Medical Test Accuracv and Statistics
1B Mathematics SL
Research Report - Specimen Paper A
Given the stimulus word, Health, started thinking of the ways one can use probability
(my favorite branch of Mathematics) in looking at medical topics. I decided to find out what
it means for a medical test to be "97% accurate" or "99% accurate". We all know that no
medical test is perfect. How much error is acceptable, and what do these errors mean? What
happens if a patient is diagnosed incorrectly?
To begin, we need to understand what the word "accurate" really means in terms
of medical tests. The accuracy of a medical test refers to two things - sensitivity and
specificity. Sensitivity is the ability of a test to correctly diagnose a condition in a patient
who has that condition. Specificity is the ability of a test to correctly determine that a
person does not have a condition. For example, if a very high percentage of people who
have a disease test positive for that disease, the test has high sensitivity. If people who do
not have a disease test negative, the test has high specificity. Therefore, a medical test
which provides a correct result 99% of the time can be said to be "99% accurate".
Let's look at what medical test accuracy means to a general (fictional)
population. Suppose that a new and terrible brain disease, Mathitis, is known to
currently affect approximately 5% of all students aged 16-18. A doctor has
developed a medical test to diagnose this affliction. The test is "99% accurate". A
large school district decides to test all of the 20000 students within this age group. The table
below shows how the results would come out. The results in green represent a correct
diagnosis, and the results in red are incorrect.
In this particular school district, 19000 students do not have the disease, though only
18820 will test NEGATIVE. One way to look at this data is that 18810 out of 18820, or
99.95%, of the NEGATIVE test results are correct. These are called true negatives. 99.95%
correct sounds pretty good, right? The table also shows, however, that although 1000
students (out of 20000) actually have the disease, 1180 will test positive. In other words,
about 16.1% of the POSITIVE results are incorrect. These results are called false positives,
and 16.1% incorrect doesn't sound very accurate!
We can also look at this (admittedly fictional) situation using conditional probability.
The formula for conditional probability is P(AEB) = In words, we would say, "the
probability of event A given that we know event B is equal to the probability of the
intersection of events A and B, divided by the probability of B." In our example, let T be the
event that a student tests positive for the disease, and let M be the event that a student has
the dreaded disease Mathitis. What if we wanted to find the probability that a student who
tests positive for the disease actually has the disease? We would use our formula for
conditional probability:
P(MIT) - P(Mnr) _ _ 990 0.839= 83.9% PCT)
( 1180 1180
k20000J
Page |2
In words, this means that about 83.9% of the students who test positive for Mathitis
actually have the disease. This is easily seen using the values in the table.
What if you are among the students who test positive for this disease? Do you kiss
your family and friends, not to mention your University acceptance, goodbye? Of course you
don't. In most cases (real, not fictional), if a person tests positive for a serious disease, then a
second, even more accurate, test is given. This raises the question, if a more accurate test
exists, why didn't the doctor just administer that test in the first place? In many cases, this
second test can be much more expensive and invasive than the initial test. It is
just not economically feasible, nor physically comfortable, to give a more
invasive test to the general population.
Now let's take a look at a real-life situation, the HIV/AIDS virus. It is
estimated that 4,1.._ about 33,000,000 people in the world are living with this virus. The
world's population is currently about 6,770,000,000, which means that about 0.487% of the
people on Earth have the HIV/AIDS virus. Let's say that an initial medical test for the virus
is 99.7% accurate, which sounds very good. In fact, "A large study of HIV testing in 752
U.S. laboratories reported a sensitivity of 99.7% and specificity of 98.5% for enzyme
immunoassay [a type of detection test], and studies in U.S. blood donors reported
specificities of 99.8% and greater than 99.99%"1 so it seems reasonable to assume that an
initial test might be 99.7% accurate. Here is what the results would look like if the test were
given to every person on Earth:
INITIAL test reads test reads Totals
HIV/AIDS TEST POSITIVE for NEGATIVE for
RESULTS HIV/AIDS
HIV/AIDS
person really has
the virus 99,000
person does not
have the virus
Totals 53, 112,000
Once again, we see that the NEGATIVE test results look pretty good. Only 0.00147%
of these are incorrect, while 99.9985% of the NEGATIVE results are correct. The
POSITIVE test results tell a different story, however. About 38.1% of the POSITIVE results
Page |3
are incorrect, or false positives. Only 61.9% of those who receive a POSITIVE diagnosis in
the original medical test actually have the HIV/AIDS virus.
Does all this mean that we shouldn't take a POSITIVE result seriously? Is this so-
called
99.7% "accurate" test flawed? The answer to both of these questions is NO. However, it is
interesting that the results from this very serious real-life disease are somewhat similar to
our fictional "Mathitis" situation. In both cases, the NEGATIVE test results seem pretty
reliable, but the POSITIVE results are not so reliable.
Why do we seem to have such a high percentage of false positives? For one thing, not
every person on Earth would generally be given an AIDS test, or a test for any other disease,
for that matter. The people who are tested for most diseases are those who are showing
symptoms or who are more.at risk for that disease than the general population. In this case,
we are looking at a relatively rare disease, yet the test is given to a huge number of people.
While anyone can get the HIV/AIDS virus, not every person on Earth is at equal risk, due to
factors such as general health, age, lifestyle, geographic location, etc. Consider a disease like
the recent "Swine Flu", or HINI virus. Not everyone needs to be tested for this flu virus.
Those given the test are people who are already showing symptoms of the illness, such as a
prolonged fever, or those who have been exposed to a person known to be infected.
In addition, the percentages would be different if we were to look at specific
populations. Sadly, about two-thirds of the HIV/AIDS cases in the world are located in Sub-
Sarahan Africa, where in some countries over 20% of the population is infected with the
virus. We would expect many more true positive results to an HIV/AIDS test given in this
part of the world than we would in areas like eastern Asia and parts of central Europe, where
fewer than
0.1% of the populations are infected. If we were to conduct our conditional probability test
on these populations, we would see quite different percentages than we would when
looking at the world-wide population.
Page |4
Of course, we have only been looking at our data in one way, using conditional
probability. We all know that statistics can be used to tell the truth, or to distort the truth.
One of my favorite quotes is, "Statistics are like bikinis. What they reveal is suggestive, but
what they conceal is vital" NAaron Levenstein. 2 Let's try looking at our data in two different
ways, and see what it reveals.
If we look at the number of false positives from our HIV/AIDS test, we have
20,211,000. This seems relatively large when compared to the total number of positive test
results. Let T be the event that a person tests positive for the virus, and let V be the event that
a person actually has the HIV/AIDS virus. So once again using conditional probability,
r 20211000
k6770000000
J 20211000
P(V'ET) o, 381 = 38.1%
PCT) ( 53112000 h - 53112000 k6770000000]
In words, this means that If a person tests POSITIVE for the virus in our initial test, there is a
38.1% probability that he does actually have the virus.
Instead, now let's compare the number of false positives, 20,211,000, to the totat
number of people given the test, 6,770,000,000 (i.e. the entire world population).
So we could say that the probability of any random person receiving a false positive result is
20211000 equal to , which is approximately 0.298%. This
supports the supposed 99.7%
6770000000 accuracy of the test. So although the conditional probability seems very
high, there are actually a very small percentage of these false positives in the entire
population.
Now let's look at what would happen when the people who receive a POSITIVE
result are given a second, more expensive, but also more accurate, test. Let's suppose that
this second test has an accuracy level of 99.9%.
Page 5
SECONDARY test reads test reads Totals
HIV/AIDS POSITIVE for NEGATIVE for
HIV/AIDS
TEST RESULTS HIV/AIDS
person really has 32,901
the virus
person does not
20,211
have the virus
Totals
We can look at these results in a few different ways, and consider the implications of
these results. First, let's consider the number 32,868,099, which represents the number of
true positives. Remember, these people have now tested positive for the virus twice. So we
can say that 32,868,099 out of the estimated 33,000,000 people who actually have the virus,
or about
99.6%, have received a correct diagnosis twice. This is a huge percentage of people who
can now begin receiving treatment and can hopefully live longer, fuller lives. I think most
of us would agree that this is the whole purpose of medical testing, and that being able to
properly treat 99.6% of the patients with this or any other disease is a good thing.
What about the patients who have been diagnosed incorrectly in either test? We have
99,000 people who received a false negative result from the first test and were not given the
second test. We have 32,901 people who received a false negative result on the second test.
We also have 20,211 people who received false positive results twice. This means that we
have a total of 152,112 people who have been mis-diagnosed. 131,901 people have the virus
and may not receive the treatment they need. Even worse, they might unknowingly pass the
virus on to others. And the 20,211 people who do no have the virus, yet have tested positive
twice?
They may be receiving unnecessary treatment and living in fear.
These are just the whole numbers, let's take a look at the statistics. We have 152,112
people who have been mis-diagnosed n at least one of the tests. This represents about
0.00225% of the world population. Is this acceptable? Is it OK for 20211 people (or
0.00000298% of the world population) to live with the mistaken notion that they have a
deadly virus? In this case, it is not so easy for us to say yes, this is
acceptable. We are lookin at an extremely small percentage of the
population, yet it doesn't seem fair.
But what can be done? Let's consider a few things. First, we need to remember that
an HIV/AIDS test, or any other test, would not be given to the entire population. So it is
unlikely that there would even be as many as 0.00000298% "double false positives" in any
test for a serious disease. Second, if a person who received a false negative continued to
display symptoms of a disease, it is assumed that that person would receive some type of
medical attention, and would be tested again. And finally, we can hope that anyone (both
true and false positives) who was receiving any type of treatment would be monitored
closely by a medical professional.
To me, the statistical nature of this topic is quite interesting and fun, from a
mathematical perspective. The personal and ethical implications, however, are
somewha>__. C staggering. In looking at the numerical data, I was constantly asking
myself, "What if I were one of the false positives?" I guess the best thing for me to
remember is that at least I am lucky enough to have access to the medical testing and
treatment need in any type of health situation. Unfortunately, there are too many people in
the world who do not have this luxury.
Page 7
BIBLIOGRAPHY
1. Annals of Internal Medicine; "Screening for HIV: A Review of the Evidence for the U.S.
Preventive Services Task Force"; Roger Chou, MD; Laurie Hoyt Huffman, MS;
Rongwei Fu, PhD; Ariel K. Smits, MD, MPH; and P. Todd Korthuis, MD, MPH;
http://www.annals.org/cgi/content/fulI/143/1/55
2. "Quotations about Statistics"; http://www.quotegarden.com/statistics.html
3. UNAIDS: Joint United Nations Programme on HIV/AIDS; "2008 Report on the Global
AIDS
Epidemic";
http://www.unaids.org/en/KnowledgeCentre/HlVData/GlobalReport/2008/2008 Global
re port.asp
4. CancerQuest - Emorv Universitv; "Sensitivity and Specificity of Medical Tests";
http://www.cancerquest.org[index.cfm?page=3305
5. CancerQuest - Emory University; "Frequently Asked Questions: False Positives and
Negatives"; http://www.cancerquest.org/index.cfm?page=4923
6. Tau Topics; "Accuracy of Diagnostic Laboratory Medical Tests: Specificity and
Sensitivity"; http://www.michaelandiudvstouffer.com/iudy/articles/specsen.html
Page | 8
