Diagnostic Test Accuracy Flashcards
How do you calculate RR?
Risk in EXPOSED
____________________
Risk in UNexposed
How do you calculate an OR?
Odds exposure
(cases)
___________________
Odds exposure
(controls)
How do you calculate odds?
Case: exposed/unexposed
(Repeat for controls)
SpPIN
Specificity - rules IN as Positive for disease (can be sure TPs - Down’s)
SnNOUT
Sensitivity - rules OUT Negatives (can be sure of TNs - HIV for transfusions)
What are prior tests?
Tests conducted before the index.
MORE prior tests = fewer differential diagnoses, shapes mixture of patients tested as diagnostic test may appear more accurate
What is a reference standard?
The best /most accurate method available of determining whether an individual in the study has the target condition
What are the main features of the results in diagnostic test accuracy studies?
First consider: the size & consequence of test errors
Measures of Accuracy
- Sensitivity
- Specificity
Precision/significance
- CI = uncertainty associated with estimates of accuracy
Sensitivity
Low SeNsitivity = Lots of FN
(high sensitivity = those WITH the disease are correctly identified as such)
Good for rule OUT tests - likely to be TN if fewer FN
Specificity
Low SPecificity = Lots of FP
(high specificity = those WITHOUT the disease are correctly identified as such)
Good for rule IN tests - likely to be TP if fewer FP
What forms of bias are DTA studies subject to?
Selection bias
- Spectrum bias (difficult to diagnose excluded)
Measurement bias
- Review bias (index & ref standard interpretation are not blinded)
- Verification bias (not all index receive ref standard)
What does a high SENSITIVITY mean?
Low proportion of FN
- a negative test is likely to be a TRUE negative
One could be reasonably confident that a patient is healthy, thereby ruling OUT the condition.
What does a high SPECIFICITY mean?
Low proportion of FP
- a positive test is likely to be a TRUE positive
One could be reasonably confident that a patient has the disease, thereby ruling IN the condition.
Why is prevalence important in DTA studies?
Sensitivity & specificity are proportions, not measures of absolute error in test results.
The proportion of test errors may be low (ie - 10%FP = 90%SPf) but in rare conditions (low prevalence) this will result in a large ABSOLUTE number of errors (FP) (as the disease negative population is remarkably larger)