Resilience as a measure of preparedness for pandemic influenza outbreaks

ABSTRACT

The global crisis generated by COVID-19 has heightened awareness of pandemic outbreaks. From a public health preparedness standpoint, it is essential to assess the impact of a pandemic and also the resilience of the affected communities, which is the ability to withstand and recover quickly after a pandemic outbreak. The infection attack rate has been the common metric to assess community response to a pandemic outbreak, while it focuses on the number of infected it does not capture other dimensions such as the recovery time. The aim of this research is to develop community resilience measures and demonstrate their estimation using a simulated pandemic outbreak in a region in the USA. Three scenarios are analysed with different combinations of virus transmissibility rates and non-pharmaceutical interventions. I The inclusion of the resilience framework in the pandemics outbreak analysis will enable decision makers to capture the multi dimensional nature of community response.

1. Introduction

The COVID-19 outbreak in early 2020 has impacted countries worldwide and until the first quarter of 2021 it has caused more than three million of deaths (Araz et al., 2020; Manski & Molinari, 2021). Similarly, influenza ranks among the most common viral infection. Infections spread when susceptible individuals are in contact with infected animals or persons. The result of infection is often fatal for senior adults and children. The centre for disease control and prevention (CDC) estimates that for the 2015–2016 season, vaccination in the USA has prevented around 5.1 million influenza illnesses, 2.5 million influenza-associated medical visits and 71,000 influenza-associated hospitalisations. In the other hand, the burden of influenza in 2015–2016 was estimated to be 25 million illnesses, 11 million medical visits, 310,000 hospitalisations, and 12,000 cases of pneumonia and influenza inflicted deaths (Center for disease control and prevention, 2016).

Main strains of influenza viruses are currently in circulation, e.g., H7N9 (China), H1N1 (worldwide), H5N1 (worldwide). A(H7N9) is a subtype of influenza A virus that is found in birds including poultry. Since March of 2013, A(H7N9) has been noted to infect humans in several regions of China, especially those who are in close contact with poultry either at farms or at markets dealing with poultry. So far, there have been four waves of infections in spring of 2013 and winters of 2013–2014, 2014–2015, and 2015–2016. The fifth wave of infections is currently in progress. Since spring 2013 till March 2017, a total of 1307 laboratory-confirmed cases of A(H7N9) infections have been recorded in different regions of China causing at least 489 deaths (37.4% fatality rate; World Health Organization, 2017b). It has also been noted in the literature that the high exposure to infected poultry contributed to the elders more infected than other age groups. Though most of the infections are confirmed to be animal-to-human transmitted, researchers believe that some cases may have been human-to-human transmitted. However, current epidemiological and virological evidence suggests that this virus has not acquired the ability of sustained transmission among humans (World Health Organization, 2017a). Apprehension of A(H7N9) showed that it might gain the ability to mutate or reassort to become human-to-human transmittable and cause a pandemic. Hence, it is important to be able to assess the level of resilience of the mitigation measure that are available to assess the possible impact and to measure how to protect the population from a pandemic outbreak. Readers may recall that there was a significant fear of a H5N1 influenza pandemic outbreak during the years 2003–2009, and till date, the virus has infected a total of 858 people worldwide, of which 453 are dead (52.8% fatality rate; World Health Organization, 2017b). The fear of H5N1 pandemic triggered research studies published in 2005 (Ferguson et al., 2005) and 2006 (Ferguson et al., 2006), which critically examined the impact of potential H5N1 outbreaks in Thailand and jointly in the USA and UK, respectively. Though an H5N1 pandemic did not occur as yet, this avian influenza virus is still in circulation and is still causing deaths, as reported by WHO (World Health Organization, 2016). In 2009, Influenza H1N1 attacked worldwide including the USA, where the burden has been estimated to be 60.8 million of infections and 12,469 deaths (Shrestha et al., 2011).

Though the impact of pandemic outbreaks are generally measured via the infections and deaths numbers, their impact on the economic activities, education, and overall healthcare delivery are significant. This has been the case of COVID-19, which in addition to the infected people, it has created economic crisis and social impacts worldwide (Nicola et al., 2020). Such negative impacts also arise from natural disasters affecting infrastructure and communities by disrupting interconnected systems and generating high death tolls and economical losses (Santos et al., 2014). In the case of pandemic outbreaks, even though physical infrastructure is not impacted directly, they disrupt the workforce availability leading to reduced productivity (Nurchis et al., 2020; Orsi & Santos, 2010). In order to avoid or minimise the effects of pandemic outbreaks, communities thus should assess and improve their resilience. The importance of communities resiliency has been highlighted by the CDC, which defined it as “the ability to prevent, withstand and recover from public health incidents” (Plough et al., 2013).

Traditionally, pandemics preparedness has been measured using the infection attack rate (IAR; Center for disease control and prevention, 2012; Riley et al., 2007). This metric quantifies the total percentage of infected individuals in a population by the end of an outbreak. During the ongoing COVID-19 crisis, this is the common metrics used by authorities and media outlets to report the pandemic. Even though the percentage of infected people is a key dimension to understand community response it leaves out other critical response variables such as recovery time, resourcefulness and adaptive capacity. Resilience is a multivariate metric and thus a more adequate tool to estimate a community’s ability to cope with a pandemic outbreak.

The measure of community or social resilience is defined incorporating these different capacities: absorptive, restorative and adaptive (Keck & Sakdapolrak, 2013; Maguire & Hagan, 2007). In the context of pandemics, the absorptive capacity is characterised by the number of infected people, and the restorative capacity is measured by the time to recover, and finally, the adaptability refers to the ability to learn from experiences and positively influence system resilience in future (Folke et al., 2010). Therefore, the goal of a resilient community is to minimise a joint measure, the extent of infected population and the recovery time.

Resiliency is a key attribute of any system to preserve and improve performance outcomes throughout a period of adversities. This has led to the exploration and implementation of this concept in different areas of study. Since its introduction in 1973 (Holling, 1973), the concept of resilience has been adopted in engineering (Youn et al., 2011), psychology (Richardson, 2002), economics (Rose, 2004), ecology (Carpenter et al., 2001), management (Ponomarov & Holcomb, 2009), among others. However, in the analysis of pandemics influenza outbreaks, resilience has not been introduced as a metric of performance. Also, there is no commonly accepted model or approach to measure resilience, and hence it is assessed using different qualitative and quantitative tools (Hosseini et al., 2016). The most common approaches to measure resilience in engineering are the system ability to absorb energy and the system average performance (Bruneau et al., 2003; Zobel, 2011). The former does not consider recovery after a disruption, which limits its implementation in community resilience. The average performance is more suitable to be adapted to assess community resilience based on the system performance. In the case of community resilience and public health resilience indexes, they assess system preparation and the ability to respond before a disruption occurs but do not focus on system performance response when a disruption affects the system. In this research, we review the available quantitative metrics with a goal of using them to assess the impact of different system interventions or policies during a pandemic outbreak. The metrics are classified in two categories: the multiple indexes and the average performance metrics.

In the first group, a wide variety of metrics are combined to express a system’s resilience level (Cutter et al., 2008). One such resilience model is the multiplication of three individual metrics that capture absorptive, restorative and adaptive capacities (Francis & Bekera, 2014). Composite metrics are developed via different methodologies that combine social, economical and environmental factors to explore multiple dimensions (Asadzadeh et al., 2015; Cutter, 2016). These metrics or sets of indexes In the second category, we discuss the average performance (or, the resilience triangle) metrics the first of which was introduced by Bruneau et al. (2003) in the context of community resilience under seismic events. Average performance metric captures both the absorptive and restorative capacities. This metric paradigm has been subsequently modified to fit linear responses (Zobel, 2011), non-linear performance functions (Cimellaro et al., 2014), multiple disruptions (Zobel & Khansa, 2014), stochastic parameters (Ayyub, 2015), and time-dependent metrics (Ouyang & Dueñas-Osorio, 2012). The popularity of this approach resides in the interpretability and ease of use in evaluating strategies to improve a system’s resilience.

Resilience was used as a measure to develop communication management strategies during events like earthquakes, flu, and terrorism (Longstaff & Yang, 2008). To our knowledge, our work as presented in this paper is the first formal effort to evaluate the impact of non-pharmaceutical interventions (NPIs) on a community’s resilience levels during an influenza pandemic. Furthermore, this is a first attempt to measure performance based resilience in a pandemic context.

The remaining sections of this paper are as follows. Section 2 describes the methods and metrics that are implemented to assess resilience, as well as the agent based simulation test bed used in this study. Section 3 explores the simulation results and the corresponding resilience measures. In Section 4, the results are used to identify the pros and cons of the methods proposed in this paper to study community resilience under pandemic outbreaks. Finally, in Section 5 the contribution of the resilience framework in healthcare is highlighted and future research areas are identified.

2. Methods

In this study, we have used the latest version of an agent-based (AB) simulation model that was developed in our previous studies of influenza pandemic (Das et al., 2008; Martinez & Das, 2014; Uribe-Sánchez et al., 2011). The AB model replicates an influenza pandemic outbreak considering epidemiological parameters (e.g., incubation and latent period, force of infection); demographic information about the population (e.g., household members and their age, sex, parental status, distance between infected and susceptible individuals), infection process, contact process, and disease natural history.

Our study considers different potential virus transmissibility scenarios as well as different types and extent of non-pharmaceutical interventions (NPIs) including isolation, household quarantine, school, and workplace closure.

We execute the AB model for various combinations of the virus transmissibility rates and the intervention methods. The model runs generate the infection pattern as well as the attack rates, which are used to estimate resilience.

2.1. AB simulation model

The AB model creates individuals and their designated households (see Figure 1) in the outbreak region. The model thereafter creates schools, workplaces and communities, and the daily activity schedules suitables for various types of individuals. The model initiates the outbreak by releasing infected people in the population and keeps track daily of the status of each person. Susceptibles that come in contact with infected individuals gather force of infection, the levels of which are used to determine if the susceptibles become infected. The infected follow through a disease natural course and either recover and become immune or die.

Figure 1.

AB simulation model.

The AB model considers three NPIs scenarios: first, when no interventions are applied; second, when an arbitrary set of NPIs (Martinez Torres, 2012) are used; and the third considers a recommended NPI (Martinez & Das, 2014). Three intervention scenarios and three virus transmissibility rates are combined into nine scenarios. for each case, the output summary file contains relevant information including number of infected, number of recovered, and number of deceased.

2.2. Interventions

Mitigation containment strategies for pandemic influenza include both pharmaceutical and non pharmaceutical interventions (PI and NPI). The PIs include vaccines and antivirals. Vaccines usually take several months to develop, produce, and distribute. The stockpile levels of antivirals at the time of an influenza pandemic outbreak are limited. Moreover, antivirals don’t develop infection immunity but can reduce its severity. The NPIs include social distancing, household quarantine, isolation, school and workplace closure. These strategies can be executed at early stages of a pandemic outbreak. For this study, only NPIs are used. In under resourced countries, this is the best available measure to deploy due to the unavailability of vaccines and antivirals. Another advantage of using NPIs is that they can provide the first line of defence before vaccines are developed.

In this study, we applied a set of NPIs with certain parameter combinations that were recommended by Martinez and Das (2014), as shown in Table 1 (NPI 2). For the purpose of comparison, we also implemented another NPI 1 (Martinez Torres, 2012) with somewhat random set of parameters. These interventions are supposed to act as a containment measure for the spread of virus. These experiments are not intended to identify the best set of parameters or NPIs combinations, the objective is to analyse the system response using a resilience assessment framework and make a contrast of the different strategies evaluated.

Table 1.

NPI parameters.

#	Measure	NPI 1	NPI 2
1	Global Threshold	10	10
2	Deployment delay	3 days	7 days
3	Case isolation threshold	1 day	1 day
4	Case isolation duration	7 days	10 days
5	Case isolation compliance for workers	75%	75%
6	Case isolation compliance for non-workers	84%	57%
7	Household quarantine threshold	1 day	1 day
8	Household quarantine duration	7 days	7 days
9	Household quarantine compliance workers	75%	53%
10	Household quarantine compliance non-workers	84%	84%
11	Cases to close a class in a school	4	1
12	Classes to close a school	6	3
13	School clsure duration	10 days	21 days
14	# Cases to close a department in a workplace	6	3
15	% of departments to close a workplace	60%	30%
16	Workplace closure duration	10 days	7 days

Global threshold refers to the number of cases needed to declare an outbreak of influenza. Deployment Delay is the time needed to fully deploy NPIs after the onset of an outbreak. Case isolation is the time that an infected individual should stay at home. Household quarantine is related to the restriction of movement of household members of an infected case (not the infected individual). School closure requires number of infected students in a class and number of classes infected in a school to close the class and the school, respectively. Workplace closure is similar to school closure but consider workplaces instead of schools.

2.3. Testbed data

In the testebed a simulation of a pandemic outbreak was carried out in a community of 1.2 million people. We do not split this population in multiple communities. The epidemiological parameters (e.g., force of infection, incubation and latent periods, basic reproduction number, and fatality rate) were estimated from the studies on H7N9 influenza outbreak in China (Liu & Fang, 2015; Silva et al., 2017; Wang et al., 2019). The U.S. census and travel data were used for households, schools, workplaces, communities, and human behaviour (e.g., contacts, compliance to isolation).

2.4. Measuring resilience

The approaches that we have selected to quantify resilience levels are via measures of average performance and cumulative loss function. These metrics consider the key dimensions of number of infected people and the recovery time. Another key feature of these metrics is the simplicity to be operationalised in a simulation model. In contrast, composite indices require multiple indicators that are difficul to operationalise to capture potential changes in the system’s policies.

We define the fraction of the healthy population at any time $\mathit{t}$ as $\mathit{H}(\mathit{t})$. This is the critical performance variable of community resilience, where $\mathit{I}(\mathit{t})$ denotes the number of infected at time t, $\mathit{Rec}(\mathit{t})$ and $\mathit{D}(\mathit{t})$ represent the number of recovered and the number of deceased due to the disease, respectively. The parameter $\mathit{T}$ represents the number of days since the outbreak started.

$$\mathit{H}(\mathit{t})=\frac{\mathit{H}(\mathit{t}-1)-\mathit{I}(\mathit{t})+\mathit{Rec}(\mathit{t})}{\mathit{Population}-{\int }_0^T\mathit{D}(\mathit{t})}$$ (1)

The global resilience metric $\mathit{R}$ is adapted from Bruneau et al. (2003). In the formulation $T_{\mathit{OE}}$ is the event time, this the moment where the disruptive event strikes the system, in the context of this research is when the outbreak occurs. The parameter $T_{\mathit{LC}}$ is the total control time, this time is greater or equal the recovery time, it is defined by the decision maker. The average performance metric is measured during the $T_{\mathit{LC}}$ . This metric is between 0 and 1, the higher the value of $\mathit{R}$ then the better the system was able to sustain and recover from the disruption.

$$\mathit{R}=\underset{{\mathit{T}}_{\mathit{OE}}}{\overset{T_{\mathit{OE}}+T_{\mathit{LC}}}{\int }}\frac{H_t}{T_{\mathit{LC}}}\mathit{dt}$$ (2)

In addition, the time dependent average performance model measures previous, current and future resilience (Ouyang & Dueñas-Osorio, 2012). In this formulation, $\mathit{H}(\mathit{t})$ is the observed or real performance function, and $\mathit{TH}(\mathit{t})$ is the desired or target performance function. The value of $\mathit{TH}(\mathit{t})$ throughout the resilience analysis is 100%. Both $\mathit{R}$ and $\mathit{R}(\mathit{t})$ measure the average performance, the difference is that $\mathit{R}$ captures resilience during the $T_{\mathit{LC}}$, while the $\mathit{R}(\mathit{t})$ measures resilience at any time $\mathit{t}$ from the disruption up to $T_{\mathit{LC}}$

$$\mathit{R}(\mathit{t})=\frac{\underset{t_0}{\overset{\mathit{t}}{\int }}\mathit{H}(\mathit{t})\mathit{dt}}{\underset{t_0}{\overset{\mathit{t}}{\int }}\mathit{TH}(\mathit{t})\mathit{dt}}$$ (3)

The metrics from (2) and (3) are calculated to compare global and partial resilience in different time intervals. The evaluation is performed in all nine cases described on the previous Section.

The cumulative loss function $\mathit{L}$ is proposed as the community capacity to absorb performance loss over time, this is the complement of the performance or the area above the performance curve. An example of the cumulative loss function is the number of deaths or infected individuals up to a time $\mathit{t}$. The following equation quantifies the loss up to time $\mathit{t}$ as the cumulative percentage of the infected population.

$$\mathit{L}=\underset{t_0}{\overset{T_{\mathit{LC}}}{\int }}(\mathit{TH}(\mathit{t})-\mathit{H}(\mathit{t}))\mathit{dt}$$ (4)

The expression in (5) is used to compare resilience among the suggested policies. The concept of relative resilience denoted as $R^{\prime }$ is introduced as the ratio of a policy cumulative loss and the optimal strategy cumulative loss. Where $H^{\ast }(\mathit{t})$ is the healthy population percentage function when the optimal or best policy is implemented. When values of $R^{\prime }$ are close to 1 the policy provides an equivalent resilience performance compared to the best policy found. This metric can be interpreted as the fraction of the minimum loss compared with the loss of a policy to evaluate similarities and differences. Values below 0.5 imply that the loss is at least 2 times the loss of the optimal policy.

$$R^{\prime }=\frac{\underset{t_0}{\overset{T_{\mathit{LC}}}{\int }}(\mathit{TH}(\mathit{t})-H^{\ast }(\mathit{t}))\mathit{dt}}{\underset{t_0}{\overset{T_{\mathit{LC}}}{\int }}(\mathit{TH}(\mathit{t})-\mathit{H}(\mathit{t}))\mathit{dt}}$$ (5)

The resilience analysis is complemented with two auxiliary variables such as the initial loss $\mathit{X}$, which is an estimation of the performance drop or absorptive capacity. Both $\mathit{X}$ and $\mathit{L}$ capture the absorptive capacity, the difference is that the former measures the maximum performance drop, while the latter estimates the performance drop throughout time. The second auxiliary variable is the recovery time $T_r$, which is the time that elapses from the pandemic disruption until the infected population is less than 1 %.

The proposed metrics are intended to capture the community capacity to absorb and recover from an influenza pandemic outbreak. The indexes reflect the system resiliency level at different times and as a response of interventions strategies to mitigate the impact and speed up the recovery process.

3. Results

The interpretation of the metrics $\mathit{X}$, $T_r$, and $\mathit{R}$ provide a general idea of the community resiliency levels and the type of resilience, i.e., whether the system is capable of absorbing the initial impact or the restorative capacity is well developed. $\mathit{X}$ is the maximum drop in $\mathit{H}(\mathit{t})$, thus the smaller the value, the higher is the absorptive capacity. The variable $T_r$ represents how quickly the community healthy population is restored to the initial or desired state. In the case of $T_r$, the smaller value it has, the faster will the community recover. The value of $\mathit{R}$ combines both variables ($\mathit{X}$ and $T_r$) in estimating the global resilience.

The scenarios comparison is divided based on the transmissibility rates. Figures 2-4depict the daily % of healthy population in the low, medium, and high transmissibility rate cases respectively. For each transmissibility rate the three policies are evaluated, these are baseline, NPI 1 and NPI 2, for a total of nine scenarios. Table 2 shows the resilience levels summary for all scenarios. A predominant observation made from Table 2 is that the cumulative loss $\mathit{L}$ is smaller in the best available NPI policy regardless of the transmissibility rate.

Figure 2.

Daily % of healthy population in the low transmissibility rate scenario.

Figure 3.

Daily % of healthy population in the medium transmissibility rate scenario.

Figure 4.

Daily % of healthy population in the high transmissibility rate scenario.

Table 2.

IAR (transmissibility) comparison and estimated resilience levels.

IAR	Policy	X (%)	$T_r$	L	R (%)	R’
	Baseline	16.00	49	3.01	97.79	0.137
Low	NPI 1	9.49	114	1.95	98.00	0.232
	NPI 2	1.34	36	0.45	99.32	–
	Baseline	31.62	48	5.05	96.63	0.344
Medium	NPI 1	21.61	127	3.73	97.5	0.466
	NPI 2	6.71	127	1.74	98.39	–
	Baseline	39.97	49	6.45	95.69	0.409
High	NPI 1	31.30	158	4.83	96.74	0.547
	NPI 2	11.26	158	2.64	97.67	–

In the analysis of all the transmissibility scenarios from Table 2, the metrics $\mathit{L}$, $\mathit{R}$, and $\mathit{X}$ are better when NPIs are implemented. For $T_r$, the behaviour was not as expected due to non-linearities of the responses, which are discussed in more details in the next section.

Since global resilience values from Table 2 look alike, the Equation (3) was implemented to measure time dependent resilience. This metric enables a comparison in different time intervals by capturing the resilience levels evolution from the disruption until the community has fully recovered. Figure 5 depicts the daily cumulative resilience in the high transmissibility rate case. The analysis of the time dependent resilience metric confirms that NPI (2) outperforms the other scenarios, and the biggest difference occurs during the maximum performance loss. After the recovery time has elapsed the time dependent resilience converges to the global resilience metric. It is noticeable that while the differences using the scenario average performance over $T_{\mathit{oe}}$ are around 2 percentage points, in the case of the time dependent metric these differences grow to 10 percentage points distance.

Figure 5.

Daily cumulative resilience in the high transmissibility scenario.

Furthermore, the time-based metrics capture changes in the resilience slope and trajectory. In the Figures 3 and 4of the medium and high transmissibility scenarios there are intervals after the recovery where the percentage of healthy population decreases or stays at the same level. These resilience losses are generated by pandemic waves.

The cumulative loss $\mathit{L}$ from (4) is captured in Table 2 for all the scenarios and Figure 6 shows the performance loss at time $T_r$ for the high transmissibility scenario. This metric highlights that differences among interventions are higher in contrast with the global and time dependent average performance. Then, the relative resilience metric (R’) from (5) is measured to compare the intervention policies. As a result, resilience levels are more aligned with the number of infected population and a clear benefit of the NPI 2 is suggested. The value of $R^{\prime }$ is equal to 0.137 in the low transmissibility scenario, which indicates that the loss of the baseline policy is about 7.25 times higher compared to the optimal policy. In the case of the high transmissbility the value of 0.547 indicates that losses are 1.82 when comparing NPI 1 and NPI 2.

Figure 6.

Daily cumulative loss in the high transmissibility scenario.

4. Discussion

The global and time dependent resilience metrics confirm the benefits of the NPI 2 (Martinez Torres, 2012). The value of $\mathit{X}$ is reduced in NPI 1 and NPI 2. This result suggests that the absorptive capacity is improved. In the case of the response variable $T_r$, the situation is not the same because the recovery time increased for the medium and high attack rates. Even though the overall system performance is enhanced by the static resilience, the dynamic counterpart is affected.

To understand the underlying reasons of the restorative capacity worsening, two conditions are analysed. First, in both NPIs scenarios after the day 50 there are infection waves where the number of infected people increases but not up to the levels of the first wave. Although the number of infected people is lower during the first wave in the NPIs scenarios, then when the virus re-emerges less people are susceptible to get infected. In the baseline scenario since most of the population is infected in the first wave, even less people are affected in the subsequent outbreaks. The second cause is that the NPIs are static or a one time intervention, thus do not kick in when outbreaks waves disrupt the system.

The presence of multiple outbreaks or infection waves are part of the Influenza virus normal pattern (Miller et al., 2009). This behaviour has been documented in the resilience literature as the multi-disruption scenario, where natural disasters such as earthquakes disrupt a community multiple times in waves or replicas (Zobel & Khansa, 2012, 2014). During the COVID-19 pandemic this pattern of several waves has been observed on multiple countries such as USA (Solis et al., 2020), Italy, India, among others. The mitigation of the upcoming waves requires that policies should be designed to minimised $\mathit{X}$ and $T_r$. This suggestion can be performed by evaluating the potential impact of the implementation of dynamic NPIs or PIs (Pharmaceutical interventions) that may reduce the probability of getting infected in any of the expected outbreaks subsequents to the first one.

The scenarios comparison was carried out with the average performance metrics, but it is possible to test and compare alternative metrics that could include additional dimensions or different interpretations. The relevance of other metrics is given by how close were the resilience measurements in the results section despite the profound differences in the number of infected people. Additionally, the severity of the pandemic outbreak can be overlooked due to the $\mathit{R}$ values from Table 2 are reported as 95% or above. This limitation was overcome by using the time dependent measurement and the cumulative loss function. From the implemented metrics it is considered that the cumulative loss function and the proposed relative resilience ratio from (5) were more accurate to point out the real differences and contribution to resilience of the three intervention strategies analysed.

The insights provided by the resilience framework to the pandemics analysis enable a broader understanding of community prevention and recovery capacities, and unifying multiple dimensions under the resilience conceptual umbrella. In addition to the analysis of the number of infected and deaths, the resilience framework adds the time component to understand the recovery time and performance loss throughout the pandemic. The COVID-19 pandemic has shown that the recovery time is important to reactivate world economies (Haldane et al., 2021).

Our study considers epidemiological parameters from H7N9 influenza outbreak in China and uses demographic parameters from USA. However, this study is not limited to H7N9 and can be interpreted as for any other virus to evaluate the system resiliency.

5. Conclusions

The inclusion of the resilience framework in the pandemic analysis enriches the discussion in how to design effective interventions to prevent, mitigate and recover from a potential pandemic outbreak. The ability to capture the multiple dimensions encapsulated in the resilience concept provides an integrative approach to enhance system response capabilities.

The use of NPI increased community resilience during a pandemic outbreak with low, medium or high transsmisibility rate. The overall improvement was generated by the absorptive capacity, even though the restorative capacity worsened for the medium and and high transmissibility rate scenarios.

The AB simulation model provides significant results to estimate the healthy population as main feature to implement our community resilience metric. However, other outputs may also been relevant to analyse, such as: number of visits to the doctor, number of hours of school abscense, lost of work productivity, etc. Those outputs could be used to estimate the average performance metric or to develop new metrics.

From the metrics perspective the suggested relative resilience ratio based on the performance loss improves the comparison analysis among the baseline, NPI 1, and NPI 2 scenarios. The resilience differences of the optimal NPIs were estimated between 1.82 and 7.25 times higher than the levels of the other policies. In contrast, the margin was lower for the average performance with a 2 percent points difference and a maximum of 10 points for the time dependent metric.

Future research in this area will require to explore the following aspects: (i) Resilience metrics comparison, (ii) Community resilience factors evaluation, (iii) Pharmaceutical interventions (PIs) and dynamic strategies, and (iv) COVID-19 pandemic response. The evaluation of alternative metrics is suggested to include dimensions left out by the average performance approach and to capture resilience levels when multiple disruptions occur. There are factors that determine community resiliency levels, such as the compliance factors, the way the individuals interact, among others. The importance of PIs and NPIs in a dynamic scenario reassembles real settings and enables the potential improvement of the absorptive and restorative capacities simultaneously. There is a need to assess resilience for communities with different demographic distribution and alternative viruses. The AB model can be adapted to any specific epidemiological parameters in case that detailed analysis to any other influenza virus would be required.

Finally, the implementation of resilience models in the pandemics context seeks to grow the framework and metrics ecosystem in healthcare settings that have been using the conceptual value of the resilience property, but have not operationalised this idea.

Disclosure statement

No potential conflict of interest was reported by the author(s).