interval estimation
95% confidence intervals can be calculated for measures of association
Width of the 95% confidence interval is an indication of ____
the precision of the estimate
The [narrower or wider] the confidence interval, the more precise the estimate
Narrower
3 steps of hypothesis testing
1) Specify a “null” and an “alternative” hypothesis
2) Compare the results that were expected under the null hypothesis with the actual observed results (this is done with a statistical test)
3) Make a decision (this decision is based on the p-value of the statistical test)
Null hypothesis (H0)
No association between exposure and disease
Alternative hypothesis (HA)
There is an association between exposure and disease
Ecological Study Pros and Cons
Pros: Can be done quickly and inexpensively (often use existing data)
Cons: ‘Ecologic fallacy’ - relationships observed at the population level may not hold true at the individual level
Cross sectional study Pros and Cons
Pros: Fast, relatively cheap
Cons: Because presence/absence of exposure and disease are assessed at the same time in subjects, it is often not possible to determine which came first: exposure or disease?.
Identifies on prevalent (existing cases) – we might be measuring the association of the exposure with the duration of disease, and not the risk of disease.
Prevalence is a function of ____ and ____
Incidence and duration of disease
Prospective cohort study Pros (4) and Cons (5)
Pros:
1) Exposure is known to precede the outcome
2) Allow calculation of incidence (risk)
3) Facilitate study of rare exposures (actively recruit subjects with the exposure)
4) Can look at associations with multiple outcomes in the same study
Cons:
1) May have to follow large numbers of subjects for a long time
2) Can be very expensive and time consuming
3) Not good for rare disease
4) Not good for disease with a long latency
5) Differential loss to follow up can introduce bias
Retrospective cohort study Pros (5) and Cons (4)
Pros:
1) Can be done over a short time (don’t have to wait for disease to occur)
2) Generally less expensive than a prospective study
3) Allow calculation of incidence (risk)
4) Facilitate study of rare exposures (actively recruit subjects with the exposure)
5) Can look at association with multiple outcomes in the same study
Cons:
1) Temporal relationship between exposure and outcome is sometimes hard to establish (did exposure happen before outcome?)
2) Not good for rare diseases
3) Differential loss to follow up can introduce bias
4) Requires access to good medical records
Case control study pros (4) and cons (4)
Pros:
1) Can be done over a short time (disease has already occurred)
2) Relatively inexpensive
3) Facilitate study of rare diseases (actively recruit subjects with the disease)
4) Can look at associations with multiple exposures in the same study
Cons:
1) Selection of appropriate controls is essential (and often difficult)
2) Temporal relationship between exposure and outcome is sometimes hard to establish (did exposure happen before outcome?)
3) Not good for rare exposures
4) Depends upon accurate assessment of exposures that happened in the past (recall bias)
Randomized control trial pros (3) and cons (4)
Pros:
1) Well-controlled studies are essentially free of bias and confounding (random allocation, blind or double-blind)
2) Exposure is known to precede the outcome
3) Allow calculation of incidence (risk)
Cons
1) Expensive and very narrow in scope
2) Not always ethical to randomly allocate individuals to treatment
3) Not good for diseases with a long latency
4) Different loss to follow up can introduce bias
Two potential source of error in analytical studies
1) Random error
2) Systematic error
Random error (3)
1) Random error is the divergence, due to sampling variation, of the measure of association in the sample from the true measure of association in the population
2) Does not bias a study. A study with a lot of random error may be wrong but we don’t call it biased
3) Statistical inference deals with random error
Systematic error (4)
1) Error that is inherent to the study method being used, which would result in a predictable and repeatable error if the study were repeated using the same method
2) Not caused by chance
3) Biases a study
4) No formal method to deal with systematic error
T/F: there is no formal method to deal with systematic error
True
T/F: systematic error does not bias a study
False, systematic error does bias a study
Validity
Refers to the absence of systematic error in a study result
A valid measure of association
A valid measure of association will have the same value as the true measure in the source population, except for error due to random variation
Bias
The extent to which a measure of association from a study differs from the true measure of association in the source population, due to systematic error. Systematic error causes bias. Bias affect validity.
T/F: Bias affects validity
True
Study population
Subjects in the study
Selected from a source population
The ‘n’ of the study is the number of subjects in the study population
Source population
Population from which the subjects were drawn
Target population
Population to which we may want to generalize our results
Internal validity
Ability to make correct inferences about the association of interest in the source population (based on the observations in the study population)
External validity
Ability to make correct inferences about the association of interest in populations beyond the source population
Which is more important? Internal validity or external validity
Internal validity
Three major types of bias
1) Selection bias
2) Misclassification bias (=information bias)
3) Confounding bias
Selection bias
Due to factors that affect the selection of study subjects, their participation in the study and/or their completion of the study.
It occurs when the measure of association for the individuals who were selected for, participated in, and completed the study is different from the true measure of association we would have obtained had we done the study on the entire source population
Misclassification (information) bias
Due to factors that affect the accuracy of information on the exposure and outcome
Results in incorrect classification of exposure or outcome status:
-Subjects with no disease are classified as diseased (and vice versa)
-Subjects with no exposure are classified as exposed (and vice versa)
Confounding
A distortion of the underlying relationship between an exposure and an outcome by a third factor
-The third factor influences both the exposure and the outcome, distorting the measure of association
-The distortion can be large and lead to overestimation or underestimation of an association; it can even change the apparent direction of an association
Reducing confounding variables BEFORE the study starts
- Match the study (esp in case-control studies): select cases and controls so that the confounding factor is equally represented in both groups
- Restriction: do not enroll any animals that have the confounding factor
- Randomization: will reduce confounding
Reducing confounding variables AFTER the study has been completed (if you realize there is a confounding variable)
-Deal with the confounding factor in the analysis by stratifying
-Stratify: partition the results based on the confounding factor
(ie: if sex is the confounder, split the data into males and females and do the analysis on each group)
(ie: if practice type is the confounder, split the data into small animal and food animal and do the analysis on each group)
