Diagnostic test performance Flashcards
Difference between test validity and reliability
Validity= ability to distinguish between disease and non-disease (Sn, Sp, PPV, NPV)
Reliability = how replicable a test is across users, populations, etc…
Definition and calculation of sensitivity
Proportion with disease that are correctly identified by the test
TP/(TP+FN)
Definition and calculation of specificity
Population without disease correctly identified by test
TN/(TN+FP)
Which test validity measures are stable regardless of disease prevalence (i.e. across populations)
Sn and Sp, LR+ and LR-
How would you use Sn and Sp in clinical practice
High Sn: SnNOut -ve results rule OUT
High Sp: SpPIn + ve results rule IN
Definition and calculation of PPV
Proportion with positive test result who actually have disease
TP/(TP+FP)
Definition and calculation of NPV
Proportion with negative test result who don’t have disease
TN/(TN+FN)
How do NPV and PPV vary with pre-test probability (i.e. disease prevalence)
Higher prevalence = higher PPV, lower NPV and v.v.
Calculation and definition of LR+
Increase in odds (i.e. ratio) of disease in positive result
Sn/(1-Sp)
Calculation and definition of LR-
Decrease in odds of having disease with negative test result
(1-Sn)/Sp
Interpretation of LRs
>5 LR+ - moderate increase
<0.2 - moderate decrease
NB: may not be symmetrical!
Converting LRs into PPV/NPV
Fagan’s nomogram
Line through pre-test prob (disease prevalence) -> LR -> post-test prob (PPV or 1-NPV)
Function of ROCs
Visualise trade-off between Sn and Sp at different cutoffs
Allow determination of optimum cutoff (closest to 0,100 point)
Allow comparison of different tests (higher AUC = better test)
Axes on ROCs
1-Sp (i.e. FP rate) on x
Sn (TP) on y
AUC value of non-discriminatory test
0.5
