Сюрприз? Реальная инфицированность коронавирусом почти в 100 раз выше официальной, летальность завышена КАРДИНАЛЬНО

DISCUSSION

We use outpatient ILI surveillance data from around the US to estimate the prevalence of SARS-CoV-2. We found a clear, anomalous surge in ILI outpatients during the COVID-19 epidemic that correlated with the progression of the epidemic in multiple states across the US. The surge of non-influenza ILI outpatients was much larger than the number of confirmed case in each state, providing evidence of large numbers of probable symptomatic COVID-19 cases that remained undetected. This result is also consistent with ILI excess observed in France in late-February/early-March (29). Additionally, this finding predicts that the slowest epidemic doubling time that could explain the ILI surge would be 4 days, and that this rate could only be achieved with unusually fast early transmission or super-spreading events and a clinical rate near 100%. Consistent with this prediction, we found that deaths due to COVID-19 within the US doubled every 3.0 days and note that this empirical growth rate for the US epidemic can account for the ILI surge with a 25% clinical rate assuming a 4 day lag from the onset of infectiousness to presentation as an outpatient with ILI. Together, these results suggest that SARS-CoV-2 spread rapidly throughout the US since its January 15th start date and was likely accompanied by a large undiagnosed population of potential COVID-19 outpatients with presumably milder distribution of clinical symptoms than estimated from prior studies of SARS-CoV-2+ inpatients.

Excess ILI appears to have peaked during the week starting on March 15th, leading the observed ILI dynamics to diverge from the overall epidemic dynamics implied by the growth rate of COVID-19 deaths in the US. If the ILI dynamics were proportional to the epidemic curve then the two could be related via a constant subclinical rate. However, the changing ratio between COVID-19 prevalence estimated by the ILI surge and the epidemic curves parameterized by the growth rate of US deaths suggests additional mechanisms may be behind the ILI slowdown. Mechanisms which can explain the difference between our simulated epidemic curves and the ILI surge include effective social distancing, disproportionate reductions in ILI care-seeking behavior relative to non-ILI care-seeking behavior, or heterogeneity in susceptibility or contact structure not captured in our SEIR model (25).

Our empirical estimate of the size of the ILI surge has several potential limitations. First, the observed ILI surge may represent more than just SARS-CoV-2 infected patients. A second epidemic of a non-seasonal pathogen that presents with ILI could confound our estimates of ILI due to SARS-CoV-2. However, this seems unlikely as additional viral surveillance through the US Centers for Disease Control and Prevention (CDC) suggests that between March 8 to March 28 other monitored respiratory viruses were at low prevalence (30). Nonetheless, were our approach to be used during winter months, additional steps would be needed to account for concomitant non-influenza seasonal pathogens. Additionally, our assumption that changes in health-care seeking behavior are similar between mild ILI and non-ILI condition may be incorrect. Although this assumption was supported by New York City emergency department surveillance data, it is possible that differential health-care seeking would be present in other locations or in the outpatient setting. Last, it is also possible that our use of ILI data has underestimated the prevalence of SARS-CoV-2 within the US. Although early clinical reports focused on cough and fever as the dominant features of COVID-19 (5), other reports have documented digestive symptoms as the complaint affecting up to half of patients with laboratory-confirmed COVID-19 (31), and alternative presentations, including asymptomatic or unnoticeable infections, could result in underestimation of SARS-CoV-2 prevalence.

Additionally, our models have several limitations. First, we assumed that ILI prevalence within states can be scaled to case counts at the state level. This is based on the assumption that the average number of cases seen by sentinel providers in a given week is representative of the average number of patients seen by all providers within that state in a given week. Errors in this assumption would cause proportional errors in our estimated case counts and syndromic case detection rate. Second, our US-wide SEIR models vary by growth rate alone and as such may not capture important heterogeneity in susceptibility or transmission as well as regional variation, intervention-induced changes in transmission, or clustering of infection outbreaks. Our models were used to illustrate that the ILI surge is consistent with an estimated growth rate and start date for the US epidemic and to specify the mutual dependency of growth rate, the lag between the onset of infection and presentation to a doctor, and clinical rates. Finer models with regional demographic and case-severity compartments are needed to translate our range of estimated prevalence, growth rate, and clinical rates into actionable models for public health managers. Last, our method of calculating the infection fatality rate relied on assumptions about the clinical rate and the delay from patients recorded as ILI to death. Our clinical rate required using patterns of care-seeking for typical seasonal causes of ILI as did our delay from ILI to death; consequently, neither should be relied on as a definitive source for COVDI-19 and estimating the clinical rate and delay from ILI to death for COVID-19 specifically will reduce the large uncertainty around our ILI-estimated infection fatality rates.

Despite these potential limitations, the ILI surge identified in syndromic surveillance time-series allowed early estimates of COVID-19 prevalence, estimates that were not possible from confirmed case data due to early logistical delays in SARS-CoV-2 testing in the US. Our prevalence estimates are supported by a serosurvey conducted in New York State. We estimated that over 8.3% of New York State residents were infected by SARS-CoV-2 by March 28; on April 23, 2020, New York State announced that 14% of residents had evidence of past infection by SARS-CoV-2 by March 29 at which time the cumulative PCR-confirmed case counts totaled only 0.3% of New York’s population (32).

Although an ILI surge tightly correlated with COVID-19 case counts across the US and consistent with the New York State serology strongly suggests that SARS-CoV-2 has potentially infected millions in the US, further laboratory confirmation of our hypotheses are still needed to guide public health decisions. Our findings make testable predictions that one would find relatively high seroprevalence in other states that have already seen an ILI surge and that seroprevalence of individuals infected in March across states is proportional to relative sizes of the states’ ILI surges. A study of ILI patients from mid-March who were never diagnosed with COVID-19 could produce a focused test of our predictions about the number and regional prevalence of undetected COVID-19 cases presenting with ILI during that time. If seroprevalence estimates beyond New York State continue to corroborate our prevalence estimates from syndromic surveillance, this would strongly suggest lower case severity rates for COVID-19 than were assumed in late March by comparing PCR-confirmed case counts to deaths. Further corroboration of our estimates of the magnitude of the ILI surge would suggest ILI and other public time-series of outpatient illness allow early and reliable estimates of crucial epidemiological parameters for rapidly unfolding, novel pandemic diseases. As not all novel pandemic diseases are expected to present with influenza-like symptoms, surveillance of other illnesses that commonly present in the outpatient setting could provide a vital tool for rapidly understanding and responding to novel infectious diseases.