Basic concepts Flashcards
Relationship between prevalence and incidence rate
Prevalence is proporational to the incidence mutiplied by the average duration of the disease.
Objectives of epidemiology
- Investigation, monitoring and surveillance of disease of known and unknown origin.
- Understand the ecology and natural history of disease (i.e. risk factors)
- Evalute tests, treatments, disease control programs in terms of performance and economic benefits.
Temporal patterns of disease within a region and between regions
Sporadic (no pattern), endemic (constant low level, predictable), epidemic (beyond expected level, not predictable), pandemic (an epidemic that crosses international borders).
Risk and open populations
Cannot calculate risk for an open population, but it can be estimated from the incidence rate.
Three types of bias in screening
- Lead time (longer survival due to early diagnosis)
- Length time (longer survival due to milder disease)
- Self selection (longer survival due to volunteer bias - usually lower risk individuals)
True prevalence vs apparent prevalence
True prevalence = actual level of disease in the population.
Apparent prevalence = level of disease in the population according to the test results
To confirm or rule-in a disease
Need a test with high specificity, to minimize the false positives. (To make sure those to test positive are truly positive)
To confirm the absence of disease or rule out a disease
Need a test with high sensitivity to minimize the false negatives. (To make sure that those who test negative are truly negative).
What happens when the cutpoint of a continuous or ordinal scale test is increased?
The specificity and false negatives increase, the sensitivity and false positive decrease
What happens when the cutpoint of a continuous or ordinal scale test is decreased?
The sensitivity and false positives increase, the specificity and false negatives decrease
Two main types of error
- Random error
2. Systematic error
Three main types of bias
- Information bias
- Selection bias
- Confounding
Causes of information (misclassification) bias
- Recall bias
- Observer bias (interviewer bias)
- Inaccurate diagnostic tests or poor questionnaire
- Non-compliance with treatment in RCT
- Using ecological level data at individual level classification
Direction of non-differential misclassification (information) bias
Always biases towards the null
Direction of differential misclassification (information) bias
Bias may occur in either direction (towards or away from the null)
Examples of selection bias
- Non-response bias
- Admission risk/Berkson’s bias
- Poor choice of comparison group
- Loss to follow up (also Hawthorne effect)
- Selective entry/survival bias
- Detection/Surveillance bias
How to avoid selection bias:
Choice of comparison groups
Good study design - select from the same source, random sampling
How to avoid selection bias:
Non-response bias
Gather information on non-responders, compare non-responders to responders
How to avoid selection bias:
Admission risk/Berkson’s bias
Consider factors that may lead to or exclude individuals from the secondary base
Include only conditions that are not associated with the exposure of interest
How to avoid selection bias:
Loss to follow up
Good study design and implementation.
Collect good contact information for study subjects.
Follow all groups completely/with equal rigor.
How to avoid selection bias:
Selective entry/survival bias
Select from animals/people that were ever in the group, not just those that are currently in the group.
How to avoid selection bias:
Detection/Surveillance bias
Use the same diagnostic technique for all participants.
To qualify as a confounder, a variable must:
- Be associated with the outcome
- Be associated with exposure
- Not be a consequence of exposure (not an intervening or intermediate variable)
Three types of epidemiologic studies
Observational (descriptive, analytical), controlled experiments (RCT), theoretical studies (math models, simulations)
Inputs needed to calculate sample size in descriptive studies
- Continuous outcome (means) = a priori estimate of population variance, desired precision of the estimate and confidence level
- Dichotomous outcome (proportions) = expected prevalence, desired precision of the estimate and confidence level
Inputs needed to calculate sample size in analytic studies
- Continuous outcome (means) = a priori estimate of the means in two groups (effect size), expected variance in the data, desired confidence level and power
- Dichotomous outcome (proportions) = a priori estimates of the proportions in two groups, desired confidence level and power
Inputs needed to adjust for clustering in descriptive studies
Initial sample size, average size of each group, intraclass correlation coefficient
Inputs needed to adjust for a finite population in descriptive studies
Initial sample size, total population size
always round up
Inputs needed to calculate sample size to detect disease or determine freedom from infection (finite population)
Minimum expected prevalence, desired confidence level, size of the population (assumes a perfect test)
Inputs needed to calculate sample size to detect disease or determine freedom from disease (infinite population)
Minimum expected prevalence, desired confidence level (assumes a perfect test, calculates approximate sample size)