Are positive predictive value & post test probability (PTP) the same?
PPV = proportion of positive tests that have the test condition =
sens x prev/sens x prev + (1-spec) x (1 - prev)
sens = sensitivity
spec = specificity
prev = prevalence
post test probability (PTP) = pre-test probability X LR
LR+ = likelihood ratio = odds that the test will be +ve in a patient with
the test condition
LR+ = sens / (1 - spec)
I'm playing with methods of validating electronic patient records (EPRs).
Others have used sens/PPV as measures of completeness & accuracy of
recording. I decided to use a Bayesian approach & measure pre-test
probability as = prevalence of the test condition & calculate LR+ from a 2x2
table where the Read Code functions as a "test" for the actual (true/valid)
presence or absence of the test condition
Thus
Condition present condition absent
Read code present true + false +
Read code absent false - true -
Thus for our diabetic data the results are;
true + 282 false + 2
false - 5 true - 13302
prev = 2.1%
sens = 98.3%
spec = 100%
LR+ = 6537
LR- = 0.017
PPV = 99.3%
PTP = 99.3%
I'm probably just being dense, but calculating PPV & PTP gives the same
result. I just hadn't realised they were the same (unless I've screwed up my
calculations....)
Comments welcome ;-)
===
Dr Alan Hassey (mailto:[log in to unmask])
RCGP Health Informatics Grp & JCG
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|