Disease incidence and prediction. Flashcards
Tests that define an exposure or a disease status are subjected to what sort of error?
Systematic and random error.
Is random or systematic error important in regards to bias?
Systematic.
What two things can result in information bias?
- Imperfect definitions of study variables.
2. Flawed data collection procedures.
What can information bias result in?
Misclassification of exposure or disease.
What two types of misclassification bias exist?
- Non differential.
2. Differential.
What sort of bias exhibits bias towards the null (in terms of the Odds Ratio)?
Non differential.
Misclassification bias is similar but different to information bias. True or false?
False, misclassification bias is a type of information bias.
Is the OR higher, lower, the same, or both in differential misclassification bias?
Both, it can be higher or lower.
Define non differential misclassification bias.
The same amount of bias towards exposed/ non exposed state is the same in both the cases and control group.
Define differential misclassification bias.
The same amount of bias towards exposed/ non exposed state is different in the cases and control group.
What defines the ‘True status’ of a disease when testing a new diagnostic test?
The current gold standard test.
Define sensitivity.
Proportion of individuals with a disease which test positive.
Define specificity.
Proportion of those without a disease who tested negative.
What is the equation for sensitivity?
TP/(TP+FN)
What is the equation for specificty?
TN/(TN+FP)
What is the equation for false positivity rate?
FP/(FP+TN) –> FP/ all non diseased
What is the equation for false negative rate?
FN/(FN+TP) -> FN/ all diseased
Define ‘Positive Predictive Value’ (PPV).
Proportion of positive tests that correctly identify diseased individuals.
Define ‘Negative Predictive Value’ (NPV)
Proportion of negative tests that correctly identify non-diseased individuals.
What is the equation for PPV?
TP/(TP+FP).
What is the equation for NPV?
TN/(TN+FN)
Which of the following vary with as disease prevalence varies?
- Specificity.
- Sensitivity.
- PPV.
- NPV.
PPV/NPV.
Despite the fact that PPV and NPV are influenced by prevalence a good test will always show a good PPV. True or false?
False. A good test can result in a poor PPV if the prevalence is low.
Is PPV or NPV more commonly used?
PPV (specificity is used more than sensitivity also).
Disease prevalence does not modify sensitivity or specificity. What does?
The chosen cut-off value of a continuous variable.
What does lowering the cut-off point of a test do? Why does it do this?
Improves sensitivity but lowers specificity.
More false positives.
What does increasing the cut-off point of a test do? Why does it do this?
Lowered sensitivity but improved specificity.
More false negatives.
When is cut point decision in disease status especially important?
When the distribution of characteristic is ‘unimodal’. This is because the ‘grey area’ is so large.
Is a test with a sensitivity of 50%/ specificity of 50% better or worse than a test of sensitivity of 1%/ specificity of 1% better?
Sensitivity of 1%/ specificity of 1%. This is because if you swap the definitions over it becomes 99%/ 99%.
What does ROC stand for and why is it used?
Receiver Operator Characteristics.
Used to analyse the trade of between sensitivity and specificity.
Would a ROC curve closer to the top left or bottom left corner represent a better test?
Top left corner.
What is the AUC for a ROC curve and what does it show?
Area under the curve.
When this is 1 the test is perfect.
What does a diagonal line of 45 degrees on a ROC curve represent?
A test with no diagnostic value (same amount of true positives and false negatives).
What can you correct for if sensitivity and specificity are known?
Misclassification, providing you have the data for correctly classified subjects.
