Understanding Your Test Results

[pattayasnowman]

We tend not to think about diagnostic test appropriately, and unfortunately too few doctors do too.

We are conditioned to think that a positive result means you have the disease and negative we don't.

A test that claims to be 99% accurate we assume that the chances are 99% certain we have the disease if we have a positive result.

But this is not necessarily the case

So what does a diagnostic test really mean.

Firstly No form of testing is perfect.

Every test can introduce both false positives (testing positive when you are really negative) and false negatives (testing negative when you are really positive).

So you what you want to know about a test is - if I test positive( +ve) what is the likely hood I have the disease or what is called

Positive predictive value (PPV) that is probability of being +ve when you test +ve

The PPV dependent on three variables

sensitivity ,specificity and prevalence,

Sensitivity = proportion of people testing +ve who have the disease

Specificity = proportion of people testing -ve who don't have the disease

Prevalence is the proportion of the population that has the disease (Say 1 in a 1000 or 0.1%)

The Prevalence is normally determined by some other method that is - other then the test in question.

For example Breast Cancer by Physical examination, Mammograms etc


So Suppose we have a test, which has a sensitivity of (99.9%) 0.999 and a specificity of (99..6%) 0.996

The PPV as a function of the prevalence rate for the subpopulation to which the testee belongs is given below.

prevalence rate 0.1% 0.3% 0.5% 1.0% 2.0% 5.0% 10.0%

PPV 20% 43% 56% 72% 84% 93% 97%

So if take a test for a disease that claims to be 99.9% "accurate' and the prevalence for that disease in the area that you live is 0.1%(1 in 1000) the test will be wrong 80% of the time

For a fuller explanation : http://uhavax.hartford.edu/bugl/treat.htm

You might also want to watch

The Bayes Theorem: What Are the Odds?

Edited September 3, 2015 by pattayasnowman

[SantiSuk]

"So if take a test for a disease that claims to be 99.9% "accurate' and the prevalence for that disease in the area that you live is 0.1%(1 in 1000) the test will be wrong 80% of the time"

Forget statistics: common sense says that sentence is complete and utter bollux

And no, I'm not gonna open up the material to be persuaded otherwise (even though I did a stats A level)

[h90]

The topic is much more complex than that.

If you test for HIV antibodies and you test positive (test repeated once) and to be sure a test on HIV DNA is made twice and twice positive.

Than maybe send your blood to a second lab and repeat it. If the same result you can be sure to have the virus inside you.

But there are other tests which are often false positive or negative because the method doesn't work well.

Other tests depend on interpretation....high blood pressure or cholesterol levels

[ldnguy]

"So if take a test for a disease that claims to be 99.9% "accurate' and the prevalence for that disease in the area that you live is 0.1%(1 in 1000) the test will be wrong 80% of the time"

Forget statistics: common sense says that sentence is complete and utter bollux

And no, I'm not gonna open up the material to be persuaded otherwise (even though I did a stats A level)

The OP is correct. You are wrongly assuming that common sense means that you understand statistics, which you obviously don't understand.Even most doctors don't understand how these tests work.

[Sheryl]

Important to distinguish between the positive predictive value, which is the percent of all people in a given population who tested positive who actually were positive, and the likelihood of any one person's test result being correct. The former is of importance in setting public health recommendations (for example, about mass screening) while the latter is what is of concern to an individual patient.

For the latter, need to consider not only what the specificity of the original test is but also whether confirmatory testing is done before giving the results and, if so, it's specificity.

Sensitivity is not "proportion of people testing +ve who have the disease", it is the proportion of people with the disease who will test positive. A very big difference. Put another way, it is the risk per test of a false negative result, while specificity is the risk of a false positive.

"So if take a test for a disease that claims to be 99.9% "accurate' and the prevalence for that disease in the area that you live is 0.1%(1 in 1000) the test will be wrong 80% of the time" is not correct.

What I think OP meant was that in a test administered to a population with an extremely low prevalence, even with a high level of sensitivity, most of the positive results would be false positives i.e. the PPV would be low. (20% in the hypothetical example of Sensitivity 99.9%, specificity 99.6% and prevalence only 0.1%).

This is true. However these false positives would be in a very small group of the individuals tested; 99.5% of all the people tested under that scenario will have gotten a negative result, and out of those negatives, 99.999% would be correct, i.e. the overwhelming majority of people will have received a correct test result. (99.6% to be exact. With most of the 0.4% with a wrong result having gotten a false positive).

Screening is just that, screening. It is not definitive diagnosis. It identifies a small subset of people in need of for further investigation to rule out the possibility that they have whatever is being screened for.

Mass screening is not usually recommended for groups with extremely low prevalence. This is why many tests are only recommended (and insurance companies or National Health Systems will only pay for them) in people above certain ages or with certain risk factors.

On an individual level though there is no need for a person to concern themselves with the PPV. It is necessary though to know the sensitivity and specificity i.e. the risk that you, personally, might still have the condition even if you test positive (false negative result) or that a positive result might be a false positive.