Contributions of modelling for the control of COVID-19 nosocomial transmission.

Introduction

Since early 2020, the COVID-19 pandemic has had major impacts on healthcare facilities worldwide. In acute care hospitals, the unprecedented influx of severe COVID-19 patients over the different epidemiological waves has caused a heavy burden, leading to care disorganisation, stress and fatigue for the staff. Healthcare workers, as well as non-COVID-19 patients, have also been at high risk of SARS-CoV-2 infections. This was evidenced by numerous reports of outbreaks within healthcare settings worldwide. A recent UK study estimated that 95,000–167,000 patients had acquired SARS-CoV-2 in hospitals from June 2020 to March 2021 [1]. Another study suggested, using modelling, that 73% of infections in UK healthcare workers resulted from nosocomial transmission during the first pandemic wave in England [2].

Long-term care facilities (LTCFs) are also particularly vulnerable to SARS-CoV-2 nosocomial transmission: they host frail and/or elderly patients and daily care involves frequent at-risk contacts between patients and health care workers. As a consequence, even in a context of high restrictions such as visit prohibition or systematic hygiene or prevention measures implementation, patients from LTCFs have suffered from particularly high morbidity and mortality from COVID-19. Indeed, LTCFs carried a disproportionately high portion of the pandemic’s mortality burden. For instance, the portion of all COVID-19-related deaths occurring in LTCFs reached up to 66% depending on the country in Europe [3], and 50% over 26 USA states [4].

In this context, healthcare professionals, hygienists and public health decision-makers have faced unprecedented challenges. They had to implement a wide range of interventions to improve the prevention, surveillance and control of SARS-CoV-2 infections and avoid local transmission, while dealing with high uncertainty related to their potential impact. In this regard, mathematical models can be useful. Indeed, over the course of the COVID-19 pandemic, they have become ubiquitous to support public health decision [5]. By reproducing in silico populations and formalising mechanisms of transmission and infection, they allow the analysis of longitudinal data and the simulation of counterfactual scenarios of interventions to evaluate their impact and efficacy (see Box 1 for a simple illustration of a model in the nosocomial context).

Classically, models group individuals of a hospital population by defining a set of compartments characterising individual status, location or infectious state. Individuals can then circulate from one compartment to another according to specified rates. A simple version of a compartmental model of SARS-CoV-2 transmission within a ward population is shown on the figure. Here, patients (P) and health care workers (H) are considered separately. The model specifically describes the natural history of SARS-CoV-2 infection of individuals. When susceptible individuals (S), whatever their status (patient or HCW), encounter infectious individuals, they can acquire the virus and become exposed (E). After a latency period, they will move to an Infectious compartment, whether symptomatic (IS) or asymptomatic (IA). Finally, after certain duration, they will recover or die (R).

Alt-text: Box 1

Contributions of mathematical modelling for healthcare settings

Since the beginning of the pandemic, teams previously working on the mathematical modelling of nosocomial risk and other infectious disease modellers have worked extensively to better understand SARS-CoV-2 spread in health care, predict future risk, and assess control strategies. From previous experience of Influenza, SARS or antimicrobial resistance, it was clear early on that the nosocomial epidemic risk was going to be high, and that modelling, informed with specifically collected data, could help to support decision and measure implementation in a critical time.

In France, a working group, composed of scientists with previous experience working on healthcare-associated infections, including modellers, epidemiologists, clinicians and virologists, was set up in the spring of 2020. Regular discussions within this group during the first wave aimed to discuss issues related to the nosocomial spread of SARS-CoV-2. This has led to several interdisciplinary collaborations on specific studies including: (a) a theoretical work underlining that the basic reproduction ratio R 0 could differ widely between healthcare settings and the general community, as well as between different healthcare settings [6]; and (b) innovative data collection on interindividual contact patterns within different types of hospital wards during the first epidemic wave (NodsCov project).

One the international scale, several groups have proposed models of SARS-CoV-2 spread in healthcare settings. Overall, models have addressed the three categories of interventions mentioned above, namely prevention, surveillance and control. However, the focus of modelling works of nosocomial COVID-19 has shifted over time, in parallel to changes in the international epidemic situation and evolution of available therapeutic (e.g., vaccines) or non-therapeutic (e.g., antigenic tests) tools (Fig. 1, Fig. 2 ). In addition, some issues were investigated less often than others; to this day, the main topics addressed by published models have been testing strategies and hospital organisation (Fig. 2).

Fig. 1

Timeline of results obtained from modelling works over the pandemic course. For each key message obtained from modelling regarding hospital organisation (in blue), testing strategies (in green), protective equipment (in red), vaccination (in yellow) or risk assessment (in grey), the month when it was first made available (through publication in a journal or public archives) is provided [2], [7], [8], [9], [10], [11], [12], [13], [14], [15], [16], [17], [18].

Fig. 2

Cumulative number of identified modelling papers addressing different research questions, as a function of time. Here, publication times refer to the dates of first publication (on any public archive).

Active surveillance

Despite high levels of prevention, many healthcare settings have experienced SARS-CoV-2 outbreaks. When these occur, following SARS-CoV-2 introduction by an asymptomatic infected health care worker, visitor or resident, the presence of the virus in the ward is often detected late after its introduction. As a result, by the time of detection, large “COVID-19 clusters” are observed and have to be managed [18]. Surveillance is key to allow early detection and avoid such situations, but may prove time-consuming and costly, especially in contexts of limitation of test availability. Many modelling works have therefore attempted to identify efficient testing strategies that optimise the use and distribution of available tests to both staff and patients, often in LTCFs where on-site PCR tests were most of the time unavailable. They have provided evidence in favour of systematic screening and isolation very early on in the epidemic (the first models on this topic were made available in May 2020) (Fig. 1). Later models on this topic have shown how group testing could prove useful in low-resource healthcare settings, or compared the efficacy and efficiency of antigenic and PCR tests.

Prevention and control

Since the beginning of the pandemic, many interventions have been implemented in healthcare settings to prevent SARS-CoV-2 transmission. These notably include reinforced contact precautions (using masks or other personal protective equipment), staff rescheduling, or vaccination. Furthermore, to best manage SARS-CoV-2 outbreaks within healthcare settings when they occur, several control strategies may be implemented. These include confirmed or suspected case isolation, test-and-trace, staff cohorting, etc.

A major objective of models has therefore been to assess the effectiveness of each of these interventions, used separately or in combination. Several models have explored the question of hospital organisation, with for instance early models underlining the potential benefits of rotating staff to preserve the workforce, and many works suggesting that the implementation of an isolation ward dedicated to suspected and certain SARS-CoV-2 cases could reduce epidemic transmission. Modelling has also helped investigate the impact of holding areas for screening upon admission or of shield immunity-based staffing. In addition, early on, many models have accounted for the use of personal protective equipment such as masks, suggesting that these could have a large impact. Finally, at the end of 2020, models have started to explore the benefits of vaccination, with the most recent models on this topic also looking at the impact of a booster dose.

Interactions with clinicians

In order to really support SARS-CoV-2 risk management in healthcare settings and address relevant questions, models need to build on interactions between modellers and clinicians in the field. Two main approaches may be considered in this regard:

Raising questions to motivate theoretical work

First, discussions with clinicians may help modellers identify the most pertinent questions given the current epidemic context and resources within healthcare settings. This may in turn motivate the development of theoretical models specifically allowing exploring these questions.

As an illustration, discussions among the aforementioned interdisciplinary working group on health care associated risk set up in France in March 2020 soon made clear that there was a need to better understand the epidemic potential of the virus within healthcare settings [6]. Indeed, this had been thoroughly investigated in the general population, with numerous estimates of the basic and effective reproductive numbers R 0 and R e, but not in hospitals. To get insight into that question, a framework to estimate R 0 from hospital outbreak data was developed, and applied to data collected in a hospital in the Paris area, providing one of the first estimates of this indicator [17].

Real time close support to decision-making

Second, decision-makers within a given healthcare institution may contact modellers with a direct request for help to inform their choices, for instance in controlling a COVID-19 outbreak. Collaborative work will then ensue, with a specific model being developed over back-and-forth discussions.

As an illustration, during the first epidemic wave in the spring of 2020, we worked closely with a psychiatric hospital that was facing a COVID-19 outbreak. Through modelling, we were able to show that, in this specific context, the isolation ward that they had opened to host all identified COVID-19 cases was a highly effective measure, provided all symptomatic patients were systematically screened for SARS-CoV-2. On the contrary, the screening areas that they were proposing to implement, in which all newly admitted patients would have been led awaiting clinical assessment and PCR test results if they were symptomatic, added very little benefit to the isolation ward [16].

Discussion

In the global fight against the COVID-19 pandemic, mathematical models have proved essential both to understand better the virus features and its transmission within populations and to support decision-making for control measure implementation. Here, we underline how they have largely contributed to better understanding, preventing and controlling the spread of SARS-CoV-2 in healthcare settings, starting from very early on in the pandemic.

However, reviewing the models that have been published over the last two years, we also identify an imbalance in the issues investigated. Many of these models have focused on long-term care facilities and on testing strategies, while the effectiveness of protective equipment (masks, for instance) has rarely been explored through modelling. This is notably due to the fact that, in a mathematical model, simulating the impact of a mask generally translates as a simple reduction in the model parameter simulating transmission. Consequently, from a modeller point of view, this question may not have appeared to justify the use of such a complex model. In addition, little data was (and is still) available to properly inform this reduction, forcing modellers to fix arbitrary parameters values. However, in the field, deciding whether masks should be used in different types of interactions, which types of masks (e.g., FFP2 vs. surgical) should be used, and to whom they should be prioritised in resource-low situations, as was the case early on in the pandemic, are all questions that may be of the utmost importance. In the absence of data, model parameterisation based on expert elicitation is also possible, in addition to sensitivity analyses.

In addition, although large contributions of models early on in the pandemic for the improvement of surveillance and case detection (Fig. 1), their use for decision-making in hospitals has been limited. This can come first, from the too high technicity of modelling articles, which are often published in specialised journals, and may not be visible to clinicians, hygienists or policy-makers on the field. Second, despite large evolution in recent years, modellers still fail in convincing a part of the medical community of the applicability of modelling results, which are mostly seen as theoretical results disconnected from the field.

Such examples and others underline that, in order to maximise modelling utility in supporting outbreak control and measures implementation in health care settings, close interactions between clinicians and modellers are key to build models addressing the most relevant questions on the field and generate more accessible results that could be considered as more realistic and readily applicable.

Conflicts of interest

The authors declare no conflict of interest