One of the most commonly used diagnostic tools, particularly during this pandemic, is the reverse transcriptase polymerase chain reaction test (RT-PCR), which uses a person’s respiratory sample to detect viral particles and determine if the person may have been exposed to a virus.
Laboratory professionals across the U.S. and the globe have used RT-PCR to find out if a person has been infected with SARS-CoV-2, the virus that causes COVID-19.
These tests have played a critical role in our nation’s response to the pandemic. But, while they are important, researchers at Johns Hopkins have found that the chance of a false negative result – when a virus is not detected in a person who actually is, or recently has been, infected – is greater than 1 in 5 and, at times, far higher.
The researchers caution that the predictive value of these tests may not always yield accurate results, and timing of the test seems to matter greatly in the accuracy.
In the report on the findings published May 13 in the journal Annals of Internal Medicine, the researchers found that the probability of a false negative result decreases from 100% on Day 1 of being infected to 67% on Day 4.
The false negative rate decreased to 20% on Day 8 (three days after a person begins experiencing symptoms).
They also found that on the day a person started experiencing actual symptoms of illness, the average false negative rate was 38%.
In addition, the false negative rate began to increase again from 21% on Day 9 to 66% on Day 21.
The study, which analyzed seven previously published studies on RT-PCR performance, adds to evidence that caution should be used in the interpretation of negative test results, particularly for individuals likely to have been exposed or who have symptoms consistent with COVID-19.
Results
Probability of a False-Negative Result Among SARS-CoV-2–Positive Patients, by Day Since Exposure
Over the 4 days of infection before the typical time of symptom onset (day 5), the probability of a false-negative result in an infected person decreases from 100% (95% CI, 100% to 100%) on day 1 to 67% (CI, 27% to 94%) on day 4, although there is considerable uncertainty in these numbers. On the day of symptom onset, the median false-negative rate was 38% (CI, 18% to 65%) (Figure 2, top).
This decreased to 20% (CI, 12% to 30%) on day 8 (3 days after symptom onset) then began to increase again, from 21% (CI, 13% to 31%) on day 9 to 66% (CI, 54% to 77%) on day 21.
Posttest Probability of Infection if RT-PCR Result is Negative (1 Minus Negative Predictive Value)
Translating these results into a posttest probability of infection, a negative result on day 3 would reduce our estimate of the relative probability that a case patient was infected by only 3% (CI, 0% to 47%) (for example, from 11.2%, the rate seen in a large study of household contacts, to 10.9%) (Figure 2, bottom). Tests done on the first day of symptom onset are more informative, reducing the inferred probability that a case patient was infected by 60% (CI, 33% to 80%).
Variation in Posttest Probability of Infection if RT-PCR Result is Negative, by Pretest Probability
The posttest probability of infection in a patient with a negative RT-PCR result varies with the pretest probability of infection – that is, how likely infection is on the basis of the magnitude of exposure or clinical presentation.
When we assumed a high pretest probability of infection (4 times the attack rate observed in a large cohort study), the posttest probability of infection was at minimum 14% (CI, 9% to 20%) 8 days after exposure (Figure 3). When we assumed a lower pretest probability of 5.5% (half the observed attack rate), the negative posttest probability of infection was still minimized 8 days after exposure (1.2% [CI, 0.7% to 2.0%]).
Sensitivity Analyses
When we repeated our analysis assuming a specificity of RT-PCR of 90% rather than 100%, results were very similar (Supplement Figure 1). We found a higher probability of infection in the setting of a negative RT-PCR result, with the greatest difference occurring on day 2 (12.4% vs. 11.3% [1.1 percentage point higher]).
When we repeated our analyses varying the incubation period, we found that an earlier onset time of symptoms led to a quicker decrease in false omission rate and a later onset time led to a slower decrease; however, curves were similar overall, and our primary inferences remained the same relative to the date of onset (Supplement Figure 2). When we repeated our analysis of the posttest probability of infection excluding a different study each time, our inferences were unchanged (Supplement Figure 3).
Discussion
Over the 4 days of infection before the typical time of symptom onset (day 5), the probability of a false-negative result in an infected person decreased from 100% on day 1 to 68% on day 4. On the day of symptom onset, the median false-negative rate was 38%. This decreased to 20% on day 8 (3 days after symptom onset) then began to increase again, from 21% on day 9 to 66% on day 21.
The false-negative rate was minimized 8 days after exposure – that is, 3 days after the onset of symptoms on average. As such, this may be the optimal time for testing if the goal is to minimize false-negative results. When the pretest probability of infection is high, the posttest probability remains high even with a negative result.
Furthermore, if testing is done immediately after exposure, the pretest probability is equal to the negative posttest probability, meaning that the test provides no additional information about the likelihood of infection.
Since the outbreak began, concerns have been raised about the poor sensitivity of RT-PCR–based tests (18); 1 study has suggested that this might be as low as 59% (19). We have designed a publicly available model that provides a framework for estimating the performance of these tests by time since exposure and can be updated as additional data become available.
Tests for SARS-CoV-2 based on RT-PCR added little diagnostic value in the days immediately after exposure. This is consistent with a window period between acquisition of infection and detectability by RT-PCR seen in other viral infections, such as HIV and hepatitis C (20, 21).
Our study suggests a window period of 3 to 5 days, and we would not recommend making decisions regarding removing contact precautions or ending quarantine on the basis of results obtained in this period in the absence of symptoms.
Although the false-negative rate is minimized 1 week after exposure, it remains high at 21%. Possible mechanisms for the high false-negative rate include variability in individual amount of viral shedding and sample collection techniques.
One consideration is whether serial testing would offer any benefit in test performance compared with a single test. If we assume independence of the test results, serial testing would almost certainly reduce the false-negative rate; however, without more data on the underlying mechanism for the high false-negative rate, this assumption may not be warranted.
For example, if the rate were due to individual variability in viral shedding, performance would likely not be improved by serial tests. Although we are aware of no large-scale studies, some preliminary reports suggest lack of independence; for example, in 1 case report of a person with infection confirmed on the basis of both radiologic findings and RT-PCR positivity from endotracheal aspirates, RT-PCR results from nasopharyngeal swabs were negative throughout the clinical course (6).
Further studies to better characterize the underlying mechanism for poor diagnostic performance of SARS-CoV-2 RT-PCR are needed to inform testing strategies.
The relationship between a false-negative result and infectiousness is unclear, and patients who test negative on samples from nasopharyngeal swabs may be less likely to transmit the virus regardless of true case status.
We found an increase in the false-negative rate starting 9 days after exposure; however, it is possible that some of the later results were not true false negatives but rather represented clearance of the infection.
Thus, interpretation later in the clinical course depends on the purpose of testing: If the goal is to clear a patient from isolation, these negative results may be correct, although more data are needed given studies showing viral replication in other sites.
However, if the goal of the test is to evaluate whether additional follow-up is needed or whether the patient should be treated as SARS-CoV-2–positive for the purpose of contact tracing, the test may not be providing the desired information and caution should be used in decision making.
Because antibodies appear later in the course of infection, a combination of antibody testing and RT-PCR might be most useful for patients more remote from symptoms or exposure.
Our study has several limitations. There was significant heterogeneity in the design and conduct of the underlying studies from which the data used in our analyses were drawn.
However, when we did a sensitivity analysis excluding each study in turn, we found that no 1 study was especially influential and inferences were largely unchanged. Sample collection techniques varied across studies (oropharyngeal vs. nasopharyngeal swabs), and several studies stated that samples were from the upper respiratory tract without providing further details.
Thus, we could not fully account for differences in sample collection techniques. Most studies tested samples at time of symptom onset rather than time of exposure, leading to high variance in estimates in the first few days after exposure.
Our model is applicable only in the setting of a known, one-time exposure, not in the setting of continuous exposure, such as in health care workers who may be exposed daily to SARS-CoV-2–positive patients.
Finally, most studies defined true-positive cases as those with at least 1 positive RT-PCR result, meaning that patients who never tested positive would not be included; this could lead to underestimation of the true false-negative rate.
Two studies included probable cases based on clinical and epidemiologic characteristics even if the patients had never had a positive RT-PCR result or serology. Because such criteria as fever, respiratory symptoms, and imaging findings are nonspecific, misclassification is likely, wherein some proportion of probable cases are actually true negatives rather than false negatives. We believe that this effect was small because excluding these studies from our analysis did not change our primary inferences.
In summary, care must be taken when interpreting RT-PCR tests for SARS-CoV-2 infection, particularly early in the course of infection and especially when using these results as a basis for removing precautions intended to prevent onward transmission.
If clinical suspicion is high, infection should not be ruled out on the basis of RT-PCR alone, and the clinical and epidemiologic situation should be carefully considered. In many cases, time of exposure is unknown and testing is done on the basis of time of symptom onset.
The false-negative rate is lowest 3 days after onset of symptoms, or approximately 8 days after exposure. Clinicians should consider waiting 1 to 3 days after symptom onset to minimize the probability of a false-negative result. Further studies to characterize test performance and research into higher-sensitivity approaches are critical.
More information: Lauren M. Kucirka et al. Variation in False-Negative Rate of Reverse Transcriptase Polymerase Chain Reaction–Based SARS-CoV-2 Tests by Time Since Exposure, Annals of Internal Medicine (2020). DOI: 10.7326/M20-1495
Code and Data Availability
The data and code used to run this analysis are publicly available at https://github.com/HopkinsIDD/covidRTPCR (17).
