More than nine months after the World Health Organization (WHO) declared COVID-19 a pandemic, the US continues to struggle with the most severe outbreak in the world. In particular, the US has suffered from a lack of COVID-19 testing capacity (i.e. undertesting), which is a key contributor to rapid community spread and unnecessary infections and mortality (Galeotti et al. 2020, Shear et al. 2020).
Along with undertesting, another vexing issue, overdiagnosis, is gaining increased attention. A New York Times investigation (Mandavilli 2020) estimated that up to 90% of the positive diagnoses in the US are associated with barely any virus, meaning that these patients are unlikely to develop severe conditions or become infectious. Note that ‘overdiagnosis’ does not mean a patient is given the wrong diagnosis; it simply means that patients with minuscule viral loads receive positive COVID-19 diagnoses, leading to potentially unnecessary quarantining and contact tracing efforts that spur additional testing.
PCR tests for COVID-19 diagnosis
To understand why overdiagnosis may happen, we need to understand the inner workings of the polymerase chain reaction (PCR) test, the gold standard of COVID-19 diagnosis. The PCR test starts with a DNA segment of each sample and duplicates the segment cycle by cycle. For example, a total of 20 cycles results in 220, approximately one million, copies of the original segment, whereas a total of 40 cycles results in 240, approximately one trillion, copies. Each positive PCR test is associated with a cycle threshold (CT), which measures the number of cycles required for the fluorescence to reach the threshold.
A COVID-19 testing laboratory needs to specify a CT cut-off, which is the maximum number of PCR cycles for each test. A higher CT cut-off means patient are more likely to receive positive diagnoses, all else equal. The New York Times investigation by Mandavilli (2020) found that in the case of US laboratories, “most tests set the limit at 40, a few at 37,” whereas most experts agree that “a more reasonable cut-off would be 30 to 35.”
In addition to their excessively high CT cut-offs, US laboratories rarely report the viral load associated with each positive PCR test. The lack of viral load reporting represents a missed opportunity because, as a Science article (Service 2020) points out, including the CT value in the diagnosis can help contact tracers “triage their efforts based on CT values” and help physicians “flag patients most at risk for severe disease and death.”
In a recent Covid Economics paper (Dai and Singh 2020a), we develop a theory to explain the simultaneous existence of overdiagnosis and undertesting, motivated by the current US COVID-19 testing practices. Our theory also explains why US policymakers have not embarked with any related effort despite calls from physician leaders for mandating viral load reporting.
A theory of overdiagnosis and undertesting
The model in Dai and Singh (2020a) centres around a COVID testing laboratory’s decisions, which include (1) the optimal CT cut-off, (2) whether the laboratory should disclose viral load (for which the CT value serves as a proxy) along with each positive diagnosis, and (3) the laboratory’s capacity expansion plan in anticipation of future demand. Rather than thinking of the laboratory as either purely profit-maximising or purely patient-centric, our model assumes the laboratory is interested in maximising a weighted sum of patient welfare and its profit from testing operations.
This model yields several key messages. To start with, unless policymakers mandate reporting of viral load information, the laboratory has no incentive to disclose it along with each positive diagnosis. The reason is that disclosing the viral load information essentially means other parties (such as contact tracers) will be able to utilise the information to triage their efforts, so the laboratory will essentially forgo the ability of using its testing results to influence future demand for testing.
Next, in the absence of mandated reporting requirements, the laboratory has an incentive to use a highly sensitive diagnostic criterion, as captured by an excessively high CT cut-off, in its diagnostic testing decisions. The reason is that the laboratory is partially driven by revenue-generating incentives. A high CT cut-off means more patients will receive positive diagnoses. A positive diagnosis will immediately generate further demand for testing, through formal contact tracing efforts, or through voluntary testing based on individuals’ assessment of their contact history with the diagnosed patient.
Of course, one may argue that a high CT cut-off means more patients will be quarantined, so it reduces potential community spread and eventually contributes to a lower demand. However, due to the uncertainty associated with the development of the pandemic, such a reduction in demand is less certain compared to the immediate increase in demand as a result of each positive diagnosis. Under a mild assumption, we show that the laboratory will find it lucrative to inflate its CT cut-off to maximise its expected payoff.
If the laboratory’s inflated diagnostic criterion has any encouraging consequence, it gives the laboratory an incentive to build a higher testing capacity than otherwise. Nevertheless, our analysis shows the increase in the laboratory’s testing capacity will not fully absorb the inflated demand, so in equilibrium, we will see an excessively utilised laboratory, corresponding to the phenomenon of undertesting.
Finally, we evaluate a policymaker's decision as to whether to require the laboratory to disclose the viral load for each positive diagnosis. Our result shows the policymaker may find it rational not to impose such a disclosure requirement for two reasons. First, when the laboratory does not disclose the viral load information, it has a tendency to over-diagnose, which leads to more individuals being quarantined and thus contributes to slower community spread. Second, in the absence of the disclosure requirement, the laboratory has an incentive to build a large testing capacity. To the extent that society is willing to tolerate the costs associated with quarantining a larger-than-necessary number of individuals, who may or may not be contagious, the policymaker will find it socially optimal not to impose the disclosure requirement.
The model we propose in Dai and Singh (2020a) is related to the several papers that examine undertesting (e.g. Dai and Singh 2020b) and overdiagnosis (e.g. Brodersen et al. 2018) separately. To our best knowledge, the model is the first that explains both phenomena, motivated by the unique setting of the ongoing COVID-19 pandemic.
To be certain, all the analyses in our paper are based on the setup that the testing capacity is supplied by the private sector, which is largely consistent with the reality of the US healthcare industry. A policy implication from our paper is that in future pandemics, social planners should explore initiatives to decouple laboratory’s diagnostic testing decisions from their incentives to build testing capacity. Such initiatives may include, for example, the establishment of non-profit testing laboratory networks.
Brodersen, J, L M Schwartz, C Heneghan, J W O’Sullivan, J K Aronson and S Woloshin (2018), “Overdiagnosis: What it is and what it isn’t”, BMJ EBM 23(1):1–3.
Dai, T and S Singh (2020a), “COVID-19 diagnosis and viral load reporting: A theory of overdiagnosis and undertesting”, Covid Economics 58: 1–21.
Dai, T and S Singh (2020b), “Conspicuous by its absence: Diagnostic expert testing under uncertainty”, Marketing Science 39(3):540–563.
Galeotti, A, P Surico and J Steiner (2020), “The value of testing”, VoxEU.org, 23 April.
Mandavilli, A (2020), “Your Coronavirus test is positive. Maybe it shouldn’t be”, New York Times, 29 August.
Service, R F (2020), “A call for diagnostic tests to report viral load”, Science 370(6512) 22.
Shear, M D, A Goodnough, S Kaplan, S Fink, K Thomas and N Weiland (2020), “The lost month: How a failure to test blinded the U.S. to Covid-19”, New York Times, 28 March.