Development of a resource modelling tool to support decision makers in pandemic influenza preparedness: The AsiaFluCap Simulator

A simulation exercise is a way to train users of a response plan and to evaluate the effectiveness of the plan. The purpose of these exercises is to enhance knowledge and understanding of the plan, and identify gaps, weaknesses and opportunities for improvement in planning and operational capabilities [11]. Most of the exercises performed in Southeast Asia were tabletop exercises where public health officials and/or key staff with emergency management responsibilities examine, discuss and manage a hypothetical pandemic situation in a round-table manner. For example, Thailand had at least one tabletop exercise at the central level and in each province. Lao PDR conducted tabletop exercises, in collaboration with the World Health Organization (WHO), with representatives from the Ministry of Health and other key players. There were also a few cross border tabletop exercises, coordinated by WHO and the Mekong Basin Disease Surveillance Network (MBDS).

In Indonesia, the first full-scale simulation exercise (of its kind in the world) was carried out, in which multiple ministries and agencies across central, provincial and district levels of the government participated. The exercise in Indonesia covered key areas for outbreak control, like surveillance (e.g. for early detection of human-to-human transmission), pharmaceutical (e.g. antiviral drugs and vaccines) and non-pharmaceutical interventions (e.g. social distancing measures), medical response (case management and isolation of cases) and risk communication (to the population and media) [5].

Most of the conducted simulation exercises focused on early containment (the early detection and control of outbreaks), but not on pandemic preparedness in later phases. There has been much less investment into preparing health systems for pandemic mitigation [12]. Nevertheless, the mitigation phase includes the peak period where the highest case number of infections is reached, and thus contains the peak demand for resources.

All preparedness plans and simulation exercises need a scenario with a hypothetical outbreak situation [11]. The development of a scenario requires a strong evidence base that provides the foundation for planning and response exercises [13]. In recent years it has become common to use mathematical modelling for guiding the response to disease outbreaks and support policy decisions [3, 14‐22]. Policy makers and other stakeholders involved in pandemic preparedness require (input from) flexible models to estimate and compare the effect of different intervention strategies during an influenza pandemic. Moreover, such mathematical tools should be easily adaptable for other outbreak scenarios to address the yearly differences in virus transmission and virulence, lack of understanding of factors affecting the spread of influenza and shortage of data on the effectiveness of interventions [23].

In the past, several simulation tools have been made publicly available [24‐33]. These tools can be divided into applications that use a static modelling approach (e.g. FluSurge [26], FluAid [25] and StatFlu [24]) and those having a dynamic modelling structure (e.g. InfluSim [30], FluTE [28] and GLEaMviz [31]). All simulation tools can be used by policy makers to model a pandemic scenario in which an influenza virus is transmitted in a population and to estimate the burden of disease (e.g. the number of hospitalisations and deaths), conditional on given disease parameters. For example, StatFlu combines static modelling using historic influenza data with an easy-to-use interface and can be used to create insight in the effects of changing assumptions related to disease severity (like the attack rate and susceptibility of age groups) [24, 31]. With the GLEaMviz Simulator users can configure a disease compartment model and a scenario to simulate, by setting compartment-specific variables, transitions, environmental effects and other conditions. In addition, with the GLEaMviz Simulator results can be explored in dynamics maps and charts that describe quantitatively the geotemporal distribution of the disease [31].

However, most models only address a few types of healthcare resources. For instance, Flusurge 2.0, which was developed by the US Centers for Disease Control and Prevention (CDC), is a tool that compares needed hospital resources during pandemic influenza with existing hospital resources, but only includes three items (hospital beds, intensive care beds, and mechanical ventilators) [34]. InfluSim, a tool that is flexible regarding several disease parameter values and with which the impact of multiple interventions can be estimated, includes only the available number of health care workers and antiviral drugs [30]. Moreover, all freely available simulation tools do not estimate operational resource capacity gaps nor their impact on public health during influenza pandemics.

For policy makers and advisors it is essential to have an indication on the regional distribution of health care resources, potential resource gaps and their impact on public health during a pandemic. Limited supplies force decision makers to determine how and when to deploy the available resources and to prioritise between resources [20, 21]. Such decision-making requires models to provide estimates on surpluses and shortages of multiple outbreak related resources in geographical regions, during various pandemic scenarios. Understanding the dynamics of resources during a pandemic, including the consequences of resource gaps on public health and critical care services, can be of additional value to preparedness plans and simulation exercises such as tabletop exercises.

In summary, there is a need for a user-friendly and flexible resource modelling tool, which includes an easy-to-use option to display simulation results in maps, that could be used by policy makers and other stakeholders involved in pandemic preparedness for evidence based health care resource planning. Our objective was to develop such resource modelling tool, the AsiaFluCap Simulator, to support public health officials in understanding and preparing for surges in resource demand during future influenza pandemics. The development of this tool was part of a larger study, the AsiaFluCap project (http://​www.​cdprg.​org), which aimed to provide a strategic framework to evaluate health system capacity in Cambodia, Indonesia, Lao PDR, Taiwan, Thailand, and Vietnam in response to different phases of pandemic influenza. The AsiaFluCap project was conducted from May 2008 to April 2011 with funding support from the European Union and the Rockefeller Foundation. The following sections outline the structure, applicability, benefits and limitations of the AsiaFluCap Simulator.

The development of modelling tools involves a concession making process. It is a trade off between the usability and accessibility of the tool versus elements which may make the model more realistic (such as the use of individual based structures which can be used to estimate the geographical spread of infectious diseases [3, 35, 36]). As our main focus was to estimate the impact on health care resource capacity during influenza pandemics, we used a resource model in combination with a relatively simple epidemiological model (developed earlier in the AsiaFluCap project [12] and previously applied in resource gap analysis in countries in Southeast Asia [37, 38]). Our other objectives were to produce an easy-to-use and easily accessible tool, allowing the use by policy makers and other health care professionals without modelling expertise.

The structure of the AsiaFluCap Simulator tool was based on a framework described by Eriksson et al. 2007 [39]. The tool consists of (I) an epidemiological model combined with (II) a resource model containing 28 health care resources, (III) a graphical user interface, and (IV) a function to export simulation results to GIS software for displaying simulation results in maps (Figure 1). Part I and II of the AsiaFluCap Simulator require, next to information on available resource capacities and population sizes in areas, influenza and resource parameter values. We developed the tool in Microsoft Excel© (Microsoft Corporation, Redmond, WA) and the user interface was developed using the programming language VBA (Visual Basic for Applications) in Visual Basic. The tool is compatible with all Microsoft Excel© versions (except Microsoft Excel Viewer©).

In order to estimate the resource demand during an influenza pandemic, one requires estimates of the total number of influenza cases, especially concerning the peak number of hospitalised cases and cases with specific treatment and care, e.g. cases requiring mechanical ventilation or intensive care units (ICUs). For this, we used a deterministic SEIR (Susceptible-Exposed-Infectious-Removed) model described by differential equations tracking number of people in each compartment over time. Full details of the epidemiological mode can be found in Krumkamp et al. 2011 [12], an additional file provides a detailed summary of the equations and assumptions [see Additional file 1. Given that the primary aim of our tool was to demonstrate relative differences in resource shortages and surpluses during different pandemic scenarios, rather then to provide accurate quantitative predictions, we used a relatively simple model structure assuming homogenously mixing and without an age-structure.

In the SEIR model the population is divided into 17 compartments, with the infectious compartment being subdivided into three groups based on clinical severity (asymptomatic, mild and severe infections). All severe cases were at risk of death, and assumed to need hospitalisation and antiviral treatment (of which a certain proportion also required mechanical ventilation). All asymptomatically and mild infected patients were assumed to recover. Hospitalisation and treatment with antivirals were assumed to reduce the infectious period and the probability of death for severe cases. We also assumed that a proportion of severe cases would require mechanical ventilation, without which they would die.

The SEIR model differs from other existing transmission models [26, 30] as three key health care resources (hospital beds, mechanical ventilators, and antiviral drugs) were included as dynamic variables. Whether infected individuals received hospitalised care, ventilation or antiviral treatment depended on the availability of these resources. The inclusion of these resources as dynamic variables allows for quantitative estimates of the impact of resource shortages on morbidity and mortality.

The epidemiological model estimations were explicitly linked to a resource model. The resource model consists of 28 health care resources with accompanying parameters reflecting the use of these resources per influenza case per (hospitalised) day. We distinguished between depleting and occupied resources. Depleting resources are items that can only be used once, like personal protective equipment (e.g. masks N-95/N-99, surgical masks, face shields, surgical gloves, and surgical coverall gowns), medications (e.g. antibiotics, antivirals, and IV fluids), vaccines, and body bags. For depleting resources it is essential to have an indication whether the capacity is sufficient for the entire pandemic outbreak and, in case of shortages, to have an idea on the moment of resource depletion.

Occupied resources are those which are occupied by influenza cases for a certain period, but can be redeployed such as human resources (e.g. medical doctors, nurses, general practitioners, internal medicine specialists, other types of doctors, pharmacists, laboratory technicians, public health personnel, volunteers, and administrative staff) and equipment (e.g. hospital beds, ICU, ambulances, other transport vehicles, x-ray machines, mechanical ventilators). For occupied resources the peak of the pandemic is the most critical point for maintaining services, as the number of cases then reaches its maximum.

To estimate the required quantities of resources (for each type of resource) during a pandemic, the output of the SEIR model (namely the estimated number of hospitalised and ventilated cases at the pandemic peak and during the total pandemic period) is multiplied in the resource model with the resource demands per capita (e.g. the number of resources needed per hospitalised case per day). Next, in order to calculate resource “surpluses” or shortages for the simulated pandemic scenario, the estimated required quantities are subtracted from the resource capacity available for outbreak control (which is a percentage of the total number of resources present in a region). To estimate resource surpluses and gaps, the resource model requires information on the resource demand per case and the total number of resources available in the region(s). In terms of resources, data on the available capacity of hospital beds, antiviral drugs, and mechanical ventilators are required in order to run the simulator (while the inclusion of other resource data is optional). An additional document provides a detailed overview of the resource model structure and assumptions [see Additional file 4].

The included resources in the tool were selected through a health system resource needs analyses which was also part of the AsiaFluCap project (http://​www.​cdprg.​org). This resource characterisation process was carried out to inform policy makers in Southeast Asia on outbreak related resources and consisted of a systematic review and Delphi consensus (conducted in February 2009 by a panel of 24 public health experts from six countries in Southeast Asia and three countries from Europe).

In order to facilitate the use in practice we designed an interface in VBA for MS Excel© (Microsoft Corporation, Redmond, WA). The interface consists of six consecutive screens in which users are provided with technical background information about the AsiaFluCap Simulator and a menu in which different options for running simulations can be selected. An additional file displays a screenshot of the main menu of the interface [see Additional file 6]. The interface can be turned on by an extra button (‘Start AsiaFluCap’) that appears in the standard menu bar of MS Excel© (when the macro security settings of MS Excel© are set to ‘medium’ or ‘low’). The default values displayed in the interface are loaded from the individual sheets in the MS Excel© file.

Depending on whether users have provided data on the availability of resources in provinces or districts, users are provided with either a menu in which they need to fill in (manually) the available quantities in one region or a menu for selecting one or multiple provinces (automated menu). In case provinces or districts with corresponding data have been provided in the data input sheet, the regions are uploaded and displayed in the automated menu of the interface. Next, users can choose to simulate a mild, moderate or severe pandemic scenario with a specific R0. In three other provided menus it is possible to change all parameter values related to resource demands per case (including depletion rates in low and high pandemic activity periods, differences in resource demands between non-ventilated and ventilated hospitalised cases, and differences in case demands during day and night shifts) and change other resource parameter values (e.g. surge capacity percentages, mean number of work hours per week for healthcare personnel, duration of work shifts, and absenteeism percentages). All values provided in the interface are printed into the designated sheets in the tool for running simulations. The interface allows for resetting the default values from the literature (these default values are also loaded from the individual sheets). Additional files display screenshots of the menus for choosing pandemic scenarios, and changing and / or resetting depletion rates [see Additional files 7 and 8].

The interface contains help buttons for explaining functions and resource parameter values, and for displaying literature references of applied disease parameter values. The tool is also provided with an online audio-video guide in which the functions in the interface are explained (http://​www.​cdprg.​org/​asiaflucap-simulator.​php). Running simulations with the AsiaFluCap Simulator takes around 15 minutes for 80 districts or provinces, depending on the available computing power. The use of the AsiaFluCap Simulator has been piloted during the AsiaFluCap project meeting in Lao PDR, with 15 policy makers and other health care professionals from six countries in Southeast Asia (Cambodia, Indonesia, Lao PDR, Taiwan, Thailand, and Vietnam). A second pilot was carried out in the Netherlands, in order to assess validity and feasibility for countries outside South-East Asia, and it involved the seven regional public health experts of the National Institute for Public Health and the Environment (RIVM) in the Netherlands.

We developed a VBA function that saves all simulation results each time per province or district, in a separate sheet in the simulator. With another VBA function these simulation results are exported, all or a selection of the results (only results related to hospital beds, mechanical ventilators, oseltamivir, physicians and nurses), in a .dbf-file (dBase IV format). A shapefile is a popular vector data format for Geographical Information Systems (GIS) software, for storing geometric location (in so-called ‘.shp’ and ‘.shx’ files) and associated information (in a ‘dbf-file’: attribute format) [50]. The simulator requires a shapefile, in which the dbf-file is automatically replaced (e.g. old file is renamed) with the new dbf-file from the tool, creating a ready-to-use shapefile containing all simulation results per province or district.

With (open-source) GIS software and the newly created shapefile, simulation results can be displayed in maps that illustrate the distribution of health care resources across districts or provinces in a country quantitatively. With GIS software it is possible to create maps of resource demands, gaps or surpluses and the impact of shortages on public health (e.g. number of deaths due to resource gaps per 100,000 population) across regions during a pandemic scenario, similar to those presented in previous resource gap analysis studies [37, 38, 49].

The use of MS Excel© and the creation of the interface simplify the process of changing parameters in the model. The influenza scenarios in the model can be easily updated in the spreadsheets with values from future data on influenza or, if needed, changed to more extreme situations (e.g. a more mild of a more severe pandemic scenario). As data on the availability of 28 resources in regions or provinces might not be available to all policy makers, the AsiaFluCap Simulator also operates if only data on the availability of the three minimal needed resources (hospital beds, mechanical ventilators and antiviral drugs) is provided. Furthermore, other types of resources (with accompanying resource parameter values) can be added to the model (which will be automatically displayed in the interface) to also estimate the demand, shortage or surplus of these resources during a pandemic scenario.

The AsiaFluCap Simulator provides the simulation results in four different sheets in the tool. The model estimates the impact on public health, such as the number of hospitalised and ventilated cases and number of deaths, during a (selected) pandemic scenario. The tool reveals resource needs and, in case data on resource availability is provided, shortages and surpluses for the 28 included health care resources. It simulates the impact of limited resources (regarding hospital beds, ventilators and antivirals) on the outcome of the pandemic. Figure 2 displays graphs for the simulations made for the region in Lao PDR, as provided by the tool. A summary of the epidemiological estimations, assuming mild and severe pandemic scenarios in this region, is provided in Table 1. These key epidemiological estimations (e.g. total number of symptomatic cases stratified by disease subgroups, hospitalised subgroups and the cumulative number of deaths) and graphs with outbreak curves are provided in the ‘Epidemiology’ sheet of the tool. The next two sheets contain calculations on the impact on health care resource capacity, divided into depletion and occupied resources.

For resources that deplete, the tool calculates the required quantities, shortages or surpluses over the total pandemic period. These calculations are displayed in tables, bar charts and graphs, which provide a clear image of the resource output. In addition to the graphs that contain outbreak curves, the figures contain (for a number of resources) bars that provide an indication on the moment of depletion. For occupied resources, the AsiaFluCap Simulator displays the estimated resource needs, gaps or shortages, per pandemic peak day, in a table. The tables indicate resource gaps in red and surpluses in black. For medical doctors and nurses the tool provides insight on the moments of deficiencies in the number of available staff during the pandemic period.

The tool provides resource estimations both for a scenario for which sufficient available resources were assumed and a scenario with actual available resources. The difference between these two estimations is clearly illustrated in the simulations made for the region in Lao PDR. During a mild baseline scenario, hospital bed capacity in Lao PDR was (almost) sufficient enough to accommodate all severe cases (although there was a large shortage of ventilators). During a severe baseline scenario, this shortage in hospital beds limited the number of hospitalisations, leading to severe outpatients (Figure 2).

The tool can be used to run different scenarios for the same region to compare the impact on resource needs during different pandemic outbreaks, but also to investigate the effect of changes in resource demands on disease burden. A summary of the sensitivity analysis performed with the tool for the region in Lao PDR is presented in an additional file [see Additional file 9]. Estimated resource needs are sensitive to the severity of the modelled pandemic scenario in terms of transmissibility (basic reproduction number) and the proportion of cases requiring hospitalisation. Furthermore, the peak numbers of hospitalisations, as well as the time of the peak, are sensitive to the chosen basic reproduction number.

The AsiaFluCap Simulator runs a pandemic scenario for each province or district separately, taking into account the population size and the actual available resource capacity in regions. When the tool is used to run simulations for more than one province, then simulation results are provided per province in one overview table which is displayed in the fourth sheet of the tool. With the use of an added button in the menu bar of MS Excel©, the simulation results can be exported (partially or in total) and included in a shapefile (for displaying the simulator estimations per province or district in maps). Figure 3 displays the estimated gaps and surpluses in hospital beds, ventilators and oseltamivir, for each province in Lao PDR, during a mild, moderate and severe pandemic scenario. With these maps, provinces with (large) shortages or (abundant) surpluses in resources can be identified at a single glance, which could act as a trigger for policy makers for remobilisation or re-allocation of resources between (neighbouring) provinces. Although most provinces display insufficient resources for responding to the three modelled baseline scenarios, we did not demonstrated the effect of changes in resource parameter values (e.g. increased surge capacity) nor included interventions like social distancing measures. The AsiaFluCap Simulator can be easily used to compare other scenarios, next to the three baseline scenarios displayed in Figure 3.