Why Pretest and Posttest Probability Matter in the Time of COVID-19

June 8, 2020

Discussions about the fine points of diagnostic testing that are normally confined to medical or public health settings are now taking place far and wide, across news outlets, on social media and within private homes. As the COVID-19 outbreak fluctuates, we are faced with the importance and urgency of mass testing, but the intricacies of such a process often confound healthcare providers and the general public alike. In this article, we will explore the concept of pretest and posttest probability, what it means for SARS-CoV-2 testing and why diagnostic testing is not as simple as a positive or negative result.

What Are Pretest and Posttest Probabilities?

The way we diagnose patients with disease has changed significantly over time. It is easy to take for granted the hundreds of diagnostic tests at our fingertips. While diagnostic tests should typically be ordered to confirm a specific suspected disease, in practice they are often ordered as a long list of tests in order to rule out a host of different illnesses. Many factors contribute to the overuse of diagnostic tests, from patient expectations to test availability and convenience. While many diagnostic tests perform well, none are perfect. To truly understand the result of a test and properly diagnose a patient, we must use pretest probability, which may also be referred to as disease prevalence. Pretest probability is the chance that the patient has the disease, estimated before the test result is known. It is based on the probability of the suspected disease in that person given their symptoms. This is particularly valuable in clinical microbiology, where knowing the patient's history is incredibly useful in making a microbiologic diagnosis. Simply stated: the probability of the patient having the suspected disease should be known prior to testing. Posttest probability is clinically very important, as it tells us a person's chance of having a disease after a test is performed. This measurement can be used to decide whether to accept a diagnosis of disease, rule one out or order more testing.

How Are Pretest and Posttest Probabilities Determined?

Pretest and posttest probabilities take into account test performance (for example, false positive and negative rates), as well as disease and community context.

Test Result Patient Has Disease Patient Does Not Have Disease
Positive True Positive False Positive
Negative False Negative True Negative

The following factors are important to these calculations:

  • Sensitivity is the ability of a test to correctly identify the disease in the population of people who have the disease.
    • Explained: If a test never missed a case of influenza, then it would be 100% sensitive. Every person that has disease caused by influenza will test positive by this test method. A highly sensitive test will be so good at detecting disease that if the test is negative, we can be pretty sure the disease is not present.
    • Sensitivity = True Positive / (True Positive + False Negative)
  • Specificity is the proportion of people who test negative for the disease among those who do not have the disease.
    • Explained: If a test is 100% specific, then every person who does not have the disease of interest will test negative. A highly specific test can help rule in a disease, but since specificity (and sensitivity) are strictly related to test performance and do not take disease prevalence into account, it is not possible to completely rule a disease in or out with these measures alone.
    • Specificity = True Negative / (True Negative + False Positive)
  • Likelihood ratios use sensitivity and specificity to create a ratio of the probability that a test is correct to the probability that it isn't.
    • Explained: Likelihood ratios are calculated to determine 2 things: 1) how useful a diagnostic test is and 2) how likely it is that a patient has a disease. The formula for a positive diagnostic test result is (sensitivity) / (1-specificity) and the formula for a negative diagnostic test result is (1-sensitivity)/(specificity). If the likelihood ratio is above 1, then the probability that the patient with a positive test has the disease of interest increases. If the likelihood ratio is below 1, then there is a decreased probability that the patient has the disease. The same interpretation of results is true for negative diagnostic test likelihood ratios.
  • Prevalence tells us how much disease is present in a community.
    • Explained: The prevalence of a disease will vary from place to place and depends on a variety of factors. While the incidence is the rate of new disease in a population, the prevalence is the number of known cases of the disease in a population at a given time. In the general population, the prevalence is also known as the pretest probability.

Mix these concepts together and you have a mathematical soup known as Bayes' Theorem. The most important thing to know about Bayes' Theorem is that it recognizes context (otherwise known as patient history or background of the disease) as one of the most important factors in clinical decision making. With Bayes' Theorem, the pretest probability is the likelihood of a result when prognostic, demographic and clinical factors are considered prior to testing. The bottom line is this: while it may be unrealistic for clinicians to calculate this in everyday practice, a diagnostic test can best be interpreted when the pretest probability (prevalence) is known, and this is why the use of mass testing (particularly with a new test) can seem a bit muddy before more is known about the disease itself.

Putting It All Together

Fagan's Nomogram usefully integrates pretest probability, likelihood ratios and posttest probability.

 Fagan’s Nomogram showing how the pretest probability and likelihood ratio of a positive diagnostic test lead to a posttest probability of 70% (black), while the pretest probability and likelihood ratio of a negative diagnostic test lead to a posttest probability of 0.50% (red).
Fagan’s Nomogram showing how the pretest probability and likelihood ratio of a positive diagnostic test lead to a posttest probability of 70% (black), while the pretest probability and likelihood ratio of a negative diagnostic test lead to a posttest probability of 0.50% (red).
Source: Andrea Prinzi, modified from Wikimedia

Let's say that the pretest probability or prevalence of a particular disease in a population is 10%. Let's also say that based on what we know about the performance of the available test for this disease, the likelihood ratio for a positive test is 20. If a straight line is drawn from the pretest probability of 10% through the likelihood of ratio result of 20, we are left with a posttest probability of about 70%. This means that the probability of the patient having the disease increases from 10% to 70% with a positive test result.

The same can be done for a negative test result. Let's consider the same population as above, where the pretest probability of a particular disease is 10%. The likelihood ratio of a negative test is .05. If a line is drawn from the pretest probability of 10% through the likelihood ratio of .05, we are left with a posttest probability of 0.5%. This means that after a negative test, a person's probability of having the disease of interest drops from 10% to 0.5%.

Since the performance of diagnostic tests differs, and much of the interpretation of these tests relies on how much of the disease is present within the population, these measurements can help us determine how to manage results and anticipate next steps when it comes to public health.

How the Pretest and Posttest Probabilities Change the Way We Think About Diagnoses

Not only does the pretest probability help a clinician interpret a test result, it helps them decide whether or not to even test a patient in the first place. In fact, testing people when it is not warranted can lead to confusing results and is generally considered irresponsible. A diagnostic test will not help (and will likely only lead to confusion) when the pretest probability of a disease is either very high or very low. For example, the probability of a cisgender man being pregnant is zero, so performing a pregnancy test on one doesn't make any sense. If the test happens to be positive, the test results would be disregarded as a false positive due to the pretest probability of pregnancy being zero. While this is an extreme example, it demonstrates why ordering a diagnostic test is not always appropriate, and how it can lead to confusion and mismanagement of a patient. In a scenario where the pretest probability of a disease is very high, ordering the test might also be useless if the results will not change the clinician's practice. For example, if a newborn's pretest probability of having bacterial meningitis is high, the clinician will treat the patient for that disease regardless of the test result.

Posttest probabilities may also be referred to as positive and negative predictive values. These measures tell us how likely it is that a person has a disease of interest based on test results and prevalence of the disease within the community. If a disease is rare, the probability that a test result reflects the patient's true disease state may still be low, even with a really good test.

Pretest and Posttest Probability and COVID-19

Taking into account all of the factors discussed above, it's clear why testing everyone for SARS-CoV-2 might prove to be confusing, and why it seems so challenging to get clear answers about the performance of available tests. Here are some important things to consider as clinicians, scientists and members of the general public:

  1. The sensitivity and specificity of SARS-CoV-2 tests are still being investigated.

Since this virus is new to us, we do not have a gold-standard test to assess the results of new SARS-CoV-2 tests. Individual institutions report the analytic accuracy of the diagnostics tests they are using, typically by comparing them to reasonably validated tests, but understanding the overall performance of these tests is still a work in progress. It's not that these tests are bad, but numerous factors can affect a patient's test results, leaving us with results like false-negatives we are still trying to wrap our heads around. For example, the promise of serological (antibody) testing is great, but the test should not be used for acute (new) disease and the detection of IgG antibodies depends on when the patient is tested and if they have had an immune response.

  1. We don't know the true prevalence of COVID-19.

If we had a perfect test, and we went out and tested everyone in the population for SARS-CoV-2, then we would know how many people truly have the disease. Current challenges related to sensitivity and specificity make this task nearly impossible at this stage of the pandemic. Some research groups have attempted to determine disease prevalence in their communities, but the studies have been met with criticism due to the large amount of uncertainty surrounding their estimates. The pretest probability is the estimated probability of a person having a disease before a test is even performed, and this is based on a clinician's personal experience, local prevalence and published data; all of which is still being actively collected for COVID-19. Additionally, knowing the true prevalence of COVID-19 disease will eventually help us determine things like posttest probability and how to appropriately manage results from screening assays like antibody tests.

  1. It may not be possible to calculate a true pretest probability for COVID-19 now, but clinical reasoning remains important.

In areas where we know that there are many COVID-19 cases, pretest probability is probably high (New York City, for example). In areas where few cases have been reported and pretest probability is lower (many rural areas, for example), other conditions should be considered. Bias can be introduced when the medical team tends to think about the first thing to come to mind, which currently is COVID-19. When this happens, we run the risk of missing other diseases that may be more likely in an area with few cases of COVID-19. Testing for a disease that has a suspected low pretest probability can prevent a patient from getting the treatment and clinical management they really need. In many areas of the United States, the prevalence of COVID-19 disease may be relatively low, with estimates ranging from 5-15 percent of the population. As disease prevalence decreases, so does the posttest probability that a patient actually has the disease, meaning the chances of getting a false-positive test are higher.

Moving Forward

As we move forward through this pandemic, it will become increasingly important to understand how much of the population has been exposed to, or infected with, SARS-CoV-2. Knowing this will help us better understand the capabilities of our diagnostic platforms, disease mortality and how the virus moves through the population. The Centers for Disease Control (CDC) is conducting serology surveillance studies in the United States population in order to better understand disease prevalence. Once we know this, calculating things like pretest probability for SARS-CoV-2 will be more manageable.

How does this relate to a real-life scenario? As cases of COVID-19 are decreasing in many areas, you may be asking, “Should I get tested for SARS-CoV-2 before I take a vacation to visit a family member?” Until we know more about the prevalence of the disease in the population (using antibody testing for example), and the performance (sensitivity, specificity) of the tests that are available, questions like these will be difficult to answer with certainty. This is why public health officials continue to encourage preventative measures like social distancing, hand hygiene and wearing masks.

Most importantly, this pandemic is teaching us many lessons about the importance of appropriate diagnostic use and interpretation of clinical tests. It has gotten easy to take testing for granted and to overuse the technologies we have at our fingertips. The COVID-19 pandemic should give us pause and encourage us to evaluate the way we use all of our tests. For the general public, it is important to understand that infectious disease diagnostics are complex, and that even tests that are easy to perform come with caveats and require interpretative skills. As important as pretest probability is for SARS-CoV-2 testing, it is equally important for other types of diagnostic testing used in the microbiology laboratory every day. Moving forward, we should continue to remember that diagnostic test results require clinical context, they should be interpreted with caution and they should primarily be used when the agent being tested for is suspected.

The American Society for Microbiology has developed step-by-step procedures to help labs develop efficient and effective verification protocols for COVID-19 serologic assays.

Author: Andrea Prinzi, Ph.D., MPH, SM(ASCP)

Andrea Prinzi, Ph.D., MPH, SM(ASCP)
Andrea Prinzi, Ph.D., MPH, SM(ASCP) is a field medical director of U.S. medical affairs and works to bridge the gap between clinical diagnostics and clinical practice.