1
$\begingroup$

Assume that a diagnostic test for EPM (equine protozoal myoeloencephalitis) has a sensitivity of 89% and a specifity of 92%. The prevalence of the disease is 1%

  1. What is the probability that a randomly selected horse will test positive on the diagnostic test if the horse is healthy?
  2. What is the probability that a randomly selected horse (from the total population) will test positive on the diagnostic test?

I think in 1. I have to use the definition of conditional probability (https://en.wikipedia.org/wiki/Conditional_probability the one that says Kolmogorov definition) to calculate the probability of the intersection (positive and healthy), but I'm a bit stumped on 2.

edit: I think I might have figured it out. I have to use the law of total probability, it seems. Still, I'd appreciate some feedback if possible, since I have nowhere to check if my answer is correct and I'm very new to this stuff.

$\endgroup$
1
$\begingroup$

(second question only):

We want to find out the positive fraction. In other words, we need to find out the percentage of true positives ($TP$) and false positives ($FP$) and add them up.

We know that $Sensitivity =\frac{TP}{Disease (D)}=\frac{TP}{TP+FN}$, and

$Specificity=\frac{TN}{No\,Disease\,(\bar D)}=\frac{TN}{TN+FP}$.

The sensitivity give you the $TP\, ratio$. The question boils down to realizing that the $FP\,ratio$ is just $1\,-\,Specificity$. Here's why: $FP\,ratio = \frac{FP}{\bar D}=\frac{FP}{TN+FP}=1-\frac{TN}{TN+FP}$.

Now we know that we want:

$p\,(+)=p(TP\cup FP) =\,p\,(TP)\,+\,p\,(FP)=TP\,ratio+FP\,ratio$.

There is one slippery point to this question, which is that it is not asking us for the probability of disease being present, given that a test turned out positive. This would have entailed using Bayes theorem and knowing the prevalence of the disease.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.