COVID-19 disease is associated with three major patterns of clinical course of infection; mild illness with upper respiratory tract presenting symptoms, non-life-threatening pneumonia and severe pneumonia with acute respiratory distress syndrome (ARDS). Given that the manifestation of COVID-19 infection is highly non-specific, diagnostic tests specific to this infection are crucial and urgently needed to confirm suspected cases, screen patients, and conduct virus surveillance. (18)
RT-PCR is accepted by scientists and medical staff as a robust and well documented technique. With RT-PCR being so common in research and medicine, the technology is already in place to test for COVID-19. RT-PCR can detect current infections of disease, allowing medical staff to determine who is currently infected and who is not. But it should be remembered that RT-PCR relies on capturing and detecting the virus and so it is possible to miss patients who have cleared virus and recovered from disease. In addition the distribution of virus across the respiratory tract varies between patients, so even if a person is infected, the virus may only be detectable in sputum or nasopharyngeal swab but not necessarily at both locations at the same time. RT-PCR for COVID-19 can only tell if a person is currently infected with this coronavirus. It can’t provide information on other diseases or symptoms particular. (19)
Many countries in the world did not underestimate the problem since the first reported cases, knowing well that COVID-19 is inevitable and started to deal early, to limit transmission and increase recovery. Social distancing was paired with the basic epidemiology that’s needed. Contact tracing — the practice of identifying and testing every person that an infected person came into contact with after they themselves contracted the virus — has been prioritized. Almost all efforts to develop the infrastructure for quarantining the exposed or isolating the infected persons (20). Epidemiological testing — where the contacts of infected people are identified, tested in turn and isolated as needed — is the only way to fully break the chains of transmission. Without it, the virus will come roaring back as soon as social distancing guidelines are relaxed. The timing of conducting the tests however is crucial to avoid false-negative results. Both positive and negative results must be utilized in conjunction with clinical observations, patient history, and epidemiological information
ROC analysis has become a popular method for evaluating the accuracy of medical diagnostic systems. It is used in clinical epidemiology to quantify how accurately medical diagnostic tests (or systems) can discriminate between two patient states, typically referred to as "diseased" and "no diseased" (5, 7, 20, and 21). The most desirable property of ROC analysis is that the accuracy indices derived from this technique are not distorted by fluctuations caused by the use of arbitrarily chosen decision criteria or cut-offs. In other words, the indices of accuracy are not influenced by the decision criterion (i.e. the tendency of a reader or observer to choose a specific threshold on the separator variable) and/or to consider the prior probability of the "signal" (5). The derived summary measure of accuracy, such as the area under the curve (AUC) determines the inherent ability of the test to discriminate between the diseased and healthy populations (7). Using this as a measure of a diagnostic performance, one can compare individual tests or judge whether the various combination of tests can improve diagnostic accuracy. In addition one can easily obtain the sensitivity at specific FPF by visualizing the curve and optimal cut- off value can be determined using ROC curve analysis (7).
In summary, despite the fantastic feature of ROC analysis in diagnostic test evaluation and the meaningful interpretation of AUC and its asymptotic properties, a proper design with broad spectrum of case and control and avoidance of bias and control for confounding are necessary for a valid and reliable conclusion in the assessment of performance of diagnostic tests. Spectrum and bias should be considered with careful consideration in study design while confounding can be controlled in analysis as well. While the adjustment of confounding is widely used in etiologic studies in epidemiology, a little attention has been focused for the control of confounding in ROC analysis of medical published diagnostic studies. (5)
Regarding performance of tests in the context of population testing, there are drawbacks both from insufficient diagnostic sensitivity (e.g. leading to missing infected individuals) and insufficient diagnostic specificity (e.g. imposing confinement measures on individuals who are not true positives). This needs to be taken into account along with the stage of the pandemic in a particular population. For example, in the control stage it may be particularly important to identify positive cases with a high level of specificity (i.e. distinguishing COVID-19 from other similar but less dangerous diseases) to avoid unnecessary burden on the healthcare system. In contrast, in the deescalation stage, sensitivity (detecting all remaining infected individuals) could be more important than specificity to make sure the disease is indeed contained. It is also important to take account of the features of the population in which the test is intended to be used, for example whether the prevalence of infection is expected to be low or high, or whether there are local virus variants. Scarcity of reference methods and materials poses difficulties for these validation studies, and also for the evaluation of device performance by manufacturers.
