In my day job, especially on Fridays, I get asked about the reliability and accuracy of diagnostic tests. Part of my work involves serology or immunoassay, the interpretation of results and giving advice to referring medical practitioners on how to interpret the results they have in front of them.
Diagnostic tests are never perfect. False positive and false negative results occur. How much of a problem these false results may cause depends on the clinical context in which a test is used. This underlines the importance of the clinical or medical context in laboratory medicine, especially for pathologists whose job is to bridge the gap between the patient and the test tube. One of the most frustrating things for pathologists and medical laboratory scientists is the absence of relevant clinical information on referral forms.
In pathology truth can be defined as the presence or absence of a disease. The aim of the test is to determine the presence or absence of the disease being investigated. Unlike school tests, which are pass or fail, diagnostic tests are positive, negative or equivocal. In the context of serology however, we use the terms reactive, nonreactive and equivocal. I mention the passing and failing because I’ve read online some comments from patients who misunderstand the pathology results they’ve been given and use the terms pass and fail.
|Disease present||Disease absent|
|Test positive (reactive)||True positive||False positive|
|Test negative (nonreactive)||False negative||True negative|
A chemical or agglutination reaction is observed, or not. “Reactive” is intended to remind people that serology is indirect and depends on the host’s immune response, not just analytical uncertainty of measurement. There are many known unknown variables.
A few serological results are reported as a “level” (e.g., anti-HBs or anti-Rubella IgG) which merit an interpretive comment. These are tightly validated enzyme-linked immunosorbent assays (ELISA), run with calibrators. Titrated tests like complement fixation tests (CFT) and agglutination tests give numbers which have limited meaning on their own and rely on seroconversion or 4-fold rise over time to make a diagnosis.
The term “Positive” is often used to denote a patient’s nominal status for a condition such as a chronic infection. For example, “HIV positive”, “CMV positive” (in transplant donor/recipient) and this is usually deduced from several test results, not just a single sample or technique. Likewise, “immune status” is sometimes reported, especially for immunisation preventable diseases. This is a bit elastic in special cases but gives a fair guide to whether immunisation is required.
A test result can be reactive or nonreactive (and in some cases equivocal), as a reactive result may be a nonspecific reaction. To call it ‘positive’ for that marker implies it is a true, and therefore a meaningful, positive. In particular IgM results can be nonspecific or crossreacting (sometimes due to polyclonal activation) and should not automatically be interpreted as indicating acute infection when they are reactive. When there is limited relevant clinical information in the patient referral, the pathologist or medical laboratory scientist can only report the result in a literal sense.
An interpretative comment is possible when there are relevant clinical details or other relevant pathology results.
Sensitivity and specificity
Sensitivity and specificity are characteristics of the test, while predictive values depend of the disease prevalence in the population being tested.
Often sensitivity and specificity of a test are inversely related.
Sensitivity = ability of a test to detect a true positive. Sensitivity = True positive/[True positive + False negative]
Specificity = ability of a test to exclude a true negative. Specificity = True negative/[True negative + False positive]
Predictive values are of importance when a positive result does not automatically mean the presence of disease. Unlike sensitivity and specificity, the predictive value varies with the prevalence of the disease within the population. Even with a highly specific test, if the disease is uncommon among those tested, a large proportion of the positive results will be false positives and the positive predictive value will be low.
Positive predictive value = proportion of positive test that are true positives and represent the presence of disease. PPV = True positive/[True positives + False positives]
Negative predictive value = proportion of negative test that are true negatives and represent the absence of disease. NPV = True negative/[True negatives + False negatives]
A test with 90% sensitivity and specificity and a disease with 10% prevalence
|Patients with disease||Patients without disease||All patients|
PPV = 90/[90 + 90] = 90/180 = 50%, NPV = 810/[810 + 10] = 810/820 = 98.7%, Which means 50% of the positive test results will be false positive results.
A test with 90% sensitivity and specificity and a disease with 1% prevalence
|Patients with disease||Patients without disease||All patients|
PPV = 9/[9 + 99] = 9/108 = 8.3%, NPV = 891/(891 + 1) = 891/892 = 99.9%, So 91 % of positive results will be false positive results
PPV and NPV for test with 90% sensitivity and specificity.
If the test is applied when the proportion of people who truly have the disease is high then the PPV improve.
Conversely, a very sensitive test (even one which is very specific) will have a large number of false positives if the prevalence of disease is low.
Sensitivity and specificity are intrinsic attributes of the test being evaluated (given similar patient and specimen characteristics), and are independent of the prevalence of disease in the population being tested.
Positive and negative predictive values are highly dependent on the population prevalence of the disease.
How can we use this to predict the presence or absence of disease in our patients?
We need to understand how the diagnostic test result influences the pretest probability of disease to the posttest probability of disease.
The degree to which a test result modifies your pre-test probability of disease is expressed by the “likelihood ratio”.
The positive likelihood ratio is the chance of a positive test result in people with the disease, divided by the chance of a positive test result in people without the disease.
The negative likelihood ratio is the chance of a negative test result in people with the disease, divided by the chance of a negative test result in people without the disease.
Experienced clinicians may disagree on the interpretation of a diagnostic test result
This reasoning “makes explicit” the reasons for such disagreement:
- Differing estimates of pretest probability?
- Differing estimates of test performance?
- Differing willingness to tolerate uncertainty?
Selecting the optimal balance of sensitivity and specificity depends on the purpose for which the test is going to be used. A screening test should be highly sensitive and a confirmatory test should be highly specific. In practice a test is either used for sensitivity or specificity.
What is the test for?
Test with high sensitivity are used to RULE OUT those without the disease. Tests with high specificity are used to RULE IN those with the disease.
You can use the post-test probability of one test as the pre-test probability of the next test – Testing in Series. Diagnostic tests performed in series or sequence allows for orderly progression up or down the probability tree until you are happy with the diagnostic decision. The specificity is increased but the sensitivity falls.
Often a battery of tests is requested at the same time – testing in parallel. Sensitivity is increased because a diagnosis is made when there is positive in either test. The result will be a high number of false positives because the specificity is reduced.
Series or Parallel
|A and B (series)||0.72||0.96|
|A or B (parallel)||0.98||0.54|
Diagnostic tests are often studied in populations different from those to whom they are applied. If the study population is very “sick” the sensitivity may be higher than when the test is applied to a more “general” population, particularly when there is diagnostic uncertainty.
Specificity may be higher in a “healthy” population (low probability). When used in patients who are “sicker” (and for whom there is more diagnostic uncertainty) more false positive results are likely – specificity bias.
So what does this mean?
It means no test is perfect. It means the referring medical practitioner and the pathologist need to be aware of contextual factors like disease prevalence and other factors that influence pretest probability. Knowing that no test is perfect and that there are other variables that influence a result including interpretation and disease prevalence, every result should be considered very carefully for what it means for the individual patient.