I've been reading up on Bayes Theorem and thought I'd try to apply it to a hypothetical medical test, but I'm not sure I'm applying it correctly.
I contrived this scenario:
- A medical test has a sensitivity of 60%. In other words, the false
negative rate is 40%. - The test also has a specificity of 90%.
- The prevalence of the particular disease in the population is 5%.
- I have no symptoms of the disease, but I take the test anyway and get a
negative result. - I want to know the probability that I received a false negative.
I was thinking this could be calculated as
$$ P(\text{False Negative Result}) = P(\text{Disease|Negative}) = \frac{P(\text{Negative|Disease})\cdot P(\text{Disease})}{P(\text{Negative})} = \frac{(1-\text{Sensitivity})\cdot P(\text{Disease})}{P(\text{Negative})}$$
So in this example…
$$ P(\text{False Negative Result}) = \frac{(1-0.60)(0.05)}{((1-0.60)(0.05) + (0.90)(1-0.05))} = 0.02 $$
But I'm not sure if I'm missing something?
Best Answer
Consider this probability tree:
p: disease prevalence and other (prior) risk factors
v: test sensitivity
f: test specificity
D: Diseased
H: Healthy
+: Positive test result
-: Negative test result
The negative predictive value (NPV) is $$P(H|-)=\frac{P(H-)}{P(H-)+P(D-)}=\frac{(1-p)f}{(1-p)f+p(1-v)};$$
the true negative rate (specificity) is $$P(-|H)=f;$$
P(true-negative result) is $$P(H-)=(1-p)f;$$
P(false-negative result) is $$P(D-)=p(1-v);$$
the false omission rate (complement of the NPV) is $$P(D|-)=\frac{P(D-)}{P(D-)+P(H-)}=\frac{p(1-v)}{p(1-v)+(1-p)f};$$
the false discovery rate (complement of the PPV) is $$P(H|+)=\dots.$$
“$P(\text{False Negative Result})$” ambiguously could mean either $P(\text{False-Negative Result})$ or $P(\text{Disease|Negative}).$
The former $(2.00\%)$ is the second outcome (of four possible outcomes) in the probability tree, whereas the latter $(2.29\%)$ is conditioned on the knowledge that only outcomes 2 & 4 are possible.
No, each term above is associated with both a test and an event. Some of them are conditional probabilities, whereas the others are not working with a reduced sample space.
Read more here: the accuracy of a medical test.