Estimated effectiveness of symptom and risk screening to prevent the spread of COVID-19

Abstract

Traveller screening is being used to limit further spread of COVID-19 following its recent emergence, and symptom screening has become a ubiquitous tool in the global response. Previously, we developed a mathematical model to understand factors governing the effectiveness of traveller screening to prevent spread of emerging pathogens (Gostic et al., 2015). Here, we estimate the impact of different screening programs given current knowledge of key COVID-19 life history and epidemiological parameters. Even under best-case assumptions, we estimate that screening will miss more than half of infected people. Breaking down the factors leading to screening successes and failures, we find that most cases missed by screening are fundamentally undetectable, because they have not yet developed symptoms and are unaware they were exposed. Our work underscores the need for measures to limit transmission by individuals who become ill after being missed by a screening program. These findings can support evidence-based policy to combat the spread of COVID-19, and prospective planning to mitigate future emerging pathogens.

Keywords: COVID-19; SARS-CoV-2; emerging infectious disease; epidemic containment; epidemic control; epidemiology; global health; human; travel screening; virus.

Conflict of interest statement

KG, AG, RM, AK, JL No competing interests declared

© 2020, Gostic et al.

Figures

Figure 1.. Model of traveller screening process, adapted from Gostic et al. (2015).

Infected travellers fall into one of five categories: (A) Cases aware of exposure risk and with fever or cough are detectable in both symptom screening and questionnaire-based risk screening. (B) Cases aware of exposure risk, but without fever or cough are only detectable using risk screening. (C) Cases with fever or cough, but unaware of exposure to SARS-CoV-2 are only detectable in symptom screening. (D–E) Subclinical cases who are unaware of exposure risk, and individuals that evade screening, are fundamentally undetectable.

Figure 2.. Individual outcome probabilities for travellers who screened at given time since infection.

Columns show three possible mean incubation periods, and rows show best-case, middle-case and worst-case estimates of the fraction of subclinical cases. Here, we assume screening occurs at both arrival and departure; see Figure 2—figure supplement 1 and Figure 2—figure supplement 2 for departure or arrival screening only. The black dashed lines separate detected cases (below) from missed cases (above). Here, we assume flight duration = 24 hr, the probability that an individual is aware of exposure risk is 0.2, the sensitivity of fever scanners is 0.7, and the probability that an individual will truthfully self-report on risk questionnaires is 0.25. Table 1 lists all other input values.

Figure 2—figure supplement 1.. Departure screening only.

Figure 2—figure supplement 2.. Arrival screening only.

Figure 3.. Population-level outcomes of screening programs in a growing epidemic.

(A) Violin plots of the fraction of infected travellers detected, accounting for current uncertainties by running 1000 simulations using parameter sets randomly drawn from the ranges shown in Table 1. Dots and vertical line segments show the median and central 95%, respectively. Text above each violin shows the median and central 95% fraction detected. (B) Mean fraction of travellers with each screening outcome. The black dashed lines separate detected cases (below) from missed cases (above). (C) Fraction of simulations in which screening successfully detects at least n cases before the first infected traveller is missed.

Figure 3—figure supplement 1.. Population-level screening outcomes given that the source epidemic is no longer growing.

(A-C) are as dscribed in Figure 3.

Figure 3—figure supplement 2.. Plausible incubation period distributions underlying the analyses in Figure 3.

Each line shows the probability density function of the gamma distribution with different plausible means and a standard deviation of 2.25. The parameter values were picked based on the best-fit gamma distributions reported in Backer et al. (2020) and Lauer et al. (2020).

Figure 4.. Sensitivity analysis showing partial rank correlation coefficient (PRCC) between each parameter and the fraction (per-simulation) of 30 infected travellers detected.

Outcomes were obtained from 1000 simulations, each using a candidate parameter sets drawn using Latin hypercube sampling. Text shows PRCC estimate, and * indicates statistical significance after Bonferroni correction (threshold = 9e-4 for 54 comparisons).

Figure 4—figure supplement 1.. PRCC analysis comparing cases where 5%, 25% or 50% of cases are subclinical.

(Middle panel is identical to Figure 4, but repeated for ease of comparison).

Figure 4—figure supplement 2.. PRCC analysis assuming the source epidemic is no longer growing.

By construction, R0 has no impact in a flat epidemic. Small PRCC estimates for R0 arise from stochasticity in simulated outcomes, but are never significantly different from zero.