A risk‐based mutual insurance premium framework for establishing indices of vulnerability to the intentional introduction of transboundary animal diseases

Abstract

Biological agents as weapons of agro‐crime or agro‐terrorism pose threats to peace and economic stability. Such agents pre‐exist worldwide as hazards, adversely affecting animal health, as well as imposing substantial burdens on many nations. Few studies have quantified the global risks and vulnerabilities of countries and regions to potential terrorist or criminal operations targeting animal health. We present here a risk‐based mutual insurance premium framework for animal health outcomes built upon the World Organisation for Animal Health (WOAH) quantitative risk assessment paradigm. Our objective was to generate dimensionless and relative domain indices related to release and exposure for several biological factors, as well as to assess the preparedness and response ability of each country. We also considered disease‐specific measures relating to pathogens, targeted animal populations, the ongoing disease situation, within‐ and among‐country peace or conflict, disease‐specific control measures, and the availability of technical tools and personnel for successful disease management. National economic, political, and research and development competencies were used to assess each WOAH Member's potential for resilience. We formulated indices of vulnerability for 25 WOAH Members selected from five worldwide regions; initially, against four transboundary infectious animal diseases that target diverse animal species. We developed these indices using variables obtained from public databases arising from multiple intergovernmental organizations. Subsequently, we compared the relative vulnerability indices among countries for each given disease using three different index building methods: arithmetic mean, distance matrix, and principal component analysis (PCA). The PCA‐based approach provided the greatest ability to discriminate among the components and among countries and regions. Due to its transparency and reliance on publicly available datasets, the risk premium framework proposed herein may readily be adjusted by policymakers and agencies and utilized to improve risk management strategies against agro‐crime or agro‐terror events, as well as for unintentional disease introductions.

1. INTRODUCTION

Multiple biological agents have been considered as potential threats that could be used intentionally against livestock populations through acts of terror, warfare, or crime (Blancou & Pearson, 2003; Dudley & Woodford, 2002; GPWG, 2018). Unfortunately, most of the world faces unknown levels of ongoing risk as a result of not‐fully characterized, yet widespread, vulnerabilities and a diverse range of naturally occurring infectious disease agents (GPWG, 2018; Moore et al., 2016; Parnell et al., 2010; Rios & Insua, 2012). The potential outcomes of intentional pathogen release include both direct and indirect effects on animal health, public health, and national and regional economic outcomes. Because of the scale and speed of international trade and travel, infectious disease agents can spread quickly, and any agro‐crime or agro‐terrorism event directed at a single location could quickly become an international crisis involving many nations and regions (Elbers & Knutsson, 2013; GPWG, 2018; Jansen et al., 2014; Knobler et al., 2002).

To respond with countermeasures to an agro‐crime or agro‐terror event targeting domestic animal populations, multi‐sectoral stakeholders must develop and maintain the capacity to plan, prepare for, and respond to the widest possible range of strategies that criminal or terrorist groups may use (GPWG, 2018; Parnell et al., 2010; Rios & Insua, 2012). There is currently a marked disparity in disease emergency management capabilities among global regions (GPWG, 2018; McDougle et al., 2020). To effectively allocate resources to foster a sustainable level of capacity that promises the highest return on any given investment, the risks and vulnerabilities must be identified and understood.

In recent years, several indices and models have been developed to assess international investment and trade risks (WB, 2020) and to forecast potential human infectious disease hotspots (Moore et al., 2016; Oppenheim et al., 2019). Studies have narrowly focused on assessing the scientific and response capabilities against biological threats to human health via terrorism, including assessing public health infrastructure capacity and reviewing the scientific and policy tools for countering bioterrorism (Knobler et al., 2002; O'Brien et al., 2021). Other reports have focused on the specifics of the risk management process (CDSE, 2020; Rios & Insua, 2012). Some efforts have concentrated on standardizing and establishing guidelines to facilitate the risk and vulnerability assessment process (OIE, 2004). Among a wide range of potential indicators of infectious disease risk face by sovereign nations, common attributes such as poor governance and corruption (Collignon et al., 2015; Moore et al., 2016; WB, 2020) and susceptibility to intelligent adversaries—that is, those motivated either by ill‐will or ill‐gained profits—emerge (Parnell et al., 2010; Rios & Insua, 2012).

Across the spectrum of infectious disease research, risk analysis, policy development, and other endeavours intended to assess and mitigate risks from biological threats, it is clear that there are gaps in our understanding of global and regional response characteristics and capabilities, as well as the risks associated with failures in a truly international and collaborative response.

To address these gaps, we aimed to create a fully transparent risk‐based framework focused initially on four exemplar infectious diseases (African swine fever [ASF], foot and mouth disease [FMD], brucellosis (BA: limited to Brucella abortus), and highly pathogenic avian influenza (HPAI), which were chosen based on (1) specificity/non‐specificity of animal host, (2) viral versus bacterial agent, (3) non‐zoonotic/zoonotic agent, and (4) potential for intentional use as a bioweapon as highlighted by the Australia Group List (AG, 2020). Since World Organisation for Animal Health (WOAH) is the only international body in charge of publicly analyzing worldwide animal health status due to its mandate and structure, which allows them to seek, collect, and disclose global animal health information from 182 WOAH Member Countries, to achieve our first goal, we assumed WOAH as the ‘‘insurer’’ and WOAH Members as the ‘‘insured’’ parties. We established this process as a risk‐based mutual insurance premium framework and constructed our initial assessments using the data of 25 WOAH Members from five different global regions.

The mutual insurance ‘premiums’ were expected to change dynamically based on risk profiles of other nations that may later be added to the insurance pool, based on subsequent annual changes in each country's risk profiles, and depending on similar developments in other mutually insured countries. Our specific aim was to calculate the zero‐summed insurance premiums (also called vulnerability indices) using variables extracted from publicly available databases that provided information on biological determinants of each disease, disease control measures that undertaken by each country, as well as the availability of technical tools and personnel, and considering those socio‐economic parameters (e.g., resilience) that are seen as necessary for disease control and management among these countries (OIE, 2014).

2. MATERIALS AND METHODS

2.1. Conceptual framework

The initial risk‐based mutual insurance premium framework was established based on (1) biological factors, (2) control measures, (3) availability of technical tools, and (4) socio‐economic parameters which are the parameters highlighted by the Guidelines for Animal Disease Control, published by the WOAH (OIE, 2014). According to this guideline, biology and epidemiology of pathogens, vulnerable host and vector population distributions and densities, ongoing infectious disease status and dynamics, applied disease detection, control and prevention measures, as well as the availability of technical tools and personnel are key elements that need to be accounted for by each OIE Member in order to optimize a successful disease control program including socio‐economic considerations that can reflect good governance, suitability of its animal health regulatory framework, level of public engagement, and availability of adequate resources, the totality of which reflects the disease prevention, control, and response capacity of the country (Table S1). In parallel to this guideline, we considered the measure of disease response success to be the degree to which a WOAH Member would be capable of bringing its disease status back to, or better than, the equivalent of where it existed prior to either an agro‐crime or agro‐terror event, or an unintentional or naturally occurring disease introduction, that is, through its preparedness and response capabilities and its mobilized efforts. Our risk‐based mutual insurance premium framework is further built upon the WOAH quantitative risk assessment framework (i.e., hazard identification—release assessment—exposure assessment—consequence assessment—risk estimation) and incorporates multivariate data from reputable international sources used to populate its risk index components (OIE, 2010).

In the end, we structured our risk‐premium framework on three main facets of risk‐based mutual insurance premium setting which are analogous to those utilized in establishing premiums for personal and property damage and third‐party liability insurance for motor vehicles (e.g., based on risks associated with the driver, the vehicle, and the location of storage and operation). These were established as follows: (1) preparedness—comprising resources and systems put in place to detect disease agent incursions and accurately establish disease status, as well as to prepare human and animal populations for incursion (e.g., through vaccination and personnel training), (2) response—comprising resources and systems needed to be able to mitigate against an incursion and take actions to reduce the further spread and to recover; also, to include aspects of disease surveillance and vaccination in determining the extent of, and limiting, the disease spread, respectively, and establishing disease status for recovery purposes, and (3) resilience—comprising the systemic aspects of governance and generalized resource availability to prepare and respond to infectious disease emergencies. External and internal conflict indices, animal populations at risk, importance of the commodity product for domestic consumption and export, and pre‐existing presence or absence of the disease agent represent features that span the three realms and also impact both the mutual insurance premium and the country's insurability (i.e., in some contexts, declared acts of aggression—like war—would be uninsurable, and pre‐existing conditions such as disease endemicity would preclude insuring against their introduction).

In this study, we combined parameters that are needed for successful disease control parameters as suggested by OIE Guidelines, developed within the OIE risk assessment framework, by assuming the agro‐crime and agro‐terror threats via four disease agents already exist; therefore, the hazard was assumed to be pre‐defined—though hypothetical in the context of an intentional threat. The potential for disease agent release was primarily assumed to be based on the motivations of an intelligent adversary to target a country where vulnerable agricultural species populations are essential for the public (e.g., domestic meat production and consumption) or its global trade balance (e.g., live animal or meat product export value). This motivation was also assumed to be affected by the ongoing disease status in that country. For instance, if a disease was previously eradicated or else under control in a country, the expected motivation for an intentional release would be higher than for a country in which the disease was endemic. In addition, release was also assumed to be more likely to occur when a country was currently engaged in either domestic or international conflict.

Exposure potential was largely assumed to be dependent on the vulnerable animal population size and its distribution across the agricultural lands of each country (i.e., animal density). Countries with higher numbers of susceptible hosts, and with more densely concentrated vulnerable populations, were assumed to have greater potential for disease exposure.

Scope and scale of the consequences of disease agent exposure were assumed to depend on the preparedness and response ability of each OIE Member. Disease preparedness and response abilities (abbreviated as: PrepRes) of the countries were assumed to be heavily dependent on both trained and qualified personnel and laboratory capacity within the veterinary domain, and on a competent authority established to (1) implement disease detection, (2) oversee existing disease prevention and control measures, (3) cover the spectrum of animal species that are controlled for each disease, and (4) provide for the availability of the technical tools required for such action. Even though disease preparedness and response are two different parameters, they are almost universally and positively related given that rapid detection and identification of an animal disease, along with characterization of herds and flocks, reveal a rapid response potential through the application of specific disease control parameters that are already established, as well as the pre‐existing ability to track and document the recovery through pre‐existing monitoring and surveillance networks.

Preparedness and response abilities of countries were also assumed to depend on their resilience capacity. Resilience capacity in the international development literature relates to the economy, governance practices, social safety, and security, as well as levels of both domestic conflict and international conflict. Resilience was also assumed to be related to international goodwill with other countries for potential support, literacy rates, and the allocation of governmental funds towards research and technology development all of which help establish the ability to prepare and plan for, absorb, recover from, and more successfully adapt to adverse events (NRC, 2012).

In summary, the framework of the risk estimation developed in this study is a dimensionless vulnerability index calculated for each participating OIE Member (policy holder) and is based on the potential magnitude of each of disease agent (hazard) release, exposure, and consequence (i.e., preparedness and response) under each given scenario (Figure 1).

FIGURE 1.

Proposed framework and related factors considered for inclusion into each subdomain for risk‐based insurance premium framework (see Table S1 for original OIE construct)

2.2. Country selection

There were too many potential WOAH Member States to be accurately measured for the initial data quality assessments of this study. Each country combines unique political, economic, geographical, and other associated technical attributes contributing asymmetric risk to the WOAH Member. Therefore, a set of five representative Members from each of five hybrid WOAH/FAO (Food and Agriculture Organization of the United Nations) regions (Americas, Europe and Central Asia, Middle East, Africa, and Asia) was selected to initially assess their attributes and vulnerabilities. These countries were chosen based on the completeness of their data in published resources such as the WOAH Performance of Veterinary Services Pathway Reports (OIE, 2017), representation of their features as components of the full spectrum of hybrid WOAH/FAO regions, and being representative of a full range of gross national income levels (i.e., low, lower‐middle, upper‐middle and high) as reported by the World Bank (WB) database of 2019 (WB, 2019). The actual names of the countries used in this practical assessment have been masked to avoid highlighting national vulnerabilities to agro‐crime or agro‐terrorism; instead, they are represented as numerically coded values (1–25). Regional distribution of these 25 countries, along with their corresponding income classifications, are presented in Figure 2.

FIGURE 2.

Regional distribution and gross national income (Organisation for Economic Co‐operation and Development [OECD]) classification of the 25 countries included in this study.

For visualization purposes, the principal component analysis (PCA) scores obtained were further scaled between 0 (lowest) and 1 (highest) for each domain, and shown as PCA (scaled) at the bottom left.

2.3. Variable selection, data curation, and processing

2.3.1. Hazard identification

In this study, we chose to assess risks related to the introduction of African swine fever virus (ASFv), foot‐and‐mouth disease virus (FMDv), highly pathogenic avian influenza virus (HPAIv), and Brucella (specifically; brucellosis caused by B. abortus) as the infectious disease agent hazards. These initial four pathogens were chosen for their potential impacts on differing subsets of the domestic animal populations, their differences in transmission capabilities (e.g., speed, vectors), the range of disease manifestations, including as potential zoonoses (e.g., B. abortus and HPAI) as well as their potential impacts on agricultural production commensurate with livestock host importance among the various WOAH Members.

In this assessment, raw data collected at the country‐level from a total of 25 countries, and using online databases such as the World Animal Health Information System of WOAH (WAHIS) and FAO (FAOSTAT) were utilized. In addition to these animal‐ and disease‐related databases, factors relating to the resilience capacity of each country were obtained from reports published by agencies and organizations such as the Institute for Economics and Peace (IEP, 2019), the RAND Corporation (Moore et al., 2016), and the World Bank (WB, 2018c).

For this study, data and interpretations were restricted to five host types (cattle, goats, sheep, pigs, and poultry [chickens]) among the broader list of hosts (including wildlife) that are known to be vulnerable to the four disease agents. Pigs were included for ASFv‐related analyses, cattle were included for B. abortus‐related analyses, cattle, sheep, goats, and pigs were included for FMDv‐related analyses, and chickens were included for HPAIv‐related analyses. Wildlife and vector‐related factors (where applicable) were excluded from the analysis due to limited data availability across all countries. In addition, other vulnerable hosts such as domestic cloven‐hoofed animals (e.g., buffaloes, camels) for FMDv, domestic and wild birds (e.g., ducks, turkeys, geese) for HPAIv, and other domestic animals susceptible to B. abortus (e.g., buffaloes, deer) were excluded due to the many great missing data points in both WAHIS and FAOSTAT databases for these host species.

2.3.2. Release domain components

Meat production quantity in metric tonnes (FAOSTAT, 2017b) and live animal export value (in US $1000) for domestic livestock species (cattle, goats, poultry [chicken], sheep, and pig) were obtained from the FAOSTAT database of 2017 (FAOSTAT, 2017c) in order to reflect the economic and public importance of the animal species to each country. These numbers were later subtotalled across relevant species representing the set of vulnerable populations for each of the four exemplar disease agents.

Disease prevalence and dynamics scores were calculated based on a disease status timeline (i.e., from a total of 11 data points) reported biannually in the WOAH's WAHIS database (WAHIS, 2014–2019), specifically from the beginning of 2014 until mid‐2019. Regarding the disease status for any given reporting period, WOAH Members were coded as zero (0) if the disease (or infective agent) was confirmed to be present, one (1) if the disease was suspected but not confirmed, two (2) if the disease was present but limited to one or more zones, and three (3) if the disease was absent (or, never reported). After ranking each of the 11 timepoints, each observation was multiplied by a weight scaled from 1 to 11 (i.e., from the oldest to the newest datum) to reflect a scaled composite index of the disease history/situation during the entire 5.5‐year period. Higher values reflected more optimal disease status, weighted to emphasize the more current data.

To represent levels of international and domestic conflict, the Ongoing International and Domestic Conflict Domain Score of each country was obtained from the Global Peace Index (GPI) Report of 2019, published by the Institute for Economics and Peace (IEP) (IEP, 2019). These scores purport to reflect the ongoing domestic and international conflict situation for each individual country and are based on parameters such as the number and duration of internal and external conflicts, as well as the status of relations with neighbouring countries (Table S2).

2.3.3. Exposure domain components

Total raw numbers of cattle, sheep, goats, pigs, and chickens for each of the 25 countries were obtained from the FAOSTAT database of 2017 (FAOSTAT, 2017a). Vulnerable animal population numbers for each disease agent were obtained by subtotalling the number of animals from each species that are known to be susceptible to each disease agent. Animal density (number per km²) and agricultural land size (km²) of each country were obtained from the World Bank database of 2016 (WB, 2016); later, vulnerable animal population densities were obtained by dividing the subtotalled number of vulnerable animals in each country by the agricultural land size (km²) to represent the potential for disease agent exposure in a hypothetical susceptible population.

2.3.4. Consequence: Preparedness and response domain components

Estimation of the animal health personnel capacity and laboratory capability of each OIE Member was determined using the total number of veterinarians and para‐professionals obtained from the WAHIS database (2017) for each country (WAHIS, 2017). The number of national laboratories established for each specific disease was obtained using the WAHIS database of 2019 (WAHIS, 2019).

Disease control and prevention parameters listed in the WOAH's OIE Notification Procedure, Terrestrial Animals (OIE, 2016) were included—as available—for each disease agent as follows: (1) disease notification, (2) precautions at the border, (3) disease monitoring, (4) disease screening, (5) general surveillance, (6) targeted surveillance, (7) movement controls inside the country, (8) stamping out, (9) zoning, (10) slaughter, (11) ante‐post‐mortem inspections, (11) control of vectors, (12) selective killing and disposal, (13) inactivation of pathogenic products, (14) compartmentalization, (15) vaccination prohibited, (16) official vaccination, (17) control of wildlife reservoirs, and (18) vector control. Data related to these parameters were obtained from the WOAH‐WAHIS database of 2017 (WAHIS, 2018). In addition to these disease control and prevention parameters, disease introduction simulation exercise occurrences (available only for FMD and HPAI) for each of the 25 countries were obtained from the WAHIS database between the years 2017 and 2019 (WAHIS, 2017–2019). As a result, 19 disease control parameters—including the disease stimulation exercises—were counted in the final analysis as the raw sum total of control and preventive measures as applied by each country. These parameters were binary coded to reflect either presence (1) or absence (0), while the final score of each country for each disease was recorded as the total number of disease control and prevention measures available and applied.

Domestic animal species subject to control measures for each disease were as listed in WOAH's OIE Disease Notification Guideline: (1) ASFv: pig, (2) Brucella (B. abortus): cattle, buffaloes, pigs, goats, sheep, camels (Camelidae), deer (Cervidae), and rabbits, (3) FMDv: cattle, buffaloes, pigs, goats, sheep, and camels, (4) HPAIv: birds (OIE, 2016) were obtained from the WAHIS database of 2017, and the total number of species subject to control under WOAH Guidelines for each disease was carried to the next step in the analysis.

Specific disease control and prevention parameters, and the number of animal species that were subject to control for each disease agent, were only obtained for domestic animals (i.e., wild and feral animal populations were excluded). The list of the disease control and prevention parameters available for each disease, along with the domestic species subject to control measures, is shown in Figure S1.

2.3.5. Consequence: Resilience domain components

Worldwide governance indicators (WGI) were obtained from the World Bank database of 2018 (Kaufmann et al., 2011; WB, 2018c). Indices that were included in the analysis were as follows: (1) control of corruption, (2) government effectiveness, (3) regulatory quality, (4) the rule of law, and (5) voice and accountability. Among these indices, control of corruption (CO) encompassed perceptions regarding how public power is exercised for private gain, including both petty and grander forms of corruption, as well as ‘self‐capture’ of State resources by elites and private interests. Government effectiveness accounted for the quality of public and civil services along with their independence from politics, the quality of policymaking and their power of implementation, and the credibility of the government's responsibilities. Regulatory quality accounted for the success of the government to rule and implement reliable policies and regulations, especially those which promote and permit private sector development. Rule‐of‐law accounted for confidence in the quality of contract enforcement, property rights, the police, and the courts, as well as the likelihood of violence and crime. Finally, voice and accountability accounted for perceptions of citizen participation in elections, freedom of expression, freedom of association, and freedom of the press. A detailed list of attributed sources for each WGI is presented in Table S2.

Ongoing domestic and international conflict scores (ODAIC), obtained from the GPI 2019 Report (IEP, 2019) and included in the release domain, were scaled there in reverse with increased measures of ongoing domestic and international peace expected to be associated with a decreased risk for the intentional release of any one of the four disease agents. From the same report, the Societal Safety and Security Domain Score (SSAS) values were included in the resilience domain in order to account for the domestic security situation related to crime rates, violent demonstrations, and terrorist activities observed within each country (Table S2).

In order to account for the strength and robustness of the monetary economy, the Economic Domain score (ED), which reflects parameters such as the GDP per capita, economic growth rate, human development index, and national poverty ratio of each country, was obtained from the Infectious Disease Vulnerability Index previously published by the RAND Corporation in 2016 (Table S2) (Moore et al., 2016). Along with the ED scores, Political‐International Domain Scores (PI) obtained from the same report were included in the analysis to account for international political factors that could play a role in determining international aid and support, that is, monies or in‐kind contributions that could be received by a country as cash funding, expertise, personnel, and materials (e.g., diagnostic tools, medicines, vaccines) and each of which could contribute to the resilience of an OIE Member during an animal health‐related crisis.

The most up‐to‐date literacy rates for adults (% literate among persons aged 15 and above) were obtained from the World Bank database (WB, 2018a). As an indicator of development in science and engineering, the total number of scientific and technical journal articles published in physics, biology, chemistry, mathematics, clinical medicine, biomedical research, engineering and technology, and the earth and space science fields were derived from the World Bank database (WB, 2018b); further, these were adjusted to reflect the human population size as articles published per capita. Human population numbers for each country were obtained from the FAOSTAT database of 2018 (FAOSTAT, 2018).

2.4. Handling missing data

A single imputation method was used to replace missing observations with either (1) the last observation carried forward when data were available in the previous year of the dataset or (2) absent any historical data, mean substitution was employed by taking the average value of the other countries in the same region and of similar income level. Missing data resulting from the lack of direct importance of an animal species, such as one with little or no economic (e.g., export) or public (e.g., meat consumption) value, or simply not existing in a given country, were also substituted with zero. This included, for example, data on swine populations in predominantly Muslim countries. Finally, missing data points in the published WAHIS database were oftentimes replaced with information obtained via direct communication with WOAH‐WAHIS personnel who may have had access to as yet unpublished data.

2.5. Data processing

Before processing the data, two additional hypothetical countries were generated representing the minimum (refMin) and maximum (refMax) empirical reference scores for each domain and component variable using the corresponding database and year of interest. This was performed to avoid interpretive issues of rescaling within each domain if and when new WOAH Members are added to the analysis. Observations within each variable were recorded as zero for refMin, whereas the maximum available value (from each dataset of interest, whether or not it was one of the 25 countries in the analysis) reported for any country was recorded for refMax. Later, the observations within each variable were scaled between 0 (lowest score, also the refMin) and 1 (highest score, also the refMax) by using an interval‐based normalization feature option found in the Orange v.3.26.0 data‐mining toolbox (Demšar et al., 2013).

Individual indices obtained for each country using WGI (WB, 2018d), GPI (IEP, 2019), or RAND (Moore et al., 2016) reports also were rescaled from 0 (lowest score) to 1 (highest score) based on their reported or theoretical scoring ranges. The only exception was the ongoing domestic and international conflict score (ODAIC) and social safety and security (SSAS) variables in GPI, which were originally reversed for interpretive purposes. Therefore, these two variables were reverse‐scaled to reflect 1 (highest score) and 0 (lowest score) for the resilience domain while the original scores for ODAIC were preserved for scaling in the release domain. The list of variables that were included in each domain, and their database source and reporting year, are provided in Table 1.

TABLE 1.

Summary of variables and source datasets used in the study

Domain	Variable	Description	Year	Source
Release	meat	Meat production quantity (tonnes)	2017	FAOSTAT
	export	Live animal export quantity ($)	2017	FAOSTAT
	5y_disease	Disease dynamic score of last 5.5 years	2014–2019	WAHIS
	odaic	Ongoing international and domestic conflict	2019	IEP
Exposure	num	Total number of vulnerable population (head)	2017	FAOSTAT
	den	Vulnerable animal population density (animal head/km2 agricultural land)*	2017	FAOSTAT
PrepRes	Personnel	Personnel capacity (total number of veterinarians and paraprofessional vets) per vulnerable animal**	2017	WAHIS
	sp_cont	Total number of vulnerable species for which disease control measures exists	2018	WAHIS
	tot_cont	Total number of disease control and prevention parameters exist for each disease	2017	WAHIS
	nl	Total number of disease‐specific national laboratories	2019	WAHIS
Resilience	coc	Control of corruption	2018	WB
	rol	Rule of law	2018	WB
	rq	Regulatory quality	2018	WB
	ge	Government effectiveness	2018	WB
	vaa	Voice and accountability	2018	WB
	odaic_r	Reverse ongoing international and domestic conflict	2019	IEP
	ssas_r	Reverse societal safety and security	2019	IEP
	e	Economic	2016	RAND
	pi	Political‐international	2016	RAND
	article	Research articles per person	2018	WB
	literacy	Literacy rate	2018	WB

^*Calculated from the vulnerable animal population, 2017, FAOSTAT and agricultural land size (km²), 2016, World Bank.

^**Calculated from the personnel capacity (including the total number of veterinarians and paraprofessionals), 2017, WOAH‐WAHIS and vulnerable animal population, 2017, FAOSTAT.

2.6. Index building

Correlations observed among the variables within each domain and for each disease were examined using Stata v.16.1 prior to performing the aggregate analyses. Individual domain indices were obtained by arithmetic mean, distance matrix, and principal component analysis (PCA) using the scaled (0–1) variables for each disease.

The arithmetic means for the variables within each domain and disease combination were calculated. Euclidian distances from each of the 25 countries to the refMin (0) values were calculated using Orange v.3.26.0 (Demšar et al., 2013). PCAs were performed for each domain and within each disease. Afterwards, the totality of those components that explained either ≥90% of the variance or else individually had eigenvalues ≥0.5 were rotated using varimax rotation. Following the rotation, the final PCA individual WOAH Member values (per domain) were obtained by multiplying the rotated component scores with their corresponding proportion of explained variance. The final vulnerability indices resulting from the risk‐based mutual insurance premium framework for each disease and for each of the 25 countries were calculated by subtracting the PrepRes and Resilience domain indices from the sum of the Release and Exposure indices as follows Vulnerability = (Release + Exposure) − (PrepRes + Resilience). These were graphically overlaid upon stacked and coloured bars representing the indices for each of the individual domains.

To compare the indices obtained from the three different methods for each of four domains and each of four diseases (i.e., arithmetic mean, distance matrix, and PCA), pairwise Pearson correlations were performed. Countries (n = 25) were further ordered from highest to lowest index value by adjusting the highest index to the one within each domain and disease; meanwhile, ties were assigned the highest score among the set. Ranks obtained from the three different methods were compared using Spearman's rank correlation test. The choice of three methods was based on the ability of the method to discriminate among the domain scores across the 25 countries, balanced against explaining the most variance.

3. RESULTS

Out of 3750 expected datapoints, 3384 (representing 40 variables) were found, while 366 (9.7%) observations were not found for a given source, or year of interest, among the 25 countries (see Figure S2). A subset of the missing observations (n = 162) was replaced with the most‐up‐to date information available from the previous years (i.e., since 2013), assuming that the status observed at the last observation was carried forward; this approach was mainly applied to the ongoing disease status of the countries (disease dynamic score of last 5.5 years [5y_disease]) and to their live animal export quantities (export). The remaining 186 observations were substituted with a mean value, coded most often as zero due to the presumed lack of importance of that animal species for the country of interest; further, variables such as live animal exports and meat production for these same countries therefore often were not reported in the databases. The missing information relating to number of national labs and personnel was replaced with the information obtained via personnel communication with WOAH‐WAHIS staff (n = 18). Following data curation and imputation, all 3750 expected datapoints were carried forward to the next set of calculations and for construction of the final vulnerability indices.

As a result, 21 variables for each of the four diseases were generated. Pairwise correlations of the variables within each domain showed that the variables used for exposure and resilience domains were highly correlated compared to those variables in the release and PrepRes domains. Details on all pairwise correlations, for each variable and by disease, are provided in Table S3. Following the scaling of raw variables between 0 (refMin) and 1(refMax), the mean, standard deviation, and minimum and maximum observation values of countries within each domain were calculated and are shown in Table 2, while the related dataset is provided in Supporting Dataset 1.

TABLE 2.

Mean, standard deviation, and minimum and maximum values of scaled variables included in the study for 25 countries for each of four diseases: (1) ASF—African swine fever, (2) brucellosis (B. abortus), (3) FMD—foot and mouth disease, and (4) HPAI—highly pathogenic avian influenza

			ASF				B. abortus				FMD				HPAI
Domain	Variable	Obs	Mean	SD	Min	Max	Mean	SD	Min	Max	Mean	SD	Min	Max	Mean	SD	Min	Max
Release	meat	25	0.01	0.02	0.00	0.07	0.05	0.16	0.00	0.80	0.02	0.04	0.00	0.20	0.06	0.14	0.00	0.71
	export	25	0.01	0.06	0.00	0.29	0.03	0.10	0.00	0.50	0.03	0.12	0.00	0.60	0.01	0.02	0.00	0.11
	5y_disease	25	0.93	0.21	0.00	1.00	0.41	0.45	0.00	1.00	0.67	0.40	0.00	1.00	0.89	0.18	0.50	1.00
	odaic	25	0.28	0.22	0.00	0.71	0.28	0.22	0.00	0.71	0.28	0.22	0.00	0.71	0.28	0.22	0.00	0.71
Exposure	num	25	0.01	0.02	0.00	0.09	0.09	0.25	0.00	1.00	0.06	0.12	0.00	0.48	0.03	0.06	0.00	0.27
	den	25	0.03	0.07	0.00	0.28	0.16	0.13	0.01	0.45	0.09	0.16	0.00	0.83	0.00	0.01	0.00	0.02
Preparedness and Response	personnel	25	0.29	0.43	0.00	1.00	0.00	0.01	0.00	0.02	0.10	0.13	0.01	0.60	0.09	0.20	0.00	1.00
	sp_cont	25	0.80	0.41	0.00	1.00	0.68	0.30	0.00	1.00	0.94	0.14	0.50	1.00	0.96	0.20	0.00	1.00
	total_cont	25	0.22	0.16	0.00	0.63	0.36	0.19	0.00	0.71	0.38	0.15	0.06	0.67	0.38	0.16	0.00	0.61
	nl	25	0.06	0.08	0.00	0.25	0.12	0.09	0.00	0.38	0.12	0.12	0.00	0.57	0.13	0.07	0.00	0.25
Resilience	coc	25	0.45	0.19	0.17	0.87	0.45	0.19	0.17	0.87	0.45	0.19	0.17	0.87	0.45	0.19	0.17	0.87
	rol	25	0.46	0.19	0.09	0.85	0.46	0.19	0.09	0.85	0.46	0.19	0.09	0.85	0.46	0.19	0.09	0.85
	rq	25	0.47	0.18	0.16	0.86	0.47	0.18	0.16	0.86	0.47	0.18	0.16	0.86	0.47	0.18	0.16	0.86
	ge	25	0.46	0.21	0.05	0.84	0.46	0.21	0.05	0.84	0.46	0.21	0.05	0.84	0.46	0.21	0.05	0.84
	vaa	25	0.44	0.19	0.11	0.80	0.44	0.19	0.11	0.80	0.44	0.19	0.11	0.80	0.44	0.19	0.11	0.80
	odaic_r	25	0.72	0.22	0.29	1.00	0.72	0.22	0.29	1.00	0.72	0.22	0.29	1.00	0.72	0.22	0.29	1.00
	ssas_r	25	0.60	0.17	0.26	0.93	0.60	0.17	0.26	0.93	0.60	0.17	0.26	0.93	0.60	0.17	0.26	0.93
	e	25	0.45	0.21	0.08	1.00	0.45	0.21	0.08	1.00	0.45	0.21	0.08	1.00	0.45	0.21	0.08	1.00
	pi	25	0.50	0.15	0.26	0.92	0.50	0.15	0.26	0.92	0.50	0.15	0.26	0.92	0.50	0.15	0.26	0.92
	article	25	0.10	0.16	0.00	0.64	0.10	0.16	0.00	0.64	0.10	0.16	0.00	0.64	0.10	0.16	0.00	0.64
	literacy	25	0.85	0.13	0.59	1.00	0.85	0.13	0.59	1.00	0.85	0.13	0.59	1.00	0.85	0.13	0.59	1.00

Note: Scaled data are generated using minRef (lowest score = 0) and maxRef (highest score = 1) values for each variable and disease. Meat (meat production quantity), export (live animal export quantity), 5y_disease (disease dynamic score of last 5.5 years), odaic (ongoing international and domestic conflict), num (total number of vulnerable population), den (vulnerable animal population density), personnel (personnel capacity per vulnerable animal), sp_cont (total number of vulnerable species for which disease control measures exists), tot_cont (total number of disease control and prevention parameters exist for each disease), nl (total number of disease‐specific national laboratories), coc (control of corruption), rol (rule of law), rq (regulatory quality), ge (government effectiveness), vaa (voice and accountability), odaic_r (reverse ongoing international and domestic conflict), ssas_r (reverse societal safety and security), e (economic), pi (political‐international), article (research articles per person), literacy (literacy rate).

Abbreviations: Obs, observations; SD, standard deviation.

All three approaches using mean, distance and PCA scoring methods for each of four domains, and for the final vulnerability index are presented in Supporting Dataset 2. PCA components (eigenvalues) and rotated component loadings are provided in Table S4. Pairwise domain and final vulnerability index comparisons for each method and disease combination yielded only positive correlations (0.61–1.00). Indices obtained from the mean method were highly correlated (0.81–1.00) with the scores obtained from the distance method across all four diseases. All possible pairwise domain and the final vulnerability index comparisons for each method are presented in Table S5.

Individual release and exposure domain indices, along with negatively oriented PrepRes and resilience indices, as well as the final vulnerability indices calculated from all four diseases, are presented in Figure 3 for FMD and Figure S3a–c for the remaining three diseases (ASF, brucellosis [B. abortus], and HPAI, respectively).

FIGURE 3.

Individual domain and the final vulnerability indices obtained for foot and mouth disease (FMD) using mean, distance, and principal components analysis (PCA) methods. Non‐scaled FMD‐PCA (lower right) components were not constrained in either a positive or negative direction and are only presented here for comparison purposes.

All Spearman's rank correlations were significant correlated (p < .0005), with rho between 0.66 and 0. See Table S6 for individual pairwise correlation results.

After ordering each of the three methods from highest scored country to the lowest scored country for each domain and disease, Spearman's rank test showed highly significant (p < .0005) correlations among all three of the scoring methods (Figure 4). Pairwise Spearman's rank correlations by scoring method within each domain by disease and with the corresponding rho and p‐values are presented in Table S6. Individual domain and risk premium scores along with the order of the scores from the highest to the lowest are provided in Supporting Dataset 2.

FIGURE 4.

Heatmap representing country ordering for each color‐coded domain obtained from distance matrix, arithmetic mean, and principal component analysis (PCA) analyses for each of the four diseases. Lower scores (lighter the colour) represent the lower number rankings by country. Reverse scaling of release and exposure domains relative to preparedness and response (PrepRes) and resilience clouds the direct interpretation of ‘highest index’ versus ‘lowest index’ ordering

4. DISCUSSION

The calculated risk premiums (final vulnerability indices) are intended to be used as comparative indices for determining the risk differences at the country, regional, and global levels, as well as for assigning risk‐based premiums for weighting investments within and among countries and regions, or for providing foreign aid or assistance as part of international development. Each overall or component premium (or discount) should only be interpreted in relation to the other mutually insured WOAH Members. While WOAH is assumed to be the international ‘‘insurer’’ in this case, the overall premium and each of its four components are neutral with respect to the countries under consideration and can therefore be used to prioritize internal and external (e.g., foreign assistance) investments in disease prevention, preparedness, and control. All data inputs are transparent, so premium components can be prioritized for future risk mitigation. Due to the sensitive nature of this biological threat assessment, no country names are included in this report. The mutual insurer intends to share specific recommendations with each of its WOAH Members, while anonymized and aggregated indices serve as regional and global benchmarks to be explored by members of the insured pool.

In this study, the PCA method, using scaled domain components, provided the greatest discriminatory ability among the 25 mutually insured countries, while also explaining the greatest amount of variability in the data. This is readily apparent in Figure 3, as well as in Figures S3a–c. In contrast to the other two much more simplified methods, the PCA approach using scaled domain components identified greater differences among the 25 countries, both within each of the four domains and within the overall vulnerability index. This ability to discriminate among the domain scores and overall across the domains is crucial in setting appropriate risk‐based mutual insurance premium frameworks and in directing resources for each country towards those domain components with the greatest potential to reduce overall vulnerability to each of the four diseases.

Within each domain, some components are very difficult to change in the shorter term (e.g., GDP and literacy) and are thus likely to remain aspirational in the medium to longer term; however, within each domain there generally are other components for which actionable short‐term goals can be achieved and with good potential to reduce vulnerability and disease consequences by reducing release potential, enhancing preparedness and response, and improving resiliency with more modest investments. In terms of exposure potential, animal population numbers and densities are less likely—or even less desirable—to change; however, vulnerable numbers of those same populations in the presence of a viable and effective vaccine can certainly be shifted through orchestrated campaigns to immunize the hosts.

In the wake of the naturally occurring and ongoing ASF and human coronavirus (COVID‐19: SARS‐CoV‐2) disease outbreaks (COVID‐19), it has become apparent that despite the extensive resources deployed and efforts administered by individual Members, as well as the international community as a whole, to varying degrees the entire world has been shown to be vulnerable to infectious disease outbreaks. And yet, post‐introduction different countries and regions have been shown to be better able to respond and recover from these disease incursions. Many of these differences are tied to non‐disease‐specific and non‐biological aspects of resilience, supporting WOAH's OIE Guidelines (Table S1) that place an emphasis on socio‐economic considerations when assessing risk. We believe that the most successful disease responses will be performed by fully functioning, autonomous, and resilient animal disease preparedness and response networks. This response includes a surveillance system capable of detecting disease incursions resulting in early risk reduction measures on the ground. Our risk premium index was framed upon the relative likelihood that an attribute or behaviour would contribute to a WOAH Member being able or unable to detect the disease or agent, to recover or not recover to the pre‐attack disease state through response efforts, and with sufficient national resources and resilience to allow for such recovery to occur in a reasonable timeframe. Therefore, herein, we utilized a multivariate framework for developing an ordinated risk‐based mutual insurance premium suited to agro‐crime and agro‐terror events, with potential for extension to natural or unintentional outbreaks (Koch et al., 2020). The premium itself necessarily incorporates factors that affect the probability of disease agent introduction and spread, along with timely detection and mitigation. Notably, a risk‐based mutual insurance premium must also include those measures of national resilience which help to limit the consequences of disease incursions.

When conceptualizing a mutual insurance paradigm providing for disease response to agro‐crime and agro‐terrorism, it can be beneficial to evoke another abstraction around natural disasters such as flood insurance. Within such a paradigm, hydrology, topography, and floodplains are used to assess the susceptibility of people and property to flood damage, and then the design of the property structures and the surrounding infrastructure are assessed to evaluate the individual's and community's resilience to damage (FEMA, 2021). The probability, estimated time to incident, and the amount of coverage are taken into consideration to calculate the present value of future costs of response (FEMA, 2021). Another concept, adverse selection, is also pertinent to this paradigm. Adverse selection is a condition whereby the insurer and the insured have inequivalent information and the parties that constitute the highest risk are those that are more likely to purchase insurance (Cohen & Siegelman, 2010). Furthermore, moral hazard, in which those insured alter behaviours once they achieve coverage, may result in those parties acting with less prudence and becoming a greater concern to the insurer (Cohen & Siegelman, 2010).

In our risk‐based mutual insurance premium framework and vulnerability indices, we have proposed that the established ‘‘premiums’’ not be used to literally provide insurance to any one WOAH Member by the WOAH (the insurer); rather, we advocate the use of such dimensionless indices as a way for individual countries (or, donor nations) to prioritize investments among the components of each of the risk domains, so as to best optimize disease risk and achieve reductions in vulnerability. These also allow for self‐study and benchmarking against those peer nations included in the mutual insurance scheme. The data needed to populate this approach are generally publicly available and robust to external scrutiny. Proprietary or sensitive data can be maintained in confidence by the insurer, who acts solely on behalf of the ‘policy holders’ under a mutual insurance scheme.

For agro‐crime and agro‐terrorism involving one of the four exemplar transboundary infectious animal diseases, the concept may be broadened to have intelligent adversaries gain knowledge into the susceptibilities of established agricultural systems, as well as the design of the respective response structure, and these entities would be more likely to strike at points of weakness with the perceived greatest return on their investment (Knobler et al., 2002; Rios & Insua, 2012). In the presence of asymmetry of information between an insurer and the insured, and an intelligent adversary's likelihood to use this knowledge differential to an advantage, a mutual (policy‐holder owned) insurance is required (Cohen & Siegelman, 2010; Knobler et al., 2002; Rios & Insua, 2012). Practically speaking, if the difficulties of investing in attempts to respond to an intentional release are set aside, it is no less vital to the global community to continuously strive to work cohesively in support of preparation for such response efforts, since localized outbreaks can rapidly escalate to regional epidemics and then global pandemics.

In this assessment, we utilized a deductive approach for the selection of variables that accounted for release, exposure, and the preparedness/response‐related consequences of infectious disease agent (hazard) introduction for each WOAH Member. On the other hand, the variables selected to represent a nation's resilience were primarily based on an inductive approach. Therefore, to blend these, we used multivariate data reduction approaches suited to either deductive and inductive reasoning by using each of PCA, distance matrix, or an arithmetic mean calculation for each of the domains and for each disease (OECD, 2008; Yoon, 2012). Of course, increasing levels of aggregation, including via data reduction approaches utilized in building dimensionless indices, will necessarily result in a loss of granular information (41).

Ideally, when vulnerability indices are calculated, seasonal impacts would be accounted for, as previously seen in the FMD outbreak reported in the United Kingdom in 2001. Biophysical factors (e.g., temperature, rain, climate) would primarily affect disease dynamics and transmission by directly affecting pathogen survivability in the environment, as well as indirectly via wildlife and vectors. However, such an approach would act to further stratify the analyses, rather than working towards providing a global risk index. Further, each pathogen would need to be parametrized to reflect varying susceptibility to environmental degradation.

Of a more obvious nature are questions about how any one individual country might benefit from this analysis and strategize to improve its risk‐based mutual insurance premium for any given disease. Overall final vulnerability indices, as calculated by the PCA (scaled) components, are the subtotal of the four component indices (release, exposure, preparedness and response, and resilience) that are included in the analyses. Variables that are included in each of the components are scaled variables, and the variables and their contributions to the final vulnerability indices are derived and presented with full transparency. Therefore, country indices could only be improved by evaluating the low component scores and the contribution of each domain's parameter to these scores. For example, a low final vulnerability index of a country might be influenced by an imperfect preparedness and response component score, which further can be traced back to a particular disease control measure score that might be the result of a lack of precautions taken at the border for that disease, especially when other countries are engaging in such practices. Therefore, simply improving precautions at the border for the disease agent in question could improve the final vulnerability index value. In a similar way, benchmarking against all other countries in the insurance pool, or by region, could be useful wherein each country can visualize its own component scores and be able to compare itself against others (anonymized) within various economic or regional strata.

In a similar vein, it is advisable to be cautious when interpreting results of a data reduction scheme such as PCA. Non‐dimensionless component scores and expert‐based methods such as the Delphi technique may seem to require the suspension of empiricism and rely too heavily on composite indices for the ‘known unknowns’. In our analysis, we used un‐weighted but rescaled values and assumed all such factors equally contributed to the final vulnerability index. This is not likely to be the case. For example, we assumed that the total number of veterinarians and paraprofessional veterinarians per 100,000 vulnerable animal population contributed the same to the preparedness and response index as the national laboratory score (which was calculated based on the number of disease‐specific laboratories and the tests conducted in each country). While the laboratory scores are taken for a specific disease, the total numbers of animal health workers were not included as the numbers that work directly with each vulnerable animal host population. Therefore, the interpretations of our analyses for each particular disease should be made carefully since the results were generated using unweighted data. On the other hand, using only a subjective weighting method such as one based on expert opinion (e.g., the Delphi technique) could likewise be misleading since different evaluation results can be obtained from different assemblages of experts with varying familiarity with different countries and regions. In conclusion, using an integrated weighting method (programmed by expert opinion) that effectively combines these two types of methods into one approach might be ideal for future analyses. Thus, the interpretations of the results and recommendations for action could be made based on not only expert judgments but also real data for such analysis.

Simulation exercises and sensitivity assessments could be used to explore how closely the results might fit with the real‐life scenarios. We believe that conducting simulation exercises to identify those parameters that contribute to the greatest uncertainties in the index outputs could be important follow‐up and adjunct to our analysis. However, as has been noted repeatedly throughout the study, there exists a distinct shortage of empirical data concerning intentional releases of agriculturally relevant biological agents, along with the contributions that resilience components—such as governance and socio‐economic factors—might play in any disease recovery scenario. One solution to this problem could be to include past disease eradication experiences, such as culling in the face of unintentional or ‘natural outbreaks’, to better understand the factors needed for recovery success. Indeed, success with culling could be an ideal indicator of greater response capacity, including of the ability to pay indemnity and garner producer support for the same. Reliable sources of data, across the full spectrum of international development, would be needed for such an approach.

One of the difficulties faced during this analysis was considering whether vaccination in place (versus prohibited vaccination) within each of the countries represented either a ‘positive’ or ‘negative’ control measure for HPAIv and FMDv. While some countries in our dataset authorized the use of these vaccines, other countries (specifically, the ones that had earlier eradicated the disease in their country) often prohibited the use of such a vaccine. Vaccination for a disease can be an important preventive measure, and especially a control parameter. However, if vaccination is not in place for a disease, that outcome itself may not be a negative driver of control measures.

Further concerning the capacity to respond to an outbreak, is whether or not a military‐related domain, civil defence, law enforcement or the private sector efforts need to be included in the resilience domain due to their well‐documented involvement in disease control management during outbreaks. Military personnel may be involved in the response of an outbreak by helping to enhance the control measures, such as working on zoning or the disposal of dead livestock. Civil defence and law enforcement personnel would be essential in disease mitigation efforts such as movement bans and controlling access to infected premises. Therefore, in future analyses, such capacities might also be additional parameters to consider. However, this becomes somewhat problematic in measuring capacity (is it simply a function of numbers of enlisted personnel? Or, is a qualitative measure of technological capacity and proficiency required to do the job correctly and efficiently?). Further, the inclusion of military‐related parameters can create the misperception that such regional and global analyses are not merely restricted to animal health‐related questions.

In conclusion, a simplified approach to vulnerability analysis, using a risk‐based and dimensionless mutual insurance premium framework resulting in vulnerability indices (as employed in this study), have the advantage that data sources are pre‐existing for most jurisdictions and there exists a wealth of well‐documented reporting and databases for inputs to many of the domains. WOAH, FAO, World Bank, and many others maintain regularly updated and well‐curated datasets. Further refinements and a transition to an index bearing actual dimensions of risk, or else attempting to assign a monetary value (or resource equivalent), would require access to additional data sources; as such, these remain the major obstacles to furthering analysis and basing more robust conclusions on validated data.

AUTHOR CONTRIBUTIONS

Writing—original draft preparation (equal), review (equal), editing (equal), conceptualization (supporting), data curation (lead), formal analysis (equal), visualization (lead), and investigation (equal): Gizem Levent. Writing—original draft preparation (equal), review (equal), editing (equal), conceptualization (supporting), and investigation (equal): Christopher Glen Laine. Writing—review and editing (supporting), funding acquisition (equal), and investigation (supporting): Melissa Berquist. Writing—review and editing (supporting), funding acquisition (supporting), and investigation (supporting): Miguel Gonzalez. Writing—Review and editing (supporting), funding acquisition (equal), and investigation (supporting): Heather Simmons. Writing—review and editing (supporting), funding acquisition (supporting), and investigation (supporting): Jimmy Tickel. Writing—original draft preparation (equal), review, editing, conceptualization (lead), formal analysis, funding acquisition (supporting), supervision (lead), investigation (equal), and project administration (lead): Harvey Morgan Scott.

CONFLICT OF INTEREST

The authors declare no conflict of interest.

ETHICS STATEMENT

The authors confirm that the ethical policies of the journal, as noted on the journal's author guidelines page, have been adhered to. No ethical approval was required as this is an article with no original research data.

Supporting information

Supporting Information

Supporting Information

Supporting Information

Levent, G. , Laine, C. G. , Berquist, M. , Gonzalez, M. , Simmons, H. , Tickel, J. , & Scott, H. M. (2022). A risk‐based mutual insurance premium framework for establishing indices of vulnerability to the intentional introduction of transboundary animal diseases. Transboundary and Emerging Diseases, 69, 3582–3596. 10.1111/tbed.14721

DATA AVAILABILITY STATEMENT

The data that support the findings of this study are openly available in publicly accessible sources along with citations and references provided within the study. Further, the specific data points utilized from these sources are provided as Supporting Information to this manuscript.