COVIDStrategyCalculator: A software to assess testing- and quarantine strategies for incoming travelers, contact person management and de-isolation

Intra-host viral dynamics

The COVIDStrategyCalculator analytically solves a stochastic transit compartment model (Fig. 1A) of the SARS-CoV-2 infection time course. The underlying transit compartment model consists of different phases that resemble relevant attributes of the infection: I.e. whether the virus is (i) detectable, the individual (ii) has symptoms and (iii) may be infectious. In an associated article⁶, we describe the mathematical details of the model, and exemplify the estimation of default parameters that capture clinically observed temporal changes and variability in test sensitivities, incubation- and infectious periods, as well as times to symptom onset.

• Download figure
• Open in new tab

Figure 1. Intra-host viral dynamics.

A. Stochastic transit compartment model of intra-host viral dynamics. Model details are provided in an associated article⁶. B-D. Model simulation with regards to time-dependent B virus detectability by PCR, C infectiousness and D the probability of experiencing symptoms.

In essence, for user-defined parameters, non-pharmaceutical interventions (‘symptom screening’ and testing) and initial states (exemplified below in paragraph Configuring and simulating an NPI strategy) COVIDStrategyCalculator internally computes the probability over time of the aforementioned attributes (i-iii) based on the stochastic intra-host model (Fig. 1A). An example is given in Figure. 1B-D: In this simulation, infection occurred at time t_O = 0. Figure 1B shows the probability that the virus is detectable by PCR (= temporal test sensitivity); Fig. 1C depicts the probability that the individual is infectious (= able to disseminate infectious particles) and Fig. 1D depicts the probability that the individual is symptomatic. For any time point, e.g. 14 days post-infection (black dashed line in Fig. 1B-D), it is possible to compute the virus detection probability, the probability that an individual is infectious and has symptoms. The detection probability (Fig. 1B) and the probability to have symptoms (Fig. 1C) are relevant for different non-pharmaceutical interventions (NPI), e.g. diagnostic testing and ‘symptom screening’, whereas the probability to be infectious (Fig. 1C) determines whether individuals may transmit the virus, and is thus used to compute the efficacy of NPIs in reducing the transmission potential.

Software outputs

The COVIDStrategyCalculator allows to compute the efficacy of user-defined NPI strategies. Efficacy refers to reducing the transmission potential emanating from an individual (= ‘risk reduction’). Mathematically, the ‘relative risk’ is computed as the residual transmission risk that remains after a user-defined NPI strategy, relative to the transmission risk in a baseline scenario, where no NPIs are imposed (an example is given in the associated article⁶, Fig. 2 therein). where the nominator integrates over the conditional probability of being infectious after release from an NPI (e.g. quarantine) at time t_end, whereas the nominator integrates over the probability of being infectious in the case where individuals had not been isolated, or put into quarantine (baseline risk). Notably, the relative risk metric above is independent of the initial prevalence (see also van der Toorn et al.⁶ for details). Importantly, in the equation above we assume no additional (e.g. behavioral) differences between the two settings.

• Download figure
• Open in new tab

Figure 2. Screenshot of COVIDStrategyCalculator.

The main window consists of four components: Time course trajectories (1), test efficacy reports (2), a result log (3) and user input (4.). Reported results include the simulation settings, the pre- & post-procedure probability of infectiousness, as well as the efficacy estimates, depicted as ‘% relative risk’ and the ‘% risk reduction’.

When simulating a user-defined NPI strategy, the profile of the % relative risk will be depicted together with the time-dependent diagnostic assay sensitivity P(positive test | infected) as shown in Figure 2 (1.). The latter is intended to visually ease the selection of times to perform diagnostic tests together with the table below the graphic Fig. 2 (2.), which assesses the assays’ ability to filter out infectious individuals P(infectious | positive test). We differentiate between the two quantities because PCR assays may allow virus detection for a prolonged time after symptom onset, during which the secreted virus often is not infectious anymore. The Table in Fig. 2 (3.) logs the summary of the conducted simulations and outputs the following numeric values; (i) the probability to be infectious at the time point where the quarantine or isolation ends , (ii) the ‘% relative transmission risk’ as defined above and (iii) the ‘% risk reduction’ (= efficacy of the NPI). For each metric (Fig. 2 (2. & 3.)), three values are reported which indicate the result in the case of a ‘typical’ infection (typical parameters as outlined below), as well as an uncertainty range based on the provided lower and upper extreme parameter values. For the graphics (Fig. 2 (1.)), the uncertainty range (lower and upper extreme dynamics) is visualized by the shaded area. We will further discuss the parameters of the software and the underlying modeling assumptions by describing how to use the COVIDStrategyCalculator.

In Fig. 2 (4.) the user can configure an arbitrary NPI strategy as outlined below.

Software utilization

Figure 3A shows a zoom of the opening window: The different tabs allow to evaluate a strategy (1.), set parameters (2.) or perform a prevalence estimation (3). In field (4.), the user can select between the different modes of the software.

• Download figure
• Open in new tab

Figure 3. User input panels of the COVIDStrategyCalculator.

A. Zoom-in on the NPI configuration window. B. Model parameter input tab. C. Prevalence estimator input tab. Details of the usage and meaning of the different input fields are provided in the text.’

Configuring and simulating an NPI strategy

The COVIDStrategyCalculator has three modes of operation, corresponding to the three common scenarios in policy making: (i) quarantine for contact management, (ii) isolation of infected individuals and (iii) quarantine for incoming travelers. The modes differ in the initial states used for simulation. In the ‘contact management’ and the ‘isolation’ mode, the simulation starts from a point distribution, whereas in the ‘incoming travelers’ mode assumes a mix of ‘infection ages’.

(i) In the contact management mode, strategies aim to reduce the risk emanating from individuals who have been in recent contact with a confirmed case. In this scenario, the time of the putative infection is known and can be entered by the user (Fig. 3A 6.). In the underlying model (Fig. 1A), all states are set to zero, except the very first compartment of the model, which is set to the initial probability of infection provided in (Fig. 3A 5.). It is important to note here that the NPI efficacy estimates (‘relative risk’, ‘risk reduction’) are independent of the initial probability of infection, as they cancel out when computing these relative quantities (eq. (9)). The user selects a duration of quarantine (7.) and whether ‘symptom screening’ is performed (checkbox in 8.). ‘Symptom screening’ would imply that individuals who develop symptoms go into isolation (in line with WHO guidelines ^1,2,5) and do not pose a risk anymore, whereas individuals that not develop symptoms are being released into society at the end of the quarantine and may continue to pose a risk. The percentage of asymptomatic infections who would be missed in ‘symptom screening’ can be provided by the user (outlined below). The expected level of adherence (= which percentage of individuals follows the proposed NPI) can be set in field (9.). The level of adherence w alters the risk calculation as follows: The user can also decide on whether diagnostic tests should be conducted during the quarantine (check-boxes in field 10.) and select whether PCR tests (default) or antigen-based rapid diagnostic testing (RDT) should be performed (field 11.). In case of a positive test, it is assumed that the individuals go into isolation and do not pose a risk anymore. Individuals that are false negative tested are assumed to stay quarantined until the end of the user-defined quarantine duration and are subsequently released into society. Therefore, the time-dependent false omission rate (= 1-sensitivity) of the test(s) is critical to determining the efficacy of testing during NPIs. Parameters related to the diagnostic test can be modified by the user as outlined below. The output of the configured NPI is shown immediately upon pressing ‘Run’.

(ii) In the isolation mode, the user can assess strategies for the duration of isolation after symptom onset. In this scenario, the time of symptom onset is known and can be entered by the user (Fig. 3A 7.). In the underlying model (Fig. 1A), all states are set to zero, except the very first ‘symptom compartment’, which is set to probability of infection provided in (Fig. 3A 5.). The user’s options are similar to the mode described above, with the exception that a ‘symptomatic’ screening is not possible.

(iii) When choosing the incoming travelers mode (field 4 in Fig. 3A), the user is taken to the prevalence estimation subroutine of the software, Figure 3C. The prevalence estimation subroutine can be used on its own, or to generate a population with mixed ‘infection age’ for the analysis of NPI strategies. The details will be outlined below. To use the mixed population, the user presses ‘estimate prevalence’ (Fig. 3C, field 24) to generate an initial probability distribution (depicted in Fig. 3C, field 25), which can be used (Fig. 3C, 26.) for simulation Figure 3A. After setting the initial probability distribution, the user can proceed as described above.

Setting model parameters and hypothesis testing

In an associated article⁶, we detailed the parameter estimation procedure to derive default parameters that capture clinically observed temporal changes and variability in test sensitivities, incubation- and infectious periods, as well as times to symptom onset. However, a user may want to assess the efficacy of an NPI strategy that uses a novel diagnostic test, or is directed against a novel virus variant, which exhibits different viral kinetics. For this reason, COVIDStrategyCalculator allows users to adapt all parameters related to viral dynamics, diagnostics, as well as the proportion of asymptomatic cases. When switching to the parameters tab Figure 3A (2.), the user is taken to the window depicted in Fig. 3B.

The first 4 rows (shaded area in Fig. 3B) allow to alter the time course of infection, i.e. (i) the average incubation time (= time before symptom onset), (ii) the fraction of this incubation time, where the infected individual is not infectious and the virus is not detectable yet, (iii) the average duration of the symptomatic phase, where the individual is infectious and the virus detectable, (iv) as well as the average duration of the post-infectious phase, where the virus is detectable, but not infectious anymore. Changing these parameters will alter the efficacy of NPI strategies (e.g. how long the duration of a quarantine needs to be), the time-dependent sensitivity of the diagnostic assay (i.e. when tests would be most beneficial), as well as the efficacy of symptomatic screening during quarantine.

The next three lines (fields 17-19 in Fig. 3B) allow the user to alter the characteristics of the diagnostic tests performed, i.e. (v) maximum sensitivity of the PCR test (field 17.), as well as (vi) its specificity (field 18.). The (vii) sensitivity of antigen-based rapid diagnostic tests (RDT) is usually measured relative to PCR results ^13-16 and can be set in field 19. of Fig. 3B. The temporal changes in test sensitivity for RDT are assumed to be comparable to those of PCR testing, but can be altered as described below. The RDT specificity is assumed to be equal to the PCR specificity, because, at least in low-prevalence / low pretest probability settings, a positive RDT result is typically confirmed by a PCR test ^17-20. Generally, upon a SARS-CoV-2 diagnosis it is assumed in the COVIDStrategyCalculator that the individuals go into isolation and do not pose a risk anymore. Thus, altering diagnostic parameters (v-vii; fields 17-19) allows to change the efficacy of the tests to identify infected individuals, while the inputs (v-vii) shift the magnitude of the false omission rate. The temporal profile of the false omission rate can be altered by changing viral dynamics parameters (i-iv; fields in the shaded area in Fig. 3B). The resultant temporal test sensitivity can be inspected in Figure 2A (1.). Lastly, (viii) the proportion of asymptomatic cases can be changed in field 20 of Figure 3B. This parameter alters the effectiveness of ‘symptom screening’ in NPI strategies. Altering any of the parameters allows the user to perform hypothesis testing: For example, ‘How much worse would an NPI perform if the test sensitivity was lower?’, ‘… if the virus incubation time was shorter but the infectious phase longer?’ or ‘… if the proportion of asymptomatic cases was larger’, etc.

Prevalence estimation and assessing NPI strategies in populations with mixed ‘infection age’

In COVIDStrategyCalculator, the user has the possibility to perform a prevalence estimation given a user provided incidence history of the past 5 weeks. The user can also use this utility to generate a population with mixed ‘infection age’ when computing NPI strategies in the ‘quarantine for incoming travelers’ mode.

To perform a prevalence estimation the user inputs the incidence history of the past 5 weeks in the region of interest (Fig. 3C, field 22). The incidence reports are typically reported as the number of cases per week and 100,000 inhabitants by national or supra-national health authorities, like the ECDC. In addition to the incidence reports, an estimate of the presumed proportion of cases that are actually reported (Fig. 3C, field 23.) can be provided. This number may vary widely between different regions and depends on national testing strategies. Upon pressing ‘estimate prevalence’, COVIDStrategyCalculator will estimate the prevalence based on the intra-host SARS-CoV-2 dynamics model (Fig. 1A), as detailed in an associated article⁶. The routine will calculate the probability distribution over the states of the model, as well as the ‘total prevalence’ and the prevalence of individuals that are currently infectious, or pre-infectious. Uncertainty ranges are calculated based on the extreme parameter values provided by the user (Fig. 3B, field 13-14.).

The estimated probability distribution can be used for assessing NPI strategies by checking ‘use’, which will take the user back to the NPI configuration window (Fig. 3A). While this feature of the software allows to estimate the prevalence in a setting of interest, it can also be used to set the initial distributions for the assessment of NPI strategies, e.g. by altering ‘incidence’ values to deduce a suitable initial distribution over the model states.

Identifying efficient combined NPI + testing strategies

Another application of the COVIDStrategyCalculator is to identify NPI strategies that use testing to shorten quarantine or isolation periods. Essentially, this requires to find testing and quarantine strategies that are equivalent or non-inferior to established strategies. In order to identify those strategies, the user starts with assessing the efficacy (relative risk, risk reduction) of an established gold standard NPI as outlined in section “Configuring and simulating an NPI strategy” (e.g. 14 days quarantine as suggested by the WHO ¹). Subsequently, the user could configure a combined strategy, where testing shortens the quarantine or isolation period, for example by placing a test at the end of the quarantine period. By repeating the procedure with different quarantine durations the user could find the shortest combined test + quarantine strategy which has a non-inferior efficacy (relative risk, risk reduction) to the gold standard NPI.