ReMed 2018 Remed 5 - Histoire de la Médecine | Page 13
Great, now that you have the results, do you
think you can interpret and use them in your practice?
Let’s start with inspection, Bulging flanks and Edema
seem to have very high sensitivities, which is a really
good starting point. But how do you deal with Bulging
flanks specificity? The variation among studies is very
considerable, going from 44% to 70%. It is even worse
for Flank dullness, going from 29% (which is very poor)
to 69% (which is quite correct). So the first difficulty is
variation in data results and the inability of making
sense of the numbers. A simple way of correcting the
problem is taking the average of the results. Although
not very precise, it may be helpful.
However, the real problem is when trying to com-
bine data. What do you say about a patient presenting to
the ER with bulging flanks, shifting dullness and fluid
wave but none of the other features? Notice that some
signs with very high sensitivity are absent. On the other
hand, some signs with very high specificity are present...
It is very confusing and unless there is an effective meth-
od to combine sensitivities and specificities, we are inca-
pable of making any decision.
2- More simple and more practical : The likelihood Ratio
A very simple, effective and, most important, practical
index to use with data is the Likelihood Ratio (LR). It
summarizes the information contained in sensitivity
and specificity to tell us how likely a given test result
is in people who have the disease compared to how
likely it is in people who do not have the disease. This
method of describing the accuracy of diagnostic infor-
mation, once mastered, is much faster and more pow-
erful than the sensitivity and specificity approach.
The first concepts that one must understand
are the Pre- and Post-test probabilities. Pre-test prob-
ability is the probability of a person to have a disease
before applying any test on him. In most cases, it cor-
responds to the prevalence of the disease in the coun-
try. For example, if the prevalence of a certain disease
is 16%, any random patient has a probability of 16% to
have that disease.
Post-test probability is the probability of a patient to have
the disease after considering the results of a diagnostic
test. The shift in the probability before and after applying
the test expresses the « strength » of the test; it tells us
how powerful our test is in recognizing the disease.
We should also remember that some findings,
when positive, increase greatly the probability, but they
change it very little when negative. In opposite, other
signs are more useful if they are absent, because the
negative finding decreases considerably the probability,
although the positive one changes probability very little.
Here, the likelihood ratio is the most useful tool
to show the strength of a test. Its definition is basical-
ly « the proportion of patients with disease who have a
particular finding divided by the proportion of patients
without disease who also have the same finding. » 4
LR =
Probability of finding in patients with the disease
Probability of finding in patients without the disease
We add the adjectives « positive » and « neg-
ative » to indicate in which case the physical sign is
present or absent. A positive LR, therefore, is the pro-
portion of patients with the disease who have a phys-
ical sign divided by the proportion of patients without
the disease who also have that sign. A negative LR, is
the proportion of patients with the disease who lack
a physical sign divided by the proportion of patients
without the disease who also lack that sign.
The formula is very simple. In a positive LR,
the numerator, i.e « the proportion of patients with
the disease who have the physical sign » is basically
the sign’s sensitivity. And the denominator, i.e « the
proportion of patients without the disease who have
the sign » is the 1- specificity.
Thus: Positive LR = Sensitivity/(1- Specificity).
It seems a bit confusing but try to think of it slowly
and you will see that it is really simple.
Same thing for the Negative LR: the nominator, i.e
« the proportion of patients with the disease lack-
ing the finding », is 1 - sensitivity. And the denom-
inator, i.e « the proportion of patients without the
disease lacking the finding », is the specificity.
Therefore: Negative LR = (1- Sensitivity)/Specificity.
Now the most most important part: what do
these information tell us ? When the LR of a test is above
1.0, it means that the finding is more likely among pa-
tients with the disease than those who l ack the disease.
Thus, a LR > 1 means that the probability of the disease
increases. By applying the same reasoning, we deduce
that when the LR is below 1, the probability of the dis-
ease decreases. Finally, when the LR is 1, or very close
to it, it means that the probability of the disease is un-
changed (because the finding is equally likely in patients
with and without the disorder).
Now let’s apply these findings to our example.
Remember that the Sensitivity of the Edema was 87%,
and the Specificity 77%. Thus the positiveLR = 0.87/
(1- 0.77) = 3.8. The negativeLR = (1- 0.87)/0.77 = 0.2.
We calculate the remaining LRs and summarize the
results in the table:
LR+
LR-
Inspection
Bulging flanks
Edema
1.9 0.4
3.8 0.2
Palpation et percussion
Flank dullness 1.9 0.3
Shifting dulness 2.3 0.4
Fluid wave 5.0 0.5
ReMed Magazine - Numéro 5
13