Modeling the effects of routine screening for accidental lab-acquired infections on the risk of potential pandemic pathogen escape from high-biosafety research facilities

Abstract

Introduction

Accidental lab-acquired infections (LAIs) risk releasing potential pandemic pathogens from high-biosafety (BSL-3 or BSL-4) research facilities. We constructed a stochastic network infectious disease model to simulate how following an initial LAI, the probability of an outbreak of a pathogen resembling wild-type SARS-COV-2 would be influenced by test-and-isolate interventions over a 100-day horizon.

Methods

We varied test frequency (0–7 tests/ week), peak sensitivity (50%–100%), and isolation delay (0–3 days), assuming perfect test specificity to minimize model complexity. For each of 192 parameter combinations, we conducted 1,000 simulations and used logistic regression to quantify how each parameter influenced the likelihood of an outbreak of 50 or more infections.

Results

Results indicated that even relatively infrequent routine testing significantly reduced the risk of outbreaks under diverse plausible scenarios, with greater reductions achieved at higher test frequencies. Once-weekly testing reduced outbreak risk by 52% under optimistic assumptions (80% sensitivity, 1-day delay) and by 29% under pessimistic assumptions (50% sensitivity, 2-day delay). Testing two and five times weekly yielded risk reductions of up to 62% and 71%, respectively, under optimistic assumptions, and 43% and 55%, respectively, under pessimistic assumptions. Logistic regression showed each additional weekly test decreased outbreak odds by 20%, each 10-point increase in test sensitivity reduced odds by 10%, and each additional isolation delay day increased odds by 15.5%. Interaction analyses revealed longer isolation delays attenuated the protective effects of greater testing frequency and sensitivity.

Discussion

These results show routine lab-worker screening with prompt isolation could substantially mitigate potential pandemic pathogen escape risks. Future research should address operational issues such as false positives, which could influence the feasibility, acceptability, and cost-effectiveness of real-world implementation.

Introduction

Accidental lab-acquired infections (LAIs) with potential pandemic pathogens (PPPs) risk sparking a pandemic (Merler et al., 2013; Shurtleff et al., 2012; Imperiale and Casadevall, 2020; Lipsitch, 2017). Wet-lab research with PPPs is generally undertaken in Biosafety Level Four (BSL-4) or Biosafety Level Three (BSL-3) facilities, which operate at maximum and near-maximum biosafety levels, respectively (HHS, 2015). Such research can advance basic science and medical countermeasure development (Duprex et al., 2015). However, it also carries major risks of accidental human exposures and infections. From 1975 to 2016, there were more than 50 documented accidental human-caused exposures to high-consequence pathogens (Manheim and Lewis, 2022), including a smallpox outbreak associated with University of Birmingham’s lab in the UK in 1977, and three lab-associated outbreaks of SARS-1 that occurred in Singapore and Taiwan in 2003 and Beijing in 2004 (Young, 2023). Though the cumulative number of deaths from these events is modest, the mere existence of these events is an indicator that a biosafety lapse leading to escape of a PPP is plausible.

While the pandemic LAI risk posed by any particular research facility in a particular year is likely to be low, these risks are much higher when cumulated across time and over many research facilities. As of 2017, 14 laboratories were conducting research on highly-pathogenic avian influenza (HPAI) viruses modified for mammalian airborne transmissibility, notably H5N1 and H7N9, collectively termed by Klotz as the “matHPAI Research Enterprise.” Based on reported accidental pathogen releases from high-biosafety labs (2003–2017), Klotz estimated an annual release probability of 0.003 per lab, resulting in a cumulative 5-year community release risk of 15.8% and pandemic risk of approximately 2.5% (Klotz and Sylvester, 2012). There are reasons to believe the cumulative risk of escape of a PPP is likely higher than Klotz’s estimate, which only focused on the risk coming from one class of pathogen studied in high-biosafety research facilities, and did not account for the increase in the number of high-biosafety research facilities from 2017 to 2024. As of 2023, there were 51 BSL4 labs in operation, three under construction, and 15 planned, located across 27 countries (Lentzos et al., 2023; Lentzos, 2021). Moreover, while the true number of BSL-3 labs worldwide is unknown, it is at least in the thousands, and likely increasing (Richards et al., 2014).

Routine testing of lab workers for LAIs is not currently standard practice at high-biosafety research facilities. Discussions with current BSL-4 researchers revealed several potential explanations (Honko, 2022). First, lab managers may feel that routine testing is unnecessary given the existing safety precautions. Second, lab managers may be concerned that routine testing could be logistically burdensome and financially costly for individual labs. Third, lab managers may be concerned about the potential practical consequences of false positive results. Fourth, biosafety risk assessments performed by regulators, research funders, and lab managers do not always sufficiently focus their risk assessments on overall societal risk, rather than individual or lab-level occupational risk (Lipsitch and Inglesby, 2014). Last, existing biosafety and occupational-health frameworks governing pathogen research emphasize risk-based occupational health programs, preplacement medical evaluation, post-exposure medical surveillance, and prompt evaluation of symptomatic workers, but they do not appear to treat routine asymptomatic screening of high-biosafety laboratory workers as a standard requirement (APHIS, 2023; Arduino and Arndt, 2020; CDC and USDA, 2021).

While routine testing of lab workers for LAIs is not currently standard practice at high-biosafety research facilities, similar activities have been undertaken before, indicating their potential feasibility. During the SARS-COV-2 pandemic, many workplaces were able to implement routine screening protocols of multiple tests per week, especially in healthcare settings and for high-revenue professional sports teams (Seguí et al., 2021). Routine screening for community infections with SARS-COV-2 has also been performed in a high-biosafety research environment, Boston University’s combined BSL-4 and BSL-3 research center, the NEIDL (Chen et al., 2023), even as it also conducted gain-of-function experiments with the SARS-COV-2 virus (Honko, 2022).

A key challenge with containing PPP LAIs is that for some pathogens, such as SARS-COV-2 and some strains of influenza, a substantial proportion of disease transmission is caused by asymptomatic or pre-symptomatic infections (Peak et al., 2020; Fraser et al., 2004), which makes it difficult to contain an emerging outbreak solely with case-based interventions (Merler et al., 2013). Screening of individuals known to be at high risk of infection can help mitigate these challenges by identifying people with asymptomatic, pre-symptomatic, or mild symptomatic infections, which facilitates case-based interventions (Larremore et al., 2021; Holmdahl et al., 2020).

This modeling study investigates the extent to which routine screening of lab workers in high-biosafety research environments for LAIs can reduce the risk of a catastrophic escape of a PPP similar to wild type SARS-COV-2 into the community. In particular, this analysis focuses on how the success of such a surveillance protocol would depend on test frequency, test sensitivity, and the average delay from a positive test result being taken until the effective isolation of the infected individuals.

Methods

Study population

The target study population is the laboratory employees of a BSL-4 research facility and the surrounding community. To approximate this study population, we simulated a 1,000 person contact network for a BSL-4 laboratory with 20 lab workers and 980 people in the surrounding community. A Barabasi-Albert network structure was chosen to allow a scale-free distribution of connections between individuals (Bjornstadt, 2018). The mean number of connections between individuals that were sufficiently close for transmission was set to eight per person, for both groups, but each individual’s number of connections was heterogenous due to the network structure, with a few individuals acting as network hubs.

Intervention strategies and analytic scenarios

Our goal was to explore the impact of a wide range of plausible test-and-isolate intervention strategies encompassing different values of test frequency, peak test sensitivity, and average delay from positive test to isolation. To achieve this, we conducted epidemic simulations encompassing a wide range of these parameter values using the contact network created for the study population. At the beginning of each simulation, we assumed a single randomly chosen lab worker developed a LAI. We then simulated disease progression, transmission, and other epidemiological outcomes for a subsequent 100 days, by which time the large majority of simulated epidemics reached the threshold for defining an outbreak or had otherwise ended. Across simulations, the frequency of routine lab worker testing was systematically varied from zero tests per week up to seven tests per week in increments of one. Similarly, the value of peak test sensitivity was systematically varied from 50% to 100% in increments of 10%. Last, the average delay from positive test to isolation was systematically varied from 0 days to 4 days in increments of 1 day. This produced 192 unique analytic scenarios.

Data

Parameters characterizing the natural history of infection with wild-type SARS-CoV-2 were drawn from the clinical literature. Full definitions and values for the parameters, difference equations, and state variables are given in Supplementary Tables S1–S3 in the Supplementary Material, respectively.

Simulation model

We conducted this analysis with a discrete time stochastic network infectious disease model, with a 1-day time-step. A schematic of the modelled health states and possible state transitions is given in Figure 1.

FIGURE 1.

Model structure. *Full definitions of each variable are available in the Supplementary Material.

Transmission mechanisms

In this model, we assume that an infection is transmitted across a given susceptible-infectious edge with probability τ per time step t. In the simplest version of such a chained binomial model, the probability of any particular susceptible becoming infected in a given time-step is $p=1-{\left(1-\tau \right)}^y$ , where y is the number of infected contacts (Bjornstadt, 2018). The spread of epidemics on networks is influenced by both the mean number of contacts (edges) k, as well as the heterogeneity in that number, which is quantified by k’s relationship with $k^2$ , as evidenced by the equation for a simple chained binomial model, $R_0=\left(\frac{\tau }{\tau +\gamma }\right)*\frac{k^2-k}{k}$ (Bjornstadt, 2018). To enhance the model, the equation was modified to account for differential transmission by individuals with pre-symptomatic and asymptomatic infection relative to symptomatic disease, as well as the greatly reduced transmission of isolated infected contacts. This was done in several ways. First, the model includes a transmission offset $\mathit{\zeta }$ that is multiplied by the number of asymptomatic infected contacts to account for the diminished infectiousness of individuals with asymptomatic infection relative to individuals with symptomatic infection. Second, the model features a transmission offset $\mathit{\psi }$ that is multiplied by the number of pre-symptomatic infected contacts to account for the diminished infectiousness of pre-symptomatic infectees relative to individuals with symptomatic infection. Third, the model includes a transmission offset $\mathit{\iota }$ multiplied by the number of isolated infected contacts which represents the (small) fraction of transmission that isolated infectious individuals exhibit even while isolated, relative to non-isolated infectious individuals. The modified transmission equation is given in Supplementary Tables S1 in the Supplementary Material.

Screening and isolation

For every time step, lab workers were assumed to be tested with probability ${\theta }_L$ . In cases when the tested individual was infectious and symptomatic, the person would be isolated with probability $\sigma$ , the peak test sensitivity. If the tested individual was infectious and pre-symptomatic, the person would be isolated with probability $\sigma *\nu$ , with $\nu$ representing the reduction in test sensitivity during the pre-symptomatic period. If the tested individual was infected but not yet infectious or even symptomatic, the person would be isolated with probability $\sigma *\mu$ , with $\mu$ representing the heavily reduced test sensitivity during this period. We additionally modeled RT-PCR clinical sensitivity as modestly lower in persistently asymptomatic infection by applying a multiplicative offset of 0.8 to the symptomatic-phase peak sensitivity (σ). This choice is consistent with pre-Delta longitudinal nasal-swab viral trajectory data suggesting proportionately lower peak viral RNA (higher Ct) in asymptomatic versus symptomatic infections (Stankiewicz Karita et al., 2022). Parameterization and implementation are provided in the (Supplementary Tables S1, S2).

In addition to test-induced isolation, the model incorporated baseline rates by which symptomatic lab workers and symptomatic general population members would isolate themselves even without testing positive, ${\mathit{\pi }}_{\mathit{L}}$ and ${\mathit{\pi }}_{\mathit{G}}$ respectively. The model was programmed so that in all simulations, after the first positive test result was recorded, the testing cadence for lab workers increased to daily. Similarly, in all simulations after more than ten positive test results were recorded infected, symptomatic members of the general population became subject to a testing cadence of one test per week. To minimize model complexity, tests were assumed to be 100% specific, with no false positives. This simplifying assumption was intended to isolate the effect of screening on outbreak risk conditional on an accidental infection rather than to model the operational burden created by false-positive results; those implementation implications are addressed qualitatively in the Discussion (Altman and Bland, 1994; Braunstein et al., 2021).

Study outcome

The key outcome in this study was whether an outbreak of 50 or more infections occurs within 100 days of the initial infection. In sensitivity analyses we also explored whether 20 or more and 10 or more infections occurred within 100 days.

Simulations and statistical analysis

We conducted one thousand simulations for each unique analytic scenario, for a total of 192,000 simulations. To execute in parallel the large number of simulations required, simulations were executed using the Harvard FAS Research Computing Cluster. As the simulations were conducted, summary data from each simulation were extracted and aggregated into a comprehensive analytic data frame. Once the comprehensive analytic data frame was created, a series of logistic regression analyses were performed to summarize the relationships between the probability of an outbreak of 50 or more and the test frequency, test sensitivity, and average isolation delay. The use of logistic regression to summarize these relationships was important because the magnitude of each relationship was unknown prior to running the simulations due to the inherent stochasticity and complex network structure of the model. Two versions of logistic regression analysis were performed: one without interaction terms between predictors (see Table 1) and one with interaction terms between predictors (see Table 2). The contact network simulation, epidemic simulations, and the statistical analysis were all conducted in the R programming language.

TABLE 1.

Conditional probability of an outbreak of 50 or more infections under various testing scenarios, given one infected researcher.

Testing scenario	Probability of outbreak of 50+	Risk ratio
No routine testing: 0 tests per week	0.42	1
Low-intensity, pessimistic: 1 test per week, peak sensitivity of 50%, isolation delay of 2 days	0.3	0.714
Low-intensity, optimistic: 1 test per week, peak sensitivity of 80%, isolation delay of 1 day	0.2	0.48
Moderate-intensity, pessimistic: 2 tests per week, peak sensitivity of 50%, isolation delay of 2 days	0.24	0.57
Moderate-intensity, optimistic: 2 tests per week, peak sensitivity of 80%, isolation delay of 1 day	0.16	0.38
High-intensity, pessimistic: 5 tests per week, peak sensitivity of 50%, isolation delay of 2 days	0.19	0.45
High-intensity, optimistic: 5 tests per week, peak sensitivity of 80%, isolation delay of 1 day	0.12	0.29

TABLE 2.

Logistic regression models of outbreaks of 50 or more infections.

Simple logistic model
​	Odds ratios	Lower bound 95% CI	Upper bound 95% CI	Z-score	P-value
Frequency	0.800	0.796	0.805	−82.124	<0.001
Sensitivity	0.990	0.989	0.990	−29.97	<0.001
Isolation delay	1.155	1.143	1.168	27.050	<0.001
Intercept	0.377	0.367	0.386	−76.291	<0.001

Logistic model with interactions
​	Odds ratios	Lower bound 95% CI	Upper bound 95% CI	Z-score	P-value
Frequency	0.757	0.749	0.764	−56.963	<0.001
Sensitivity	0.992	0.991	0.994	−10.528	<0.001
Isolation delay	1.057	1.040	1.075	6.450	<0.001
Frequency X sensitivity	0.998	0.998	0.999	−9.612	<0.001
Frequency X isolation delay	1.033	1.028	1.038	13.281	<0.001
Sensitivity X isolation delay	1.001	1.000	1.002	3.155	<0.001
Intercept	0.437	0.423	0.451	−49.940	<0.001

Sensitivity analyses

To ensure the analysis results were robust across different thresholds for defining whether an outbreak had occurred, three thresholds were selected for analysis: outbreaks with 50 or more infections, 20 or more infections, and 10 or more infections. In addition, among the subsets of epidemics that exceeded these thresholds, we calculated the proportion of epidemics that were still growing, as indicated by a positive exponential growth rate over the final 20 days. To ensure the pathogen behaved as expected given the epidemiological parameters meant to approximate wild-type SARS-COV-2, we calculated the observed basic reproductive number R0 in the second week after the initial infection in simulations when there was no routine testing, and confirmed that it was consistent with an R0 of 2.0, similar to the WHO’s R0 estimate of 1.95 for the very early stages of the SARS-COV-2 pandemic in China in January 2020 (Liu et al., 2020), as well as to the 2.28 R0 value estimated from the early part of the SARS-COV-2 outbreak on the Diamond Princess passenger cruise ship in 2020 (Zhang et al., 2020).

Role of the funding source

No external funding was used for this research.

Human subjects considerations

No human subjects were involved in this research.

Results

Risk reductions under plausible testing scenarios

By simulating a range of plausible scenarios, we found that a screening intervention with a range of test frequencies, sensitivity, and average delay from positive test to isolation could substantially reduce the probability of an outbreak of 50 or more infections (see Table 1; Figure 2). The probability of such an outbreak in the simulations of the baseline scenario with no routine testing was 0.42. In simulations capturing a low-intensity, pessimistic testing scenario of one test per week, peak sensitivity of 50%, and average isolation delay of 2 days, the probability of a 50+ infection outbreak fell by 29% (RR: 0.71) to 0.30, compared with the baseline risk of 0.42. In simulations capturing a low-intensity, optimistic testing scenario of one test per week, peak sensitivity of 80%, and average isolation delay of 1 day, the probability of a 50+ infection outbreak fell by 52% (RR: 0.48), to 0.2, compared with the baseline risk of 0.42. In simulations capturing a moderate-intensity, pessimistic scenario of two tests per week, peak sensitivity of 50%, and average isolation delay of 2 days, the probability of an outbreak fell by 43% (RR: 0.57), to 0.24, compared with the baseline risk of 0.42. In simulations capturing a moderate-intensity, optimistic scenario of two tests per week, peak sensitivity of 80%, and average isolation delay of 1 day, the probability of an outbreak fell by 62% (RR: 0.38), to 0.16, compared with the baseline risk of 0.42. In simulations capturing a high-intensity, pessimistic scenario of five tests per week, peak sensitivity of 50%, and average isolation delay of 2 days, the probability of an outbreak fell 55% (RR: 0.45), to 0.19, compared with the baseline risk of 0.42. Last, in simulations capturing a high-intensity, optimistic scenario of five tests per week, peak sensitivity of 80%, and average isolation delay of 1 day, the probability of an outbreak fell by 71% (RR: 0.29), to 0.12, compared with the baseline risk of 0.42.

FIGURE 2.

Risk reduction under various testing scenarios*. *Pessimistic testing assumptions: Peak sensitivity of 50%, average isolation delay of 2 days. *Optimistic testing assumptions: Peak sensitivity of 80%, average isolation delay of 1 Day.

Factors predicting outbreaks

At the outbreak threshold of 50 or more infections, each of the three predictors was found to have a significant independent effect on the likelihood of an outbreak occurring, in both the models with and without interaction terms, but the strongest effect was observed for test frequency. In the simple logistic model, each additional day per week of test frequency was associated with a 20% reduction in the odds of an outbreak (see Table 2; Figure 3). In the model with interaction terms, the effect of increased test frequency was moderated by an interaction effect with the average isolation delay (see Table 2). In the logistic model with interactions (see Table 2), increasing test frequency to one, two, and 5 days per week relative to a baseline of no routine testing when peak test sensitivity was 75% (the mean value in the simulations) and there was no isolation delay reduced the odds of an outbreak by 24%, 43%, and 75%, respectively. As the average isolation delay increased, the increase in test frequency led to smaller reductions in the odds of an outbreak. For example, in a scenario with a relatively long average isolation delay of 2 days and a mean test sensitivity of 75%, increasing test frequency to one, two, and 5 days per week relative to a baseline of no testing reduced the odds of an outbreak by 19%, 35%, and 66%, respectively (see Table 2). In the model with interaction terms, the effect of increased test frequency was modestly magnified by an interaction effect with test sensitivity, such that the effect of increased testing was greater at higher levels of test sensitivity.

FIGURE 3.

Marginal Probability of an Outbreak of Various Sizes by Test Frequency*. *Each line represents a smoothed conditional mean probability of an outbreak at a given threshold. Each point represents a unique parameter combination for which 1,000 simulations were run. Minor horizontal jitter was added to the points for visibility, since test frequency was varied in increments of precisely one test per week.

In the simple logistic model, a 10-percentage point increase in test sensitivity was associated with a 10% reduction in the odds of an outbreak (see Table 2; Figure 4). In the model with interaction terms, the effect of increased test sensitivity was slightly moderated by an interaction effect with the average isolation delay (see Table 2). As the average isolation delay increased, the increase in test sensitivity led to smaller reductions in the odds of an outbreak. Similarly, the effect of increased test sensitivity was somewhat magnified by an interaction effect with test frequency, such that the effect of increased test sensitivity was greater at higher levels of test frequency.

FIGURE 4.

Marginal Probability of an Outbreak of Various Sizes by Test Sensitivity. *Each line represents a smoothed conditional mean probability of an outbreak at a given threshold. Each point represents a unique parameter combination for which 1,000 simulations were run. Minor horizontal jitter was added to the points for visibility, since test sensitivity % was always an exact multiple of 10%.

In the simple logistic regression model, an increase of 1 day in the average delay from a positive test to effective isolation was associated with a 15.5% increase in the odds of an outbreak (see Table 2; Figure 5). In the model with interaction terms, isolation delays influenced the likelihood of outbreaks partially through an attenuated main effect (OR of 1.057 compared to 1.155) and partially through interactions with test frequency and test sensitivity that attenuated the protective effects of increasing those variables when isolation delays were longer (see Table 2, Supplementary Tables S4, S5).

FIGURE 5.

Marginal Probability of an Outbreak of Various Sizes by Average Isolation Delay. *Each line represents a smoothed conditional mean probability of an outbreak at a given threshold. Each point represents a unique parameter combination for which 1,000 simulations were run. Minor horizontal jitter was added to the points for visibility, since average isolation delay in days was varied in increments of precisely 1 day.

These results were qualitatively similar at the three outbreak thresholds of 50, 20 and 10 or more infections, as well as across models both without and with interaction terms. Among the subsets of simulations with 50 or more cumulative infections, 20 or more cumulative infections, and 10 or more cumulative infections, respectively, the percentages of epidemics that were still growing as indicated by a positive exponential growth rate over the final 20 days were 32%, 36%, and 32% (See Table 3).

TABLE 3.

Proportion of outbreaks of various thresholds still growing at 100 Days.

Proportion of outbreaks of 50+ growing at 100 days	Proportion of outbreaks of 20+ growing at 100 days	Proportion of outbreaks of 10+ growing at 100 days
0.32	0.36	0.32

Discussion

In this study, we found that the risk reduction from routine screening for LAIs with PPPs be substantial under a wide variety of assumptions about test frequency, test sensitivity, and the average delay from positive test to effective isolation, as well as at various thresholds for defining an outbreak. For example, under our logistic model with interaction (Table 2), assuming peak test sensitivity of 75% and average isolation delays of between zero to 2 days, increasing test frequency to one, two, and 5 days per week relative to a baseline of no testing reduced the odds of an outbreak of 50 or more by 19–24%-, 35%–43%, and 66%–65%, respectively.

It is worth noting that the baseline risk of an outbreak of 50 or more infections conditional on the first lab worker’s infection in the absence of routine testing was high, at 0.42, and while routine testing and isolation interventions substantially reduced the risk, it still remained significant. For example, even under a high-intensity, optimistic scenario of five tests per week, peak sensitivity of 80%, and average isolation delay of 1 day, where the probability of an outbreak was reduced by 71% (RR: 0.29) compared with the baseline risk of 0.42 to a new risk of 0.12, the risk remained far from 0. Among the subsets of simulations with 50 or more cumulative infections, 32% of were still growing over the final 20 days and risked becoming a serious epidemic. This suggests the difficulty of containing the epidemic with the modeled interventions once sustained community transmission is underway.

Test frequency was found to be the most important factor in influencing outbreak risk. This is notable because test frequency is a highly modifiable feature of a screening program’s design. Testing just one or 2 days per week yielded substantial risk reductions in this modeling study, with even greater cumulative risk reductions achievable at higher testing frequencies. While increased test sensitivity was also found to influence outbreak risk, this effect is of modest practical importance for two reasons. First, the incremental reduction in the odds of an outbreak resulting from a given increase in test sensitivity was moderate. Second, in practice, test sensitivity can only be improved within a limited range. The length of average delay from a positive test to effective isolation was observed to be important because longer isolation delays are associated with higher outbreak risks. A longer delay from a positive test to effective isolation will reduce the protective value of greater test frequency and sensitivity by reducing the proportion of an infected person’s infectious time that they spend in isolation.

There may be tradeoffs in program design between optimizing test frequency, test sensitivity, and isolation delays. It would be unwise to infer generalized rules from one modeling study about how all potential tradeoffs in routine screening program design between different test characteristics should be navigated. However, this modeling study does offer evidence that when faced with such tradeoffs, program designers will most likely achieve the greatest benefits by prioritizing high test frequency, within reason.

No new technology would likely be necessary to conduct routine testing for LAIs in high-biosafety research environments. Sensitive, commercially available molecular PCR tests already exist for the major PPPs that are studied in high-biosafety research environments that are of greatest concern, particularly coronaviruses and influenzas (Fleming, 2023; Levine, 2023). Moreover, the key regions of pathogens’ genetic material that are necessary to trigger a positive test result are generally conserved even for pathogens modified in the laboratory (Fleming, 2023; Levine, 2023). If individual research facilities sought more specific tests that could distinguish between viral strains modified in a lab versus those circulating in nature, they would likely have the in-house expertise to develop them (Fleming, 2023; Levine, 2023). To minimize costs and logistical burdens, these more specific tests could be held in reserve for follow-up investigation after an initial positive test result from a commercially available test.

The primary implementation issue not captured by the model is false-positive test results. Positive predictive value depends not only on assay sensitivity and specificity but also on the underlying prevalence of true infection (Altman and Bland, 1994). In a setting where the absolute probability of a true accidental infection during any single testing interval is very low, even a highly specific assay may yield positive results for which the probability of a true accidental infection is meaningfully below 1 (Altman and Bland, 1994; Braunstein et al., 2021). This consideration helps explain why our model does not extend routine screening to the asymptomatic general population and why any laboratory-worker screening program would require carefully designed follow-up procedures. Although false positives would not materially change the model’s estimate of epidemic-risk reduction conditional on a true accidental infection, they could strongly influence the feasibility, acceptability, and cost-effectiveness of implementation. Unnecessary isolation can impose psychological, financial, and workflow costs on workers and institutions, and repeated false alarms may reduce confidence in the screening system or increase the perceived need for additional confirmation before action is taken (Surkova et al., 2020; Lim et al., 2019). For these reasons, any real-world screening program would likely need pre-specified confirmatory testing and occupational-health review before longer isolation or broader response measures are triggered (APHIS, 2023; CDC and USDA, 2021; Braunstein et al., 2021).

Isolation is another implementation issue that the model necessarily simplifies. In the model, effective isolation begins after the specified post-test delay and ends when the infected individual completes the relevant isolated disease states. Using the model transition parameters reported in Supplementary Tables S1 that are drawn from the clinical literature on wild-type SARS-CoV-2, a worker first isolated in the exposed noninfectious stage would on average remain isolated for about 5.2 days in ENISO, 1.3 days in IPISO, and 10 days in ISISO, for a total expected duration of approximately 16.5 days. A worker first isolated in the presymptomatic infectious stage would be expected to remain isolated for about 11.3 days, and a worker first isolated after symptom onset for about 10 days. In practice, however, high-containment facilities would need pathogen-specific operational protocols specifying where isolation occurs, what clinical monitoring is provided, what support services are available, and what criteria trigger release (APHIS, 2023; Arduino and Arndt, 2020; CDC and USDA, 2021). The model effectively treats release from isolation as occurring at the end of the individual’s clinical course, but a real-world program might additionally use symptom resolution and/or testing consistent with reduced infectiousness (CDC and USDA, 2021).

This study has several limitations. First, it does not quantify the likely additional risk reductions that could be achieved by layering additional interventions on top of routine testing and isolation of infected lab workers, such as contact tracing and the use of medical countermeasures. Second, while this study can inform us regarding the relative reductions in the risk of escape of a PPP into the community (conditional on an accidental infection of a lab worker) that could be achieved through a routine testing intervention, this study cannot tell us the absolute risk reductions such an intervention would achieve, because the baseline absolute risks of such accidental infections have not been well-established by the literature. Third, this study does not demonstrate the cost-effectiveness or operational acceptability of the interventions studied, including the consequences of false-positive results, confirmatory-testing workflows, and the institutional requirements associated with worker isolation. Fourth, this study assessed the efficacy of the intervention for a PPP with epidemiological characteristics similar to wild-type SARS-COV-2, and did not address pathogens with different epidemiological traits. Last, this analysis did not attempt to model heterogeneity across research facilities in terms of contact structure, frequency of LAIs, and implementation of the testing and isolation intervention.

Additional research will be needed to answer related questions. First, future research should explore how the effectiveness of such a routine screening system might be influenced by pathogens’ epidemiologic characteristics. Second, future research should attempt to quantify the financial, human resource, and productivity costs of implementing such a screening program, as well as its cost-effectiveness. Third, future research should address other policy questions such as who would finance and operate such a program and what regulatory framework it would operate under. Fourth, future research should address likely implementation questions that would arise such as how to optimally design a screening system under resource constraints and how to maximize fidelity in its implementation. Last, future research should work to more firmly establish the baseline level of risk posed by laboratory escapes of PPPs, perhaps by using alternative approaches like anonymous serology that can measure undetected past exposures.

The large potential benefits of routine screening identified in this modeling study provide evidence in favor of both further modeling research as well as implementation research in the context of a pilot routine screening program. As the evidence base develops, policymakers may wish to consider updating biosafety regulations or guidelines to promote routine screening for LAIs for facilities working with PPPs.

Conclusion

This modeling study finds that routine screening of lab workers in high-biosafety research environments for accidental lab-acquired infections (LAIs) coupled with isolation of infected workers could substantially reduce the risk of a catastrophic escape of a potential pandemic pathogen (PPP) similar to wild type SARS-COV-2, with the extent of risk reduction dependent on the intensity of the testing regime. Test frequency was found to be the most important factor influencing outbreak risk, while test turnaround time and sensitivity also influenced outbreak risk. The risk reduction from routine screening remained large under a range of assumptions about test frequency, test sensitivity, and the average delay from positive test to effective isolation, as well as at various thresholds for defining an outbreak.

Funding Statement

The author(s) declared that financial support was not received for this work and/or its publication.

Data availability statement

The datasets presented in this study can be found in online repositories. The names of the repository/repositories and accession number(s) can be found below: https://github.com/byroncohen/Biosafety_Routine_LAI_Screening.

Author contributions

BC: Conceptualization, Methodology, Writing – original draft, Writing – review and editing. BR: Methodology, Writing – review and editing. WH: Conceptualization, Methodology, Writing – review and editing. NM: Conceptualization, Methodology, Supervision, Writing – review and editing. KC: Conceptualization, Methodology, Supervision, Writing – review and editing.

Conflict of interest

The author(s) declared that this work was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Generative AI statement

The author(s) declared that generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Supplementary material

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fbioe.2026.1714582/full#supplementary-material